blob: cac4a3f763230f7e448ca6de68a0859a42b13af8 [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"
Chris Mason79154b12007-03-22 15:59:16 -040034
Chris Mason0f7d52f2007-04-09 10:42:37 -040035#define BTRFS_ROOT_TRANS_TAG 0
36
Miao Xie4a9d8bd2013-05-17 03:53:43 +000037static unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = {
38 [TRANS_STATE_RUNNING] = 0U,
39 [TRANS_STATE_BLOCKED] = (__TRANS_USERSPACE |
40 __TRANS_START),
41 [TRANS_STATE_COMMIT_START] = (__TRANS_USERSPACE |
42 __TRANS_START |
43 __TRANS_ATTACH),
44 [TRANS_STATE_COMMIT_DOING] = (__TRANS_USERSPACE |
45 __TRANS_START |
46 __TRANS_ATTACH |
47 __TRANS_JOIN),
48 [TRANS_STATE_UNBLOCKED] = (__TRANS_USERSPACE |
49 __TRANS_START |
50 __TRANS_ATTACH |
51 __TRANS_JOIN |
52 __TRANS_JOIN_NOLOCK),
53 [TRANS_STATE_COMPLETED] = (__TRANS_USERSPACE |
54 __TRANS_START |
55 __TRANS_ATTACH |
56 __TRANS_JOIN |
57 __TRANS_JOIN_NOLOCK),
58};
59
Eric Sandeen48a3b632013-04-25 20:41:01 +000060static void put_transaction(struct btrfs_transaction *transaction)
Chris Mason79154b12007-03-22 15:59:16 -040061{
Josef Bacik13c5a932011-04-11 15:45:29 -040062 WARN_ON(atomic_read(&transaction->use_count) == 0);
63 if (atomic_dec_and_test(&transaction->use_count)) {
Josef Bacika4abeea2011-04-11 17:25:13 -040064 BUG_ON(!list_empty(&transaction->list));
Arne Jansen00f04b82011-09-14 12:37:00 +020065 WARN_ON(transaction->delayed_refs.root.rb_node);
Josef Bacik6df9a952013-06-27 13:22:46 -040066 while (!list_empty(&transaction->pending_chunks)) {
67 struct extent_map *em;
68
69 em = list_first_entry(&transaction->pending_chunks,
70 struct extent_map, list);
71 list_del_init(&em->list);
72 free_extent_map(em);
73 }
Chris Mason2c90e5d2007-04-02 10:50:19 -040074 kmem_cache_free(btrfs_transaction_cachep, transaction);
Chris Mason78fae272007-03-25 11:35:08 -040075 }
Chris Mason79154b12007-03-22 15:59:16 -040076}
77
Josef Bacik817d52f2009-07-13 21:29:25 -040078static noinline void switch_commit_root(struct btrfs_root *root)
79{
Josef Bacik817d52f2009-07-13 21:29:25 -040080 free_extent_buffer(root->commit_root);
81 root->commit_root = btrfs_root_node(root);
Josef Bacik817d52f2009-07-13 21:29:25 -040082}
83
Miao Xie0860adf2013-05-15 07:48:27 +000084static inline void extwriter_counter_inc(struct btrfs_transaction *trans,
85 unsigned int type)
86{
87 if (type & TRANS_EXTWRITERS)
88 atomic_inc(&trans->num_extwriters);
89}
90
91static inline void extwriter_counter_dec(struct btrfs_transaction *trans,
92 unsigned int type)
93{
94 if (type & TRANS_EXTWRITERS)
95 atomic_dec(&trans->num_extwriters);
96}
97
98static inline void extwriter_counter_init(struct btrfs_transaction *trans,
99 unsigned int type)
100{
101 atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0));
102}
103
104static inline int extwriter_counter_read(struct btrfs_transaction *trans)
105{
106 return atomic_read(&trans->num_extwriters);
Miao Xie178260b2013-02-20 09:16:24 +0000107}
108
Chris Masond352ac62008-09-29 15:18:18 -0400109/*
110 * either allocate a new transaction or hop into the existing one
111 */
Miao Xie0860adf2013-05-15 07:48:27 +0000112static noinline int join_transaction(struct btrfs_root *root, unsigned int type)
Chris Mason79154b12007-03-22 15:59:16 -0400113{
114 struct btrfs_transaction *cur_trans;
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200115 struct btrfs_fs_info *fs_info = root->fs_info;
Josef Bacika4abeea2011-04-11 17:25:13 -0400116
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200117 spin_lock(&fs_info->trans_lock);
Chris Masond43317d2011-11-06 03:26:19 -0500118loop:
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100119 /* The file system has been taken offline. No new transactions. */
Miao Xie87533c42013-01-29 10:14:48 +0000120 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200121 spin_unlock(&fs_info->trans_lock);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100122 return -EROFS;
123 }
124
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200125 cur_trans = fs_info->running_transaction;
Josef Bacika4abeea2011-04-11 17:25:13 -0400126 if (cur_trans) {
David Sterba871383b2012-04-02 18:31:37 +0200127 if (cur_trans->aborted) {
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200128 spin_unlock(&fs_info->trans_lock);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100129 return cur_trans->aborted;
David Sterba871383b2012-04-02 18:31:37 +0200130 }
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000131 if (btrfs_blocked_trans_types[cur_trans->state] & type) {
Miao Xie178260b2013-02-20 09:16:24 +0000132 spin_unlock(&fs_info->trans_lock);
133 return -EBUSY;
134 }
Josef Bacika4abeea2011-04-11 17:25:13 -0400135 atomic_inc(&cur_trans->use_count);
Josef Bacik13c5a932011-04-11 15:45:29 -0400136 atomic_inc(&cur_trans->num_writers);
Miao Xie0860adf2013-05-15 07:48:27 +0000137 extwriter_counter_inc(cur_trans, type);
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200138 spin_unlock(&fs_info->trans_lock);
Josef Bacika4abeea2011-04-11 17:25:13 -0400139 return 0;
Chris Mason79154b12007-03-22 15:59:16 -0400140 }
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200141 spin_unlock(&fs_info->trans_lock);
Josef Bacika4abeea2011-04-11 17:25:13 -0400142
Miao Xie354aa0f2012-09-20 01:54:00 -0600143 /*
144 * If we are ATTACH, we just want to catch the current transaction,
145 * and commit it. If there is no transaction, just return ENOENT.
146 */
147 if (type == TRANS_ATTACH)
148 return -ENOENT;
149
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000150 /*
151 * JOIN_NOLOCK only happens during the transaction commit, so
152 * it is impossible that ->running_transaction is NULL
153 */
154 BUG_ON(type == TRANS_JOIN_NOLOCK);
155
Josef Bacika4abeea2011-04-11 17:25:13 -0400156 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS);
157 if (!cur_trans)
158 return -ENOMEM;
Chris Masond43317d2011-11-06 03:26:19 -0500159
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200160 spin_lock(&fs_info->trans_lock);
161 if (fs_info->running_transaction) {
Chris Masond43317d2011-11-06 03:26:19 -0500162 /*
163 * someone started a transaction after we unlocked. Make sure
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000164 * to redo the checks above
Chris Masond43317d2011-11-06 03:26:19 -0500165 */
Josef Bacika4abeea2011-04-11 17:25:13 -0400166 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
Chris Masond43317d2011-11-06 03:26:19 -0500167 goto loop;
Miao Xie87533c42013-01-29 10:14:48 +0000168 } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
Dan Carpentere4b50e12012-06-19 13:30:11 +0300169 spin_unlock(&fs_info->trans_lock);
Josef Bacik7b8b92a2012-05-31 15:52:43 -0400170 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
171 return -EROFS;
Josef Bacika4abeea2011-04-11 17:25:13 -0400172 }
Chris Masond43317d2011-11-06 03:26:19 -0500173
Josef Bacika4abeea2011-04-11 17:25:13 -0400174 atomic_set(&cur_trans->num_writers, 1);
Miao Xie0860adf2013-05-15 07:48:27 +0000175 extwriter_counter_init(cur_trans, type);
Josef Bacika4abeea2011-04-11 17:25:13 -0400176 init_waitqueue_head(&cur_trans->writer_wait);
177 init_waitqueue_head(&cur_trans->commit_wait);
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000178 cur_trans->state = TRANS_STATE_RUNNING;
Josef Bacika4abeea2011-04-11 17:25:13 -0400179 /*
180 * One for this trans handle, one so it will live on until we
181 * commit the transaction.
182 */
183 atomic_set(&cur_trans->use_count, 2);
Josef Bacika4abeea2011-04-11 17:25:13 -0400184 cur_trans->start_time = get_seconds();
185
186 cur_trans->delayed_refs.root = RB_ROOT;
187 cur_trans->delayed_refs.num_entries = 0;
188 cur_trans->delayed_refs.num_heads_ready = 0;
189 cur_trans->delayed_refs.num_heads = 0;
190 cur_trans->delayed_refs.flushing = 0;
191 cur_trans->delayed_refs.run_delayed_start = 0;
Jan Schmidt20b297d2012-05-20 15:43:53 +0200192
193 /*
194 * although the tree mod log is per file system and not per transaction,
195 * the log must never go across transaction boundaries.
196 */
197 smp_mb();
Julia Lawall31b1a2b2012-11-03 10:58:34 +0000198 if (!list_empty(&fs_info->tree_mod_seq_list))
199 WARN(1, KERN_ERR "btrfs: tree_mod_seq_list not empty when "
Jan Schmidt20b297d2012-05-20 15:43:53 +0200200 "creating a fresh transaction\n");
Julia Lawall31b1a2b2012-11-03 10:58:34 +0000201 if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log))
202 WARN(1, KERN_ERR "btrfs: tree_mod_log rb tree not empty when "
Jan Schmidt20b297d2012-05-20 15:43:53 +0200203 "creating a fresh transaction\n");
Jan Schmidtfc36ed7e2013-04-24 16:57:33 +0000204 atomic64_set(&fs_info->tree_mod_seq, 0);
Jan Schmidt20b297d2012-05-20 15:43:53 +0200205
Josef Bacika4abeea2011-04-11 17:25:13 -0400206 spin_lock_init(&cur_trans->delayed_refs.lock);
Chris Masonbb721702013-01-29 18:44:12 -0500207 atomic_set(&cur_trans->delayed_refs.procs_running_refs, 0);
208 atomic_set(&cur_trans->delayed_refs.ref_seq, 0);
209 init_waitqueue_head(&cur_trans->delayed_refs.wait);
Josef Bacika4abeea2011-04-11 17:25:13 -0400210
211 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
Josef Bacik569e0f32013-02-13 11:09:14 -0500212 INIT_LIST_HEAD(&cur_trans->ordered_operations);
Josef Bacik6df9a952013-06-27 13:22:46 -0400213 INIT_LIST_HEAD(&cur_trans->pending_chunks);
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200214 list_add_tail(&cur_trans->list, &fs_info->trans_list);
Josef Bacika4abeea2011-04-11 17:25:13 -0400215 extent_io_tree_init(&cur_trans->dirty_pages,
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200216 fs_info->btree_inode->i_mapping);
217 fs_info->generation++;
218 cur_trans->transid = fs_info->generation;
219 fs_info->running_transaction = cur_trans;
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100220 cur_trans->aborted = 0;
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200221 spin_unlock(&fs_info->trans_lock);
Josef Bacik15ee9bc2007-08-10 16:22:09 -0400222
Chris Mason79154b12007-03-22 15:59:16 -0400223 return 0;
224}
225
Chris Masond352ac62008-09-29 15:18:18 -0400226/*
Chris Masond3977122009-01-05 21:25:51 -0500227 * this does all the record keeping required to make sure that a reference
228 * counted root is properly recorded in a given transaction. This is required
229 * to make sure the old root from before we joined the transaction is deleted
230 * when the transaction commits
Chris Masond352ac62008-09-29 15:18:18 -0400231 */
Chris Mason75857172011-06-13 20:00:16 -0400232static int record_root_in_trans(struct btrfs_trans_handle *trans,
Josef Bacika4abeea2011-04-11 17:25:13 -0400233 struct btrfs_root *root)
Chris Mason6702ed42007-08-07 16:15:09 -0400234{
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400235 if (root->ref_cows && root->last_trans < trans->transid) {
Chris Mason6702ed42007-08-07 16:15:09 -0400236 WARN_ON(root == root->fs_info->extent_root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400237 WARN_ON(root->commit_root != root->node);
Yan Zheng31153d82008-07-28 15:32:19 -0400238
Chris Mason75857172011-06-13 20:00:16 -0400239 /*
240 * see below for in_trans_setup usage rules
241 * we have the reloc mutex held now, so there
242 * is only one writer in this function
243 */
244 root->in_trans_setup = 1;
245
246 /* make sure readers find in_trans_setup before
247 * they find our root->last_trans update
248 */
249 smp_wmb();
250
Josef Bacika4abeea2011-04-11 17:25:13 -0400251 spin_lock(&root->fs_info->fs_roots_radix_lock);
252 if (root->last_trans == trans->transid) {
253 spin_unlock(&root->fs_info->fs_roots_radix_lock);
254 return 0;
255 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400256 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
257 (unsigned long)root->root_key.objectid,
258 BTRFS_ROOT_TRANS_TAG);
Josef Bacika4abeea2011-04-11 17:25:13 -0400259 spin_unlock(&root->fs_info->fs_roots_radix_lock);
Chris Mason75857172011-06-13 20:00:16 -0400260 root->last_trans = trans->transid;
261
262 /* this is pretty tricky. We don't want to
263 * take the relocation lock in btrfs_record_root_in_trans
264 * unless we're really doing the first setup for this root in
265 * this transaction.
266 *
267 * Normally we'd use root->last_trans as a flag to decide
268 * if we want to take the expensive mutex.
269 *
270 * But, we have to set root->last_trans before we
271 * init the relocation root, otherwise, we trip over warnings
272 * in ctree.c. The solution used here is to flag ourselves
273 * with root->in_trans_setup. When this is 1, we're still
274 * fixing up the reloc trees and everyone must wait.
275 *
276 * When this is zero, they can trust root->last_trans and fly
277 * through btrfs_record_root_in_trans without having to take the
278 * lock. smp_wmb() makes sure that all the writes above are
279 * done before we pop in the zero below
280 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400281 btrfs_init_reloc_root(trans, root);
Chris Mason75857172011-06-13 20:00:16 -0400282 smp_wmb();
283 root->in_trans_setup = 0;
Chris Mason6702ed42007-08-07 16:15:09 -0400284 }
285 return 0;
286}
287
Chris Mason75857172011-06-13 20:00:16 -0400288
289int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
290 struct btrfs_root *root)
291{
292 if (!root->ref_cows)
293 return 0;
294
295 /*
296 * see record_root_in_trans for comments about in_trans_setup usage
297 * and barriers
298 */
299 smp_rmb();
300 if (root->last_trans == trans->transid &&
301 !root->in_trans_setup)
302 return 0;
303
304 mutex_lock(&root->fs_info->reloc_mutex);
305 record_root_in_trans(trans, root);
306 mutex_unlock(&root->fs_info->reloc_mutex);
307
308 return 0;
309}
310
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000311static inline int is_transaction_blocked(struct btrfs_transaction *trans)
312{
313 return (trans->state >= TRANS_STATE_BLOCKED &&
Josef Bacik501407a2013-06-10 16:47:23 -0400314 trans->state < TRANS_STATE_UNBLOCKED &&
315 !trans->aborted);
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000316}
317
Chris Masond352ac62008-09-29 15:18:18 -0400318/* wait for commit against the current transaction to become unblocked
319 * when this is done, it is safe to start a new transaction, but the current
320 * transaction might not be fully on disk.
321 */
Chris Mason37d1aee2008-07-31 10:48:37 -0400322static void wait_current_trans(struct btrfs_root *root)
Chris Mason79154b12007-03-22 15:59:16 -0400323{
Chris Masonf9295742008-07-17 12:54:14 -0400324 struct btrfs_transaction *cur_trans;
Chris Mason79154b12007-03-22 15:59:16 -0400325
Josef Bacika4abeea2011-04-11 17:25:13 -0400326 spin_lock(&root->fs_info->trans_lock);
Chris Masonf9295742008-07-17 12:54:14 -0400327 cur_trans = root->fs_info->running_transaction;
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000328 if (cur_trans && is_transaction_blocked(cur_trans)) {
Josef Bacik13c5a932011-04-11 15:45:29 -0400329 atomic_inc(&cur_trans->use_count);
Josef Bacika4abeea2011-04-11 17:25:13 -0400330 spin_unlock(&root->fs_info->trans_lock);
Li Zefan72d63ed2011-07-14 03:17:00 +0000331
332 wait_event(root->fs_info->transaction_wait,
Josef Bacik501407a2013-06-10 16:47:23 -0400333 cur_trans->state >= TRANS_STATE_UNBLOCKED ||
334 cur_trans->aborted);
Chris Masonf9295742008-07-17 12:54:14 -0400335 put_transaction(cur_trans);
Josef Bacika4abeea2011-04-11 17:25:13 -0400336 } else {
337 spin_unlock(&root->fs_info->trans_lock);
Chris Masonf9295742008-07-17 12:54:14 -0400338 }
Chris Mason37d1aee2008-07-31 10:48:37 -0400339}
340
Yan, Zhenga22285a2010-05-16 10:48:46 -0400341static int may_wait_transaction(struct btrfs_root *root, int type)
Chris Mason37d1aee2008-07-31 10:48:37 -0400342{
Josef Bacika4abeea2011-04-11 17:25:13 -0400343 if (root->fs_info->log_root_recovering)
344 return 0;
345
346 if (type == TRANS_USERSPACE)
Yan, Zhenga22285a2010-05-16 10:48:46 -0400347 return 1;
Josef Bacika4abeea2011-04-11 17:25:13 -0400348
349 if (type == TRANS_START &&
350 !atomic_read(&root->fs_info->open_ioctl_trans))
351 return 1;
352
Yan, Zhenga22285a2010-05-16 10:48:46 -0400353 return 0;
354}
355
Miao Xie08e007d2012-10-16 11:33:38 +0000356static struct btrfs_trans_handle *
Miao Xie0860adf2013-05-15 07:48:27 +0000357start_transaction(struct btrfs_root *root, u64 num_items, unsigned int type,
Miao Xie08e007d2012-10-16 11:33:38 +0000358 enum btrfs_reserve_flush_enum flush)
Yan, Zhenga22285a2010-05-16 10:48:46 -0400359{
360 struct btrfs_trans_handle *h;
361 struct btrfs_transaction *cur_trans;
Josef Bacikb5009942011-06-07 15:07:51 -0400362 u64 num_bytes = 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400363 int ret;
Arne Jansenc5567232011-09-14 15:44:05 +0200364 u64 qgroup_reserved = 0;
liuboacce9522011-01-06 19:30:25 +0800365
Miao Xie87533c42013-01-29 10:14:48 +0000366 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
liuboacce9522011-01-06 19:30:25 +0800367 return ERR_PTR(-EROFS);
Josef Bacik2a1eb462011-04-13 15:15:59 -0400368
369 if (current->journal_info) {
Miao Xie0860adf2013-05-15 07:48:27 +0000370 WARN_ON(type & TRANS_EXTWRITERS);
Josef Bacik2a1eb462011-04-13 15:15:59 -0400371 h = current->journal_info;
372 h->use_count++;
Miao Xieb7d5b0a2012-11-01 07:32:18 +0000373 WARN_ON(h->use_count > 2);
Josef Bacik2a1eb462011-04-13 15:15:59 -0400374 h->orig_rsv = h->block_rsv;
375 h->block_rsv = NULL;
376 goto got_it;
377 }
Josef Bacikb5009942011-06-07 15:07:51 -0400378
379 /*
380 * Do the reservation before we join the transaction so we can do all
381 * the appropriate flushing if need be.
382 */
383 if (num_items > 0 && root != root->fs_info->chunk_root) {
Arne Jansenc5567232011-09-14 15:44:05 +0200384 if (root->fs_info->quota_enabled &&
385 is_fstree(root->root_key.objectid)) {
386 qgroup_reserved = num_items * root->leafsize;
387 ret = btrfs_qgroup_reserve(root, qgroup_reserved);
388 if (ret)
389 return ERR_PTR(ret);
390 }
391
Josef Bacikb5009942011-06-07 15:07:51 -0400392 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
Miao Xie08e007d2012-10-16 11:33:38 +0000393 ret = btrfs_block_rsv_add(root,
394 &root->fs_info->trans_block_rsv,
395 num_bytes, flush);
Josef Bacikb5009942011-06-07 15:07:51 -0400396 if (ret)
Miao Xie843fcf32013-01-28 12:36:22 +0000397 goto reserve_fail;
Josef Bacikb5009942011-06-07 15:07:51 -0400398 }
Yan, Zhenga22285a2010-05-16 10:48:46 -0400399again:
400 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
Miao Xie843fcf32013-01-28 12:36:22 +0000401 if (!h) {
402 ret = -ENOMEM;
403 goto alloc_fail;
404 }
Yan, Zhenga22285a2010-05-16 10:48:46 -0400405
Josef Bacik98114652012-09-14 11:22:38 -0400406 /*
407 * If we are JOIN_NOLOCK we're already committing a transaction and
408 * waiting on this guy, so we don't need to do the sb_start_intwrite
409 * because we're already holding a ref. We need this because we could
410 * have raced in and did an fsync() on a file which can kick a commit
411 * and then we deadlock with somebody doing a freeze.
Miao Xie354aa0f2012-09-20 01:54:00 -0600412 *
413 * If we are ATTACH, it means we just want to catch the current
414 * transaction and commit it, so we needn't do sb_start_intwrite().
Josef Bacik98114652012-09-14 11:22:38 -0400415 */
Miao Xie0860adf2013-05-15 07:48:27 +0000416 if (type & __TRANS_FREEZABLE)
Josef Bacik60376ce2012-09-14 10:34:40 -0400417 sb_start_intwrite(root->fs_info->sb);
Jan Karab2b5ef52012-06-12 16:20:45 +0200418
Yan, Zhenga22285a2010-05-16 10:48:46 -0400419 if (may_wait_transaction(root, type))
Chris Mason37d1aee2008-07-31 10:48:37 -0400420 wait_current_trans(root);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400421
Josef Bacika4abeea2011-04-11 17:25:13 -0400422 do {
Miao Xie354aa0f2012-09-20 01:54:00 -0600423 ret = join_transaction(root, type);
Miao Xie178260b2013-02-20 09:16:24 +0000424 if (ret == -EBUSY) {
Josef Bacika4abeea2011-04-11 17:25:13 -0400425 wait_current_trans(root);
Miao Xie178260b2013-02-20 09:16:24 +0000426 if (unlikely(type == TRANS_ATTACH))
427 ret = -ENOENT;
428 }
Josef Bacika4abeea2011-04-11 17:25:13 -0400429 } while (ret == -EBUSY);
430
Tsutomu Itohdb5b4932011-03-23 08:14:16 +0000431 if (ret < 0) {
Miao Xie354aa0f2012-09-20 01:54:00 -0600432 /* We must get the transaction if we are JOIN_NOLOCK. */
433 BUG_ON(type == TRANS_JOIN_NOLOCK);
Miao Xie843fcf32013-01-28 12:36:22 +0000434 goto join_fail;
Tsutomu Itohdb5b4932011-03-23 08:14:16 +0000435 }
Chris Mason0f7d52f2007-04-09 10:42:37 -0400436
Yan, Zhenga22285a2010-05-16 10:48:46 -0400437 cur_trans = root->fs_info->running_transaction;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400438
439 h->transid = cur_trans->transid;
440 h->transaction = cur_trans;
Chris Mason79154b12007-03-22 15:59:16 -0400441 h->blocks_used = 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400442 h->bytes_reserved = 0;
Arne Jansend13603e2011-09-13 11:40:09 +0200443 h->root = root;
Chris Mason56bec292009-03-13 10:10:06 -0400444 h->delayed_ref_updates = 0;
Josef Bacik2a1eb462011-04-13 15:15:59 -0400445 h->use_count = 1;
Josef Bacik0e721102012-06-26 16:13:18 -0400446 h->adding_csums = 0;
Yan, Zhengf0486c62010-05-16 10:46:25 -0400447 h->block_rsv = NULL;
Josef Bacik2a1eb462011-04-13 15:15:59 -0400448 h->orig_rsv = NULL;
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100449 h->aborted = 0;
Miao Xie4b824902013-02-20 09:13:32 +0000450 h->qgroup_reserved = 0;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200451 h->delayed_ref_elem.seq = 0;
Miao Xiea698d0752012-09-20 01:51:59 -0600452 h->type = type;
Josef Bacikc6b305a2012-12-18 09:16:16 -0500453 h->allocating_chunk = false;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200454 INIT_LIST_HEAD(&h->qgroup_ref_list);
Josef Bacikea658ba2012-09-11 16:57:25 -0400455 INIT_LIST_HEAD(&h->new_bgs);
Chris Masonb7ec40d2009-03-12 20:12:45 -0400456
Yan, Zhenga22285a2010-05-16 10:48:46 -0400457 smp_mb();
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000458 if (cur_trans->state >= TRANS_STATE_BLOCKED &&
459 may_wait_transaction(root, type)) {
Yan, Zhenga22285a2010-05-16 10:48:46 -0400460 btrfs_commit_transaction(h, root);
461 goto again;
462 }
Josef Bacik9ed74f22009-09-11 16:12:44 -0400463
Josef Bacikb5009942011-06-07 15:07:51 -0400464 if (num_bytes) {
Josef Bacik8c2a3ca2012-01-10 10:31:31 -0500465 trace_btrfs_space_reservation(root->fs_info, "transaction",
Liu Bo2bcc0322012-03-29 09:57:44 -0400466 h->transid, num_bytes, 1);
Josef Bacikb5009942011-06-07 15:07:51 -0400467 h->block_rsv = &root->fs_info->trans_block_rsv;
468 h->bytes_reserved = num_bytes;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400469 }
Miao Xie4b824902013-02-20 09:13:32 +0000470 h->qgroup_reserved = qgroup_reserved;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400471
Josef Bacik2a1eb462011-04-13 15:15:59 -0400472got_it:
Josef Bacika4abeea2011-04-11 17:25:13 -0400473 btrfs_record_root_in_trans(h, root);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400474
475 if (!current->journal_info && type != TRANS_USERSPACE)
476 current->journal_info = h;
Chris Mason79154b12007-03-22 15:59:16 -0400477 return h;
Miao Xie843fcf32013-01-28 12:36:22 +0000478
479join_fail:
Miao Xie0860adf2013-05-15 07:48:27 +0000480 if (type & __TRANS_FREEZABLE)
Miao Xie843fcf32013-01-28 12:36:22 +0000481 sb_end_intwrite(root->fs_info->sb);
482 kmem_cache_free(btrfs_trans_handle_cachep, h);
483alloc_fail:
484 if (num_bytes)
485 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
486 num_bytes);
487reserve_fail:
488 if (qgroup_reserved)
489 btrfs_qgroup_free(root, qgroup_reserved);
490 return ERR_PTR(ret);
Chris Mason79154b12007-03-22 15:59:16 -0400491}
492
Chris Masonf9295742008-07-17 12:54:14 -0400493struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
Yan, Zhenga22285a2010-05-16 10:48:46 -0400494 int num_items)
Chris Masonf9295742008-07-17 12:54:14 -0400495{
Miao Xie08e007d2012-10-16 11:33:38 +0000496 return start_transaction(root, num_items, TRANS_START,
497 BTRFS_RESERVE_FLUSH_ALL);
Chris Masonf9295742008-07-17 12:54:14 -0400498}
Miao Xie8407aa42012-09-07 01:43:32 -0600499
Miao Xie08e007d2012-10-16 11:33:38 +0000500struct btrfs_trans_handle *btrfs_start_transaction_lflush(
Miao Xie8407aa42012-09-07 01:43:32 -0600501 struct btrfs_root *root, int num_items)
502{
Miao Xie08e007d2012-10-16 11:33:38 +0000503 return start_transaction(root, num_items, TRANS_START,
504 BTRFS_RESERVE_FLUSH_LIMIT);
Miao Xie8407aa42012-09-07 01:43:32 -0600505}
506
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400507struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
Chris Masonf9295742008-07-17 12:54:14 -0400508{
Miao Xie8407aa42012-09-07 01:43:32 -0600509 return start_transaction(root, 0, TRANS_JOIN, 0);
Chris Masonf9295742008-07-17 12:54:14 -0400510}
511
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400512struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
Josef Bacik0af3d002010-06-21 14:48:16 -0400513{
Miao Xie8407aa42012-09-07 01:43:32 -0600514 return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0);
Josef Bacik0af3d002010-06-21 14:48:16 -0400515}
516
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400517struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
Sage Weil9ca9ee02008-08-04 10:41:27 -0400518{
Miao Xie8407aa42012-09-07 01:43:32 -0600519 return start_transaction(root, 0, TRANS_USERSPACE, 0);
Sage Weil9ca9ee02008-08-04 10:41:27 -0400520}
521
Miao Xied4edf392013-02-20 09:17:06 +0000522/*
523 * btrfs_attach_transaction() - catch the running transaction
524 *
525 * It is used when we want to commit the current the transaction, but
526 * don't want to start a new one.
527 *
528 * Note: If this function return -ENOENT, it just means there is no
529 * running transaction. But it is possible that the inactive transaction
530 * is still in the memory, not fully on disk. If you hope there is no
531 * inactive transaction in the fs when -ENOENT is returned, you should
532 * invoke
533 * btrfs_attach_transaction_barrier()
534 */
Miao Xie354aa0f2012-09-20 01:54:00 -0600535struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
Josef Bacik60376ce2012-09-14 10:34:40 -0400536{
Miao Xie354aa0f2012-09-20 01:54:00 -0600537 return start_transaction(root, 0, TRANS_ATTACH, 0);
Josef Bacik60376ce2012-09-14 10:34:40 -0400538}
539
Miao Xied4edf392013-02-20 09:17:06 +0000540/*
Wang Sheng-Hui90b6d282013-06-14 16:21:24 +0800541 * btrfs_attach_transaction_barrier() - catch the running transaction
Miao Xied4edf392013-02-20 09:17:06 +0000542 *
543 * It is similar to the above function, the differentia is this one
544 * will wait for all the inactive transactions until they fully
545 * complete.
546 */
547struct btrfs_trans_handle *
548btrfs_attach_transaction_barrier(struct btrfs_root *root)
549{
550 struct btrfs_trans_handle *trans;
551
552 trans = start_transaction(root, 0, TRANS_ATTACH, 0);
553 if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT)
554 btrfs_wait_for_commit(root, 0);
555
556 return trans;
557}
558
Chris Masond352ac62008-09-29 15:18:18 -0400559/* wait for a transaction commit to be fully complete */
Li Zefanb9c83002011-07-14 03:17:14 +0000560static noinline void wait_for_commit(struct btrfs_root *root,
Chris Mason89ce8a62008-06-25 16:01:31 -0400561 struct btrfs_transaction *commit)
562{
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000563 wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED);
Chris Mason89ce8a62008-06-25 16:01:31 -0400564}
565
Sage Weil46204592010-10-29 15:41:32 -0400566int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
567{
568 struct btrfs_transaction *cur_trans = NULL, *t;
Miao Xie8cd28072012-11-26 08:42:07 +0000569 int ret = 0;
Sage Weil46204592010-10-29 15:41:32 -0400570
Sage Weil46204592010-10-29 15:41:32 -0400571 if (transid) {
572 if (transid <= root->fs_info->last_trans_committed)
Josef Bacika4abeea2011-04-11 17:25:13 -0400573 goto out;
Sage Weil46204592010-10-29 15:41:32 -0400574
Miao Xie8cd28072012-11-26 08:42:07 +0000575 ret = -EINVAL;
Sage Weil46204592010-10-29 15:41:32 -0400576 /* find specified transaction */
Josef Bacika4abeea2011-04-11 17:25:13 -0400577 spin_lock(&root->fs_info->trans_lock);
Sage Weil46204592010-10-29 15:41:32 -0400578 list_for_each_entry(t, &root->fs_info->trans_list, list) {
579 if (t->transid == transid) {
580 cur_trans = t;
Josef Bacika4abeea2011-04-11 17:25:13 -0400581 atomic_inc(&cur_trans->use_count);
Miao Xie8cd28072012-11-26 08:42:07 +0000582 ret = 0;
Sage Weil46204592010-10-29 15:41:32 -0400583 break;
584 }
Miao Xie8cd28072012-11-26 08:42:07 +0000585 if (t->transid > transid) {
586 ret = 0;
Sage Weil46204592010-10-29 15:41:32 -0400587 break;
Miao Xie8cd28072012-11-26 08:42:07 +0000588 }
Sage Weil46204592010-10-29 15:41:32 -0400589 }
Josef Bacika4abeea2011-04-11 17:25:13 -0400590 spin_unlock(&root->fs_info->trans_lock);
Miao Xie8cd28072012-11-26 08:42:07 +0000591 /* The specified transaction doesn't exist */
Sage Weil46204592010-10-29 15:41:32 -0400592 if (!cur_trans)
Miao Xie8cd28072012-11-26 08:42:07 +0000593 goto out;
Sage Weil46204592010-10-29 15:41:32 -0400594 } else {
595 /* find newest transaction that is committing | committed */
Josef Bacika4abeea2011-04-11 17:25:13 -0400596 spin_lock(&root->fs_info->trans_lock);
Sage Weil46204592010-10-29 15:41:32 -0400597 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
598 list) {
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000599 if (t->state >= TRANS_STATE_COMMIT_START) {
600 if (t->state == TRANS_STATE_COMPLETED)
Josef Bacik3473f3c2011-06-09 10:15:17 -0400601 break;
Sage Weil46204592010-10-29 15:41:32 -0400602 cur_trans = t;
Josef Bacika4abeea2011-04-11 17:25:13 -0400603 atomic_inc(&cur_trans->use_count);
Sage Weil46204592010-10-29 15:41:32 -0400604 break;
605 }
606 }
Josef Bacika4abeea2011-04-11 17:25:13 -0400607 spin_unlock(&root->fs_info->trans_lock);
Sage Weil46204592010-10-29 15:41:32 -0400608 if (!cur_trans)
Josef Bacika4abeea2011-04-11 17:25:13 -0400609 goto out; /* nothing committing|committed */
Sage Weil46204592010-10-29 15:41:32 -0400610 }
611
Sage Weil46204592010-10-29 15:41:32 -0400612 wait_for_commit(root, cur_trans);
Sage Weil46204592010-10-29 15:41:32 -0400613 put_transaction(cur_trans);
Josef Bacika4abeea2011-04-11 17:25:13 -0400614out:
Sage Weil46204592010-10-29 15:41:32 -0400615 return ret;
616}
617
Chris Mason37d1aee2008-07-31 10:48:37 -0400618void btrfs_throttle(struct btrfs_root *root)
619{
Josef Bacika4abeea2011-04-11 17:25:13 -0400620 if (!atomic_read(&root->fs_info->open_ioctl_trans))
Sage Weil9ca9ee02008-08-04 10:41:27 -0400621 wait_current_trans(root);
Chris Mason37d1aee2008-07-31 10:48:37 -0400622}
623
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400624static int should_end_transaction(struct btrfs_trans_handle *trans,
625 struct btrfs_root *root)
626{
Josef Bacik1be41b72013-06-12 13:56:06 -0400627 if (root->fs_info->global_block_rsv.space_info->full &&
628 btrfs_should_throttle_delayed_refs(trans, root))
629 return 1;
Josef Bacik36ba0222011-10-18 12:15:48 -0400630
Josef Bacik1be41b72013-06-12 13:56:06 -0400631 return !!btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400632}
633
634int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
635 struct btrfs_root *root)
636{
637 struct btrfs_transaction *cur_trans = trans->transaction;
638 int updates;
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100639 int err;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400640
Josef Bacika4abeea2011-04-11 17:25:13 -0400641 smp_mb();
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000642 if (cur_trans->state >= TRANS_STATE_BLOCKED ||
643 cur_trans->delayed_refs.flushing)
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400644 return 1;
645
646 updates = trans->delayed_ref_updates;
647 trans->delayed_ref_updates = 0;
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100648 if (updates) {
649 err = btrfs_run_delayed_refs(trans, root, updates);
650 if (err) /* Error code will also eval true */
651 return err;
652 }
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400653
654 return should_end_transaction(trans, root);
655}
656
Chris Mason89ce8a62008-06-25 16:01:31 -0400657static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
Miao Xiea698d0752012-09-20 01:51:59 -0600658 struct btrfs_root *root, int throttle)
Chris Mason79154b12007-03-22 15:59:16 -0400659{
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400660 struct btrfs_transaction *cur_trans = trans->transaction;
Chris Masonab78c842008-07-29 16:15:18 -0400661 struct btrfs_fs_info *info = root->fs_info;
Josef Bacik1be41b72013-06-12 13:56:06 -0400662 unsigned long cur = trans->delayed_ref_updates;
Miao Xiea698d0752012-09-20 01:51:59 -0600663 int lock = (trans->type != TRANS_JOIN_NOLOCK);
Dave Jones4edc2ca2012-04-12 16:03:56 -0400664 int err = 0;
Chris Masond6e4a422007-04-06 15:37:36 -0400665
Josef Bacik2a1eb462011-04-13 15:15:59 -0400666 if (--trans->use_count) {
667 trans->block_rsv = trans->orig_rsv;
668 return 0;
669 }
670
Jan Schmidtedf39272012-06-28 18:04:55 +0200671 /*
672 * do the qgroup accounting as early as possible
673 */
674 err = btrfs_delayed_refs_qgroup_accounting(trans, info);
675
Josef Bacikb24e03d2011-10-14 14:40:17 -0400676 btrfs_trans_release_metadata(trans, root);
Josef Bacik4c13d752011-08-30 11:31:29 -0400677 trans->block_rsv = NULL;
Arne Jansenc5567232011-09-14 15:44:05 +0200678
679 if (trans->qgroup_reserved) {
Liu Bo7c2ec3f2013-03-13 07:43:03 -0600680 /*
681 * the same root has to be passed here between start_transaction
682 * and end_transaction. Subvolume quota depends on this.
683 */
684 btrfs_qgroup_free(trans->root, trans->qgroup_reserved);
Arne Jansenc5567232011-09-14 15:44:05 +0200685 trans->qgroup_reserved = 0;
686 }
687
Josef Bacikea658ba2012-09-11 16:57:25 -0400688 if (!list_empty(&trans->new_bgs))
689 btrfs_create_pending_block_groups(trans, root);
690
Josef Bacik1be41b72013-06-12 13:56:06 -0400691 trans->delayed_ref_updates = 0;
692 if (btrfs_should_throttle_delayed_refs(trans, root)) {
693 cur = max_t(unsigned long, cur, 1);
Chris Masonc3e69d52009-03-13 10:17:05 -0400694 trans->delayed_ref_updates = 0;
Josef Bacik1be41b72013-06-12 13:56:06 -0400695 btrfs_run_delayed_refs(trans, root, cur);
Chris Mason56bec292009-03-13 10:10:06 -0400696 }
Chris Masonbb721702013-01-29 18:44:12 -0500697
Josef Bacik0e721102012-06-26 16:13:18 -0400698 btrfs_trans_release_metadata(trans, root);
699 trans->block_rsv = NULL;
Chris Mason56bec292009-03-13 10:10:06 -0400700
Josef Bacikea658ba2012-09-11 16:57:25 -0400701 if (!list_empty(&trans->new_bgs))
702 btrfs_create_pending_block_groups(trans, root);
703
Josef Bacika4abeea2011-04-11 17:25:13 -0400704 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000705 should_end_transaction(trans, root) &&
706 ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) {
707 spin_lock(&info->trans_lock);
708 if (cur_trans->state == TRANS_STATE_RUNNING)
709 cur_trans->state = TRANS_STATE_BLOCKED;
710 spin_unlock(&info->trans_lock);
Josef Bacika4abeea2011-04-11 17:25:13 -0400711 }
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400712
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000713 if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) {
Josef Bacik81317fd2011-07-24 15:45:34 -0400714 if (throttle) {
715 /*
716 * We may race with somebody else here so end up having
717 * to call end_transaction on ourselves again, so inc
718 * our use_count.
719 */
720 trans->use_count++;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400721 return btrfs_commit_transaction(trans, root);
Josef Bacik81317fd2011-07-24 15:45:34 -0400722 } else {
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400723 wake_up_process(info->transaction_kthread);
Josef Bacik81317fd2011-07-24 15:45:34 -0400724 }
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400725 }
726
Miao Xie0860adf2013-05-15 07:48:27 +0000727 if (trans->type & __TRANS_FREEZABLE)
Josef Bacik98114652012-09-14 11:22:38 -0400728 sb_end_intwrite(root->fs_info->sb);
Josef Bacik6df78812012-09-05 08:08:30 -0600729
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400730 WARN_ON(cur_trans != info->running_transaction);
Josef Bacik13c5a932011-04-11 15:45:29 -0400731 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
732 atomic_dec(&cur_trans->num_writers);
Miao Xie0860adf2013-05-15 07:48:27 +0000733 extwriter_counter_dec(cur_trans, trans->type);
Chris Mason89ce8a62008-06-25 16:01:31 -0400734
Sage Weil99d16cb2010-10-29 15:37:34 -0400735 smp_mb();
Chris Mason79154b12007-03-22 15:59:16 -0400736 if (waitqueue_active(&cur_trans->writer_wait))
737 wake_up(&cur_trans->writer_wait);
Chris Mason79154b12007-03-22 15:59:16 -0400738 put_transaction(cur_trans);
Josef Bacik9ed74f22009-09-11 16:12:44 -0400739
740 if (current->journal_info == trans)
741 current->journal_info = NULL;
Chris Masonab78c842008-07-29 16:15:18 -0400742
Yan, Zheng24bbcf02009-11-12 09:36:34 +0000743 if (throttle)
744 btrfs_run_delayed_iputs(root);
745
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100746 if (trans->aborted ||
Miao Xie87533c42013-01-29 10:14:48 +0000747 test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
Dave Jones4edc2ca2012-04-12 16:03:56 -0400748 err = -EIO;
Jan Schmidtedf39272012-06-28 18:04:55 +0200749 assert_qgroups_uptodate(trans);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100750
Dave Jones4edc2ca2012-04-12 16:03:56 -0400751 kmem_cache_free(btrfs_trans_handle_cachep, trans);
752 return err;
Chris Mason79154b12007-03-22 15:59:16 -0400753}
754
Chris Mason89ce8a62008-06-25 16:01:31 -0400755int btrfs_end_transaction(struct btrfs_trans_handle *trans,
756 struct btrfs_root *root)
757{
Wang Shilong98ad43be2013-04-14 14:08:49 +0000758 return __btrfs_end_transaction(trans, root, 0);
Chris Mason89ce8a62008-06-25 16:01:31 -0400759}
760
761int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
762 struct btrfs_root *root)
763{
Wang Shilong98ad43be2013-04-14 14:08:49 +0000764 return __btrfs_end_transaction(trans, root, 1);
Miao Xie16cdcec2011-04-22 18:12:22 +0800765}
766
767int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans,
768 struct btrfs_root *root)
769{
Miao Xiea698d0752012-09-20 01:51:59 -0600770 return __btrfs_end_transaction(trans, root, 1);
Chris Mason89ce8a62008-06-25 16:01:31 -0400771}
772
Chris Masond352ac62008-09-29 15:18:18 -0400773/*
774 * when btree blocks are allocated, they have some corresponding bits set for
775 * them in one of two extent_io trees. This is used to make sure all of
Chris Mason690587d2009-10-13 13:29:19 -0400776 * those extents are sent to disk but does not wait on them
Chris Masond352ac62008-09-29 15:18:18 -0400777 */
Chris Mason690587d2009-10-13 13:29:19 -0400778int btrfs_write_marked_extents(struct btrfs_root *root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000779 struct extent_io_tree *dirty_pages, int mark)
Chris Mason79154b12007-03-22 15:59:16 -0400780{
Chris Mason777e6bd2008-08-15 15:34:15 -0400781 int err = 0;
Chris Mason7c4452b2007-04-28 09:29:35 -0400782 int werr = 0;
Josef Bacik17283662011-09-26 13:58:47 -0400783 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
Josef Bacike6138872012-09-27 17:07:30 -0400784 struct extent_state *cached_state = NULL;
Chris Mason777e6bd2008-08-15 15:34:15 -0400785 u64 start = 0;
Chris Mason5f39d392007-10-15 16:14:19 -0400786 u64 end;
Chris Mason7c4452b2007-04-28 09:29:35 -0400787
Josef Bacik17283662011-09-26 13:58:47 -0400788 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
Josef Bacike6138872012-09-27 17:07:30 -0400789 mark, &cached_state)) {
790 convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT,
791 mark, &cached_state, GFP_NOFS);
792 cached_state = NULL;
Josef Bacik17283662011-09-26 13:58:47 -0400793 err = filemap_fdatawrite_range(mapping, start, end);
794 if (err)
795 werr = err;
796 cond_resched();
797 start = end + 1;
Chris Mason7c4452b2007-04-28 09:29:35 -0400798 }
Chris Mason690587d2009-10-13 13:29:19 -0400799 if (err)
800 werr = err;
801 return werr;
802}
803
804/*
805 * when btree blocks are allocated, they have some corresponding bits set for
806 * them in one of two extent_io trees. This is used to make sure all of
807 * those extents are on disk for transaction or log commit. We wait
808 * on all the pages and clear them from the dirty pages state tree
809 */
810int btrfs_wait_marked_extents(struct btrfs_root *root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000811 struct extent_io_tree *dirty_pages, int mark)
Chris Mason690587d2009-10-13 13:29:19 -0400812{
Chris Mason690587d2009-10-13 13:29:19 -0400813 int err = 0;
814 int werr = 0;
Josef Bacik17283662011-09-26 13:58:47 -0400815 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
Josef Bacike6138872012-09-27 17:07:30 -0400816 struct extent_state *cached_state = NULL;
Chris Mason690587d2009-10-13 13:29:19 -0400817 u64 start = 0;
818 u64 end;
Chris Mason690587d2009-10-13 13:29:19 -0400819
Josef Bacik17283662011-09-26 13:58:47 -0400820 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
Josef Bacike6138872012-09-27 17:07:30 -0400821 EXTENT_NEED_WAIT, &cached_state)) {
822 clear_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT,
823 0, 0, &cached_state, GFP_NOFS);
Josef Bacik17283662011-09-26 13:58:47 -0400824 err = filemap_fdatawait_range(mapping, start, end);
825 if (err)
826 werr = err;
827 cond_resched();
828 start = end + 1;
Chris Mason777e6bd2008-08-15 15:34:15 -0400829 }
Chris Mason7c4452b2007-04-28 09:29:35 -0400830 if (err)
831 werr = err;
832 return werr;
Chris Mason79154b12007-03-22 15:59:16 -0400833}
834
Chris Mason690587d2009-10-13 13:29:19 -0400835/*
836 * when btree blocks are allocated, they have some corresponding bits set for
837 * them in one of two extent_io trees. This is used to make sure all of
838 * those extents are on disk for transaction or log commit
839 */
Sergei Trofimovich171170c2013-08-14 23:27:46 +0300840static int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000841 struct extent_io_tree *dirty_pages, int mark)
Chris Mason690587d2009-10-13 13:29:19 -0400842{
843 int ret;
844 int ret2;
Miao Xiec6adc9c2013-05-28 10:05:39 +0000845 struct blk_plug plug;
Chris Mason690587d2009-10-13 13:29:19 -0400846
Miao Xiec6adc9c2013-05-28 10:05:39 +0000847 blk_start_plug(&plug);
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000848 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
Miao Xiec6adc9c2013-05-28 10:05:39 +0000849 blk_finish_plug(&plug);
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000850 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
Chris Masonbf0da8c2011-11-04 12:29:37 -0400851
852 if (ret)
853 return ret;
854 if (ret2)
855 return ret2;
856 return 0;
Chris Mason690587d2009-10-13 13:29:19 -0400857}
858
Chris Masond0c803c2008-09-11 16:17:57 -0400859int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
860 struct btrfs_root *root)
861{
862 if (!trans || !trans->transaction) {
863 struct inode *btree_inode;
864 btree_inode = root->fs_info->btree_inode;
865 return filemap_write_and_wait(btree_inode->i_mapping);
866 }
867 return btrfs_write_and_wait_marked_extents(root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000868 &trans->transaction->dirty_pages,
869 EXTENT_DIRTY);
Chris Masond0c803c2008-09-11 16:17:57 -0400870}
871
Chris Masond352ac62008-09-29 15:18:18 -0400872/*
873 * this is used to update the root pointer in the tree of tree roots.
874 *
875 * But, in the case of the extent allocation tree, updating the root
876 * pointer may allocate blocks which may change the root of the extent
877 * allocation tree.
878 *
879 * So, this loops and repeats and makes sure the cowonly root didn't
880 * change while the root pointer was being updated in the metadata.
881 */
Chris Mason0b86a832008-03-24 15:01:56 -0400882static int update_cowonly_root(struct btrfs_trans_handle *trans,
883 struct btrfs_root *root)
884{
885 int ret;
886 u64 old_root_bytenr;
Yan, Zheng86b9f2e2009-11-12 09:36:50 +0000887 u64 old_root_used;
Chris Mason0b86a832008-03-24 15:01:56 -0400888 struct btrfs_root *tree_root = root->fs_info->tree_root;
889
Yan, Zheng86b9f2e2009-11-12 09:36:50 +0000890 old_root_used = btrfs_root_used(&root->root_item);
Chris Mason0b86a832008-03-24 15:01:56 -0400891 btrfs_write_dirty_block_groups(trans, root);
Chris Mason56bec292009-03-13 10:10:06 -0400892
Chris Masond3977122009-01-05 21:25:51 -0500893 while (1) {
Chris Mason0b86a832008-03-24 15:01:56 -0400894 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
Yan, Zheng86b9f2e2009-11-12 09:36:50 +0000895 if (old_root_bytenr == root->node->start &&
896 old_root_used == btrfs_root_used(&root->root_item))
Chris Mason0b86a832008-03-24 15:01:56 -0400897 break;
Chris Mason87ef2bb2008-10-30 11:23:27 -0400898
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400899 btrfs_set_root_node(&root->root_item, root->node);
Chris Mason0b86a832008-03-24 15:01:56 -0400900 ret = btrfs_update_root(trans, tree_root,
901 &root->root_key,
902 &root->root_item);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100903 if (ret)
904 return ret;
Chris Mason56bec292009-03-13 10:10:06 -0400905
Yan, Zheng86b9f2e2009-11-12 09:36:50 +0000906 old_root_used = btrfs_root_used(&root->root_item);
Yan Zheng4a8c9a62009-07-22 10:07:05 -0400907 ret = btrfs_write_dirty_block_groups(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100908 if (ret)
909 return ret;
Chris Mason0b86a832008-03-24 15:01:56 -0400910 }
Yan Zheng276e6802009-07-30 09:40:40 -0400911
912 if (root != root->fs_info->extent_root)
913 switch_commit_root(root);
914
Chris Mason0b86a832008-03-24 15:01:56 -0400915 return 0;
916}
917
Chris Masond352ac62008-09-29 15:18:18 -0400918/*
919 * update all the cowonly tree roots on disk
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100920 *
921 * The error handling in this function may not be obvious. Any of the
922 * failures will cause the file system to go offline. We still need
923 * to clean up the delayed refs.
Chris Masond352ac62008-09-29 15:18:18 -0400924 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400925static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
926 struct btrfs_root *root)
Chris Mason79154b12007-03-22 15:59:16 -0400927{
Chris Mason79154b12007-03-22 15:59:16 -0400928 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Mason0b86a832008-03-24 15:01:56 -0400929 struct list_head *next;
Yan Zheng84234f32008-10-29 14:49:05 -0400930 struct extent_buffer *eb;
Chris Mason56bec292009-03-13 10:10:06 -0400931 int ret;
Yan Zheng84234f32008-10-29 14:49:05 -0400932
Chris Mason56bec292009-03-13 10:10:06 -0400933 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100934 if (ret)
935 return ret;
Chris Mason87ef2bb2008-10-30 11:23:27 -0400936
Yan Zheng84234f32008-10-29 14:49:05 -0400937 eb = btrfs_lock_root_node(fs_info->tree_root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100938 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
939 0, &eb);
Yan Zheng84234f32008-10-29 14:49:05 -0400940 btrfs_tree_unlock(eb);
941 free_extent_buffer(eb);
Chris Mason79154b12007-03-22 15:59:16 -0400942
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100943 if (ret)
944 return ret;
945
Chris Mason56bec292009-03-13 10:10:06 -0400946 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100947 if (ret)
948 return ret;
Chris Mason87ef2bb2008-10-30 11:23:27 -0400949
Stefan Behrens733f4fb2012-05-25 16:06:10 +0200950 ret = btrfs_run_dev_stats(trans, root->fs_info);
Stefan Behrens8dabb742012-11-06 13:15:27 +0100951 WARN_ON(ret);
952 ret = btrfs_run_dev_replace(trans, root->fs_info);
953 WARN_ON(ret);
Stefan Behrens733f4fb2012-05-25 16:06:10 +0200954
Jan Schmidt546adb02012-06-14 16:37:44 +0200955 ret = btrfs_run_qgroups(trans, root->fs_info);
956 BUG_ON(ret);
957
958 /* run_qgroups might have added some more refs */
959 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
960 BUG_ON(ret);
961
Chris Masond3977122009-01-05 21:25:51 -0500962 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
Chris Mason0b86a832008-03-24 15:01:56 -0400963 next = fs_info->dirty_cowonly_roots.next;
964 list_del_init(next);
965 root = list_entry(next, struct btrfs_root, dirty_list);
Chris Mason87ef2bb2008-10-30 11:23:27 -0400966
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100967 ret = update_cowonly_root(trans, root);
968 if (ret)
969 return ret;
Chris Mason79154b12007-03-22 15:59:16 -0400970 }
Yan Zheng276e6802009-07-30 09:40:40 -0400971
972 down_write(&fs_info->extent_commit_sem);
973 switch_commit_root(fs_info->extent_root);
974 up_write(&fs_info->extent_commit_sem);
975
Stefan Behrens8dabb742012-11-06 13:15:27 +0100976 btrfs_after_dev_replace_commit(fs_info);
977
Chris Mason79154b12007-03-22 15:59:16 -0400978 return 0;
979}
980
Chris Masond352ac62008-09-29 15:18:18 -0400981/*
982 * dead roots are old snapshots that need to be deleted. This allocates
983 * a dirty root struct and adds it into the list of dead roots that need to
984 * be deleted
985 */
Josef Bacikcfad3922013-07-25 15:11:47 -0400986void btrfs_add_dead_root(struct btrfs_root *root)
Chris Mason5eda7b52007-06-22 14:16:25 -0400987{
Josef Bacika4abeea2011-04-11 17:25:13 -0400988 spin_lock(&root->fs_info->trans_lock);
Josef Bacikcfad3922013-07-25 15:11:47 -0400989 if (list_empty(&root->root_list))
990 list_add_tail(&root->root_list, &root->fs_info->dead_roots);
Josef Bacika4abeea2011-04-11 17:25:13 -0400991 spin_unlock(&root->fs_info->trans_lock);
Chris Mason5eda7b52007-06-22 14:16:25 -0400992}
993
Chris Masond352ac62008-09-29 15:18:18 -0400994/*
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400995 * update all the cowonly tree roots on disk
Chris Masond352ac62008-09-29 15:18:18 -0400996 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400997static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
998 struct btrfs_root *root)
Chris Mason0f7d52f2007-04-09 10:42:37 -0400999{
Chris Mason0f7d52f2007-04-09 10:42:37 -04001000 struct btrfs_root *gang[8];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001001 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Mason0f7d52f2007-04-09 10:42:37 -04001002 int i;
1003 int ret;
Chris Mason54aa1f42007-06-22 14:16:25 -04001004 int err = 0;
1005
Josef Bacika4abeea2011-04-11 17:25:13 -04001006 spin_lock(&fs_info->fs_roots_radix_lock);
Chris Masond3977122009-01-05 21:25:51 -05001007 while (1) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001008 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1009 (void **)gang, 0,
Chris Mason0f7d52f2007-04-09 10:42:37 -04001010 ARRAY_SIZE(gang),
1011 BTRFS_ROOT_TRANS_TAG);
1012 if (ret == 0)
1013 break;
1014 for (i = 0; i < ret; i++) {
1015 root = gang[i];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001016 radix_tree_tag_clear(&fs_info->fs_roots_radix,
1017 (unsigned long)root->root_key.objectid,
1018 BTRFS_ROOT_TRANS_TAG);
Josef Bacika4abeea2011-04-11 17:25:13 -04001019 spin_unlock(&fs_info->fs_roots_radix_lock);
Yan Zheng31153d82008-07-28 15:32:19 -04001020
Chris Masone02119d2008-09-05 16:13:11 -04001021 btrfs_free_log(trans, root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001022 btrfs_update_reloc_root(trans, root);
Yan, Zhengd68fc572010-05-16 10:49:58 -04001023 btrfs_orphan_commit_root(trans, root);
Chris Masone02119d2008-09-05 16:13:11 -04001024
Li Zefan82d59022011-04-20 10:33:24 +08001025 btrfs_save_ino_cache(root, trans);
1026
Liu Bof1ebcc72011-11-14 20:48:06 -05001027 /* see comments in should_cow_block() */
1028 root->force_cow = 0;
1029 smp_wmb();
1030
Yan Zheng978d9102009-06-15 20:01:02 -04001031 if (root->commit_root != root->node) {
Li Zefan581bb052011-04-20 10:06:11 +08001032 mutex_lock(&root->fs_commit_mutex);
Josef Bacik817d52f2009-07-13 21:29:25 -04001033 switch_commit_root(root);
Li Zefan581bb052011-04-20 10:06:11 +08001034 btrfs_unpin_free_ino(root);
1035 mutex_unlock(&root->fs_commit_mutex);
1036
Yan Zheng978d9102009-06-15 20:01:02 -04001037 btrfs_set_root_node(&root->root_item,
1038 root->node);
1039 }
Chris Mason9f3a7422007-08-07 15:52:19 -04001040
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001041 err = btrfs_update_root(trans, fs_info->tree_root,
Chris Mason0f7d52f2007-04-09 10:42:37 -04001042 &root->root_key,
1043 &root->root_item);
Josef Bacika4abeea2011-04-11 17:25:13 -04001044 spin_lock(&fs_info->fs_roots_radix_lock);
Chris Mason54aa1f42007-06-22 14:16:25 -04001045 if (err)
1046 break;
Chris Mason0f7d52f2007-04-09 10:42:37 -04001047 }
1048 }
Josef Bacika4abeea2011-04-11 17:25:13 -04001049 spin_unlock(&fs_info->fs_roots_radix_lock);
Chris Mason54aa1f42007-06-22 14:16:25 -04001050 return err;
Chris Mason0f7d52f2007-04-09 10:42:37 -04001051}
1052
Chris Masond352ac62008-09-29 15:18:18 -04001053/*
Eric Sandeende78b512013-01-31 18:21:12 +00001054 * defrag a given btree.
1055 * Every leaf in the btree is read and defragged.
Chris Masond352ac62008-09-29 15:18:18 -04001056 */
Eric Sandeende78b512013-01-31 18:21:12 +00001057int btrfs_defrag_root(struct btrfs_root *root)
Chris Masone9d0b132007-08-10 14:06:19 -04001058{
1059 struct btrfs_fs_info *info = root->fs_info;
Chris Masone9d0b132007-08-10 14:06:19 -04001060 struct btrfs_trans_handle *trans;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001061 int ret;
Chris Masone9d0b132007-08-10 14:06:19 -04001062
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001063 if (xchg(&root->defrag_running, 1))
Chris Masone9d0b132007-08-10 14:06:19 -04001064 return 0;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001065
Chris Mason6b800532007-10-15 16:17:34 -04001066 while (1) {
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001067 trans = btrfs_start_transaction(root, 0);
1068 if (IS_ERR(trans))
1069 return PTR_ERR(trans);
1070
Eric Sandeende78b512013-01-31 18:21:12 +00001071 ret = btrfs_defrag_leaves(trans, root);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001072
Chris Masone9d0b132007-08-10 14:06:19 -04001073 btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00001074 btrfs_btree_balance_dirty(info->tree_root);
Chris Masone9d0b132007-08-10 14:06:19 -04001075 cond_resched();
1076
David Sterba7841cb22011-05-31 18:07:27 +02001077 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
Chris Masone9d0b132007-08-10 14:06:19 -04001078 break;
David Sterba210549e2013-02-09 23:38:06 +00001079
1080 if (btrfs_defrag_cancelled(root->fs_info)) {
1081 printk(KERN_DEBUG "btrfs: defrag_root cancelled\n");
1082 ret = -EAGAIN;
1083 break;
1084 }
Chris Masone9d0b132007-08-10 14:06:19 -04001085 }
1086 root->defrag_running = 0;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001087 return ret;
Chris Masone9d0b132007-08-10 14:06:19 -04001088}
1089
Chris Masond352ac62008-09-29 15:18:18 -04001090/*
1091 * new snapshots need to be created at a very specific time in the
Miao Xieaec80302013-03-04 09:44:29 +00001092 * transaction commit. This does the actual creation.
1093 *
1094 * Note:
1095 * If the error which may affect the commitment of the current transaction
1096 * happens, we should return the error number. If the error which just affect
1097 * the creation of the pending snapshots, just return 0.
Chris Masond352ac62008-09-29 15:18:18 -04001098 */
Chris Mason80b67942008-02-01 16:35:04 -05001099static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
Chris Mason3063d292008-01-08 15:46:30 -05001100 struct btrfs_fs_info *fs_info,
1101 struct btrfs_pending_snapshot *pending)
1102{
1103 struct btrfs_key key;
Chris Mason80b67942008-02-01 16:35:04 -05001104 struct btrfs_root_item *new_root_item;
Chris Mason3063d292008-01-08 15:46:30 -05001105 struct btrfs_root *tree_root = fs_info->tree_root;
1106 struct btrfs_root *root = pending->root;
Sage Weil6bdb72d2010-03-15 17:27:13 +00001107 struct btrfs_root *parent_root;
Liu Bo98c99422011-09-11 10:52:24 -04001108 struct btrfs_block_rsv *rsv;
Sage Weil6bdb72d2010-03-15 17:27:13 +00001109 struct inode *parent_inode;
Miao Xie42874b32012-09-06 04:03:32 -06001110 struct btrfs_path *path;
1111 struct btrfs_dir_item *dir_item;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001112 struct dentry *dentry;
Chris Mason3063d292008-01-08 15:46:30 -05001113 struct extent_buffer *tmp;
Chris Mason925baed2008-06-25 16:01:30 -04001114 struct extent_buffer *old;
Alexander Block8ea05e32012-07-25 17:35:53 +02001115 struct timespec cur_time = CURRENT_TIME;
Miao Xieaec80302013-03-04 09:44:29 +00001116 int ret = 0;
Yan, Zhengd68fc572010-05-16 10:49:58 -04001117 u64 to_reserve = 0;
Sage Weil6bdb72d2010-03-15 17:27:13 +00001118 u64 index = 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001119 u64 objectid;
Li Zefanb83cc962010-12-20 16:04:08 +08001120 u64 root_flags;
Alexander Block8ea05e32012-07-25 17:35:53 +02001121 uuid_le new_uuid;
Chris Mason3063d292008-01-08 15:46:30 -05001122
Miao Xie42874b32012-09-06 04:03:32 -06001123 path = btrfs_alloc_path();
1124 if (!path) {
Miao Xieaec80302013-03-04 09:44:29 +00001125 pending->error = -ENOMEM;
1126 return 0;
Miao Xie42874b32012-09-06 04:03:32 -06001127 }
1128
Chris Mason80b67942008-02-01 16:35:04 -05001129 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
1130 if (!new_root_item) {
Miao Xieaec80302013-03-04 09:44:29 +00001131 pending->error = -ENOMEM;
Miao Xie6fa97002012-09-06 04:00:32 -06001132 goto root_item_alloc_fail;
Chris Mason80b67942008-02-01 16:35:04 -05001133 }
Yan, Zhenga22285a2010-05-16 10:48:46 -04001134
Miao Xieaec80302013-03-04 09:44:29 +00001135 pending->error = btrfs_find_free_objectid(tree_root, &objectid);
1136 if (pending->error)
Miao Xie6fa97002012-09-06 04:00:32 -06001137 goto no_free_objectid;
Chris Mason3063d292008-01-08 15:46:30 -05001138
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001139 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
Yan, Zhengd68fc572010-05-16 10:49:58 -04001140
1141 if (to_reserve > 0) {
Miao Xieaec80302013-03-04 09:44:29 +00001142 pending->error = btrfs_block_rsv_add(root,
1143 &pending->block_rsv,
1144 to_reserve,
1145 BTRFS_RESERVE_NO_FLUSH);
1146 if (pending->error)
Miao Xie6fa97002012-09-06 04:00:32 -06001147 goto no_free_objectid;
Yan, Zhengd68fc572010-05-16 10:49:58 -04001148 }
1149
Miao Xieaec80302013-03-04 09:44:29 +00001150 pending->error = btrfs_qgroup_inherit(trans, fs_info,
1151 root->root_key.objectid,
1152 objectid, pending->inherit);
1153 if (pending->error)
Miao Xie6fa97002012-09-06 04:00:32 -06001154 goto no_free_objectid;
Arne Jansen6f72c7e2011-09-14 15:58:21 +02001155
Chris Mason3063d292008-01-08 15:46:30 -05001156 key.objectid = objectid;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001157 key.offset = (u64)-1;
1158 key.type = BTRFS_ROOT_ITEM_KEY;
Chris Mason3063d292008-01-08 15:46:30 -05001159
Miao Xie6fa97002012-09-06 04:00:32 -06001160 rsv = trans->block_rsv;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001161 trans->block_rsv = &pending->block_rsv;
Liu Bo2382c5c2013-02-22 04:33:36 +00001162 trans->bytes_reserved = trans->block_rsv->reserved;
Sage Weil6bdb72d2010-03-15 17:27:13 +00001163
Yan, Zhenga22285a2010-05-16 10:48:46 -04001164 dentry = pending->dentry;
Miao Xiee9662f72013-02-28 10:01:15 +00001165 parent_inode = pending->dir;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001166 parent_root = BTRFS_I(parent_inode)->root;
Chris Mason75857172011-06-13 20:00:16 -04001167 record_root_in_trans(trans, parent_root);
Yan, Zhenga22285a2010-05-16 10:48:46 -04001168
Sage Weil6bdb72d2010-03-15 17:27:13 +00001169 /*
1170 * insert the directory item
1171 */
Sage Weil6bdb72d2010-03-15 17:27:13 +00001172 ret = btrfs_set_inode_index(parent_inode, &index);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001173 BUG_ON(ret); /* -ENOMEM */
Miao Xie42874b32012-09-06 04:03:32 -06001174
1175 /* check if there is a file/dir which has the same name. */
1176 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1177 btrfs_ino(parent_inode),
1178 dentry->d_name.name,
1179 dentry->d_name.len, 0);
1180 if (dir_item != NULL && !IS_ERR(dir_item)) {
Chris Masonfe66a052012-02-20 08:40:56 -05001181 pending->error = -EEXIST;
Miao Xieaec80302013-03-04 09:44:29 +00001182 goto dir_item_existed;
Miao Xie42874b32012-09-06 04:03:32 -06001183 } else if (IS_ERR(dir_item)) {
1184 ret = PTR_ERR(dir_item);
Miao Xie8732d442012-09-17 23:52:38 -06001185 btrfs_abort_transaction(trans, root, ret);
1186 goto fail;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001187 }
Miao Xie42874b32012-09-06 04:03:32 -06001188 btrfs_release_path(path);
Sage Weil6bdb72d2010-03-15 17:27:13 +00001189
Chris Masone9993762011-06-17 16:14:09 -04001190 /*
1191 * pull in the delayed directory update
1192 * and the delayed inode item
1193 * otherwise we corrupt the FS during
1194 * snapshot
1195 */
1196 ret = btrfs_run_delayed_items(trans, root);
Miao Xie8732d442012-09-17 23:52:38 -06001197 if (ret) { /* Transaction aborted */
1198 btrfs_abort_transaction(trans, root, ret);
1199 goto fail;
1200 }
Chris Masone9993762011-06-17 16:14:09 -04001201
Chris Mason75857172011-06-13 20:00:16 -04001202 record_root_in_trans(trans, root);
Sage Weil6bdb72d2010-03-15 17:27:13 +00001203 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1204 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
Li Zefan08fe4db2011-03-28 02:01:25 +00001205 btrfs_check_and_init_root_item(new_root_item);
Sage Weil6bdb72d2010-03-15 17:27:13 +00001206
Li Zefanb83cc962010-12-20 16:04:08 +08001207 root_flags = btrfs_root_flags(new_root_item);
1208 if (pending->readonly)
1209 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1210 else
1211 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1212 btrfs_set_root_flags(new_root_item, root_flags);
1213
Alexander Block8ea05e32012-07-25 17:35:53 +02001214 btrfs_set_root_generation_v2(new_root_item,
1215 trans->transid);
1216 uuid_le_gen(&new_uuid);
1217 memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
1218 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1219 BTRFS_UUID_SIZE);
Stefan Behrens70023da2013-04-17 09:11:47 +00001220 if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1221 memset(new_root_item->received_uuid, 0,
1222 sizeof(new_root_item->received_uuid));
1223 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1224 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1225 btrfs_set_root_stransid(new_root_item, 0);
1226 btrfs_set_root_rtransid(new_root_item, 0);
1227 }
Qu Wenruo3cae2102013-07-16 11:19:18 +08001228 btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1229 btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
Alexander Block8ea05e32012-07-25 17:35:53 +02001230 btrfs_set_root_otransid(new_root_item, trans->transid);
Alexander Block8ea05e32012-07-25 17:35:53 +02001231
Chris Mason925baed2008-06-25 16:01:30 -04001232 old = btrfs_lock_root_node(root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001233 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001234 if (ret) {
1235 btrfs_tree_unlock(old);
1236 free_extent_buffer(old);
Miao Xie8732d442012-09-17 23:52:38 -06001237 btrfs_abort_transaction(trans, root, ret);
1238 goto fail;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001239 }
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001240
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001241 btrfs_set_lock_blocking(old);
Chris Mason3063d292008-01-08 15:46:30 -05001242
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001243 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001244 /* clean up in any case */
Chris Mason925baed2008-06-25 16:01:30 -04001245 btrfs_tree_unlock(old);
1246 free_extent_buffer(old);
Miao Xie8732d442012-09-17 23:52:38 -06001247 if (ret) {
1248 btrfs_abort_transaction(trans, root, ret);
1249 goto fail;
1250 }
Chris Mason3063d292008-01-08 15:46:30 -05001251
Liu Bof1ebcc72011-11-14 20:48:06 -05001252 /* see comments in should_cow_block() */
1253 root->force_cow = 1;
1254 smp_wmb();
1255
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001256 btrfs_set_root_node(new_root_item, tmp);
Yan, Zhenga22285a2010-05-16 10:48:46 -04001257 /* record when the snapshot was created in key.offset */
1258 key.offset = trans->transid;
1259 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
Chris Mason925baed2008-06-25 16:01:30 -04001260 btrfs_tree_unlock(tmp);
Chris Mason3063d292008-01-08 15:46:30 -05001261 free_extent_buffer(tmp);
Miao Xie8732d442012-09-17 23:52:38 -06001262 if (ret) {
1263 btrfs_abort_transaction(trans, root, ret);
1264 goto fail;
1265 }
Chris Mason0660b5a2008-11-17 20:37:39 -05001266
Yan, Zhenga22285a2010-05-16 10:48:46 -04001267 /*
1268 * insert root back/forward references
1269 */
1270 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1271 parent_root->root_key.objectid,
Li Zefan33345d012011-04-20 10:31:50 +08001272 btrfs_ino(parent_inode), index,
Yan, Zhenga22285a2010-05-16 10:48:46 -04001273 dentry->d_name.name, dentry->d_name.len);
Miao Xie8732d442012-09-17 23:52:38 -06001274 if (ret) {
1275 btrfs_abort_transaction(trans, root, ret);
1276 goto fail;
1277 }
Yan, Zhenga22285a2010-05-16 10:48:46 -04001278
1279 key.offset = (u64)-1;
1280 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001281 if (IS_ERR(pending->snap)) {
1282 ret = PTR_ERR(pending->snap);
Miao Xie8732d442012-09-17 23:52:38 -06001283 btrfs_abort_transaction(trans, root, ret);
1284 goto fail;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001285 }
Yan, Zhengd68fc572010-05-16 10:49:58 -04001286
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001287 ret = btrfs_reloc_post_snapshot(trans, pending);
Miao Xie8732d442012-09-17 23:52:38 -06001288 if (ret) {
1289 btrfs_abort_transaction(trans, root, ret);
1290 goto fail;
1291 }
Miao Xie361048f2012-09-06 04:00:57 -06001292
1293 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
Miao Xie8732d442012-09-17 23:52:38 -06001294 if (ret) {
1295 btrfs_abort_transaction(trans, root, ret);
1296 goto fail;
1297 }
Miao Xie42874b32012-09-06 04:03:32 -06001298
1299 ret = btrfs_insert_dir_item(trans, parent_root,
1300 dentry->d_name.name, dentry->d_name.len,
1301 parent_inode, &key,
1302 BTRFS_FT_DIR, index);
1303 /* We have check then name at the beginning, so it is impossible. */
Chris Mason9c520572012-12-17 14:26:57 -05001304 BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
Miao Xie8732d442012-09-17 23:52:38 -06001305 if (ret) {
1306 btrfs_abort_transaction(trans, root, ret);
1307 goto fail;
1308 }
Miao Xie42874b32012-09-06 04:03:32 -06001309
1310 btrfs_i_size_write(parent_inode, parent_inode->i_size +
1311 dentry->d_name.len * 2);
1312 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
Josef Bacikbe6aef62012-10-22 15:43:12 -04001313 ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode);
Stefan Behrensdd5f9612013-08-15 17:11:20 +02001314 if (ret) {
Miao Xie8732d442012-09-17 23:52:38 -06001315 btrfs_abort_transaction(trans, root, ret);
Stefan Behrensdd5f9612013-08-15 17:11:20 +02001316 goto fail;
1317 }
1318 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, new_uuid.b,
1319 BTRFS_UUID_KEY_SUBVOL, objectid);
1320 if (ret) {
1321 btrfs_abort_transaction(trans, root, ret);
1322 goto fail;
1323 }
1324 if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1325 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
1326 new_root_item->received_uuid,
1327 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1328 objectid);
1329 if (ret && ret != -EEXIST) {
1330 btrfs_abort_transaction(trans, root, ret);
1331 goto fail;
1332 }
1333 }
Chris Mason3063d292008-01-08 15:46:30 -05001334fail:
Miao Xieaec80302013-03-04 09:44:29 +00001335 pending->error = ret;
1336dir_item_existed:
Liu Bo98c99422011-09-11 10:52:24 -04001337 trans->block_rsv = rsv;
Liu Bo2382c5c2013-02-22 04:33:36 +00001338 trans->bytes_reserved = 0;
Miao Xie6fa97002012-09-06 04:00:32 -06001339no_free_objectid:
1340 kfree(new_root_item);
1341root_item_alloc_fail:
Miao Xie42874b32012-09-06 04:03:32 -06001342 btrfs_free_path(path);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001343 return ret;
Chris Mason3063d292008-01-08 15:46:30 -05001344}
1345
Chris Masond352ac62008-09-29 15:18:18 -04001346/*
1347 * create all the snapshots we've scheduled for creation
1348 */
Chris Mason80b67942008-02-01 16:35:04 -05001349static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1350 struct btrfs_fs_info *fs_info)
Chris Mason3063d292008-01-08 15:46:30 -05001351{
Miao Xieaec80302013-03-04 09:44:29 +00001352 struct btrfs_pending_snapshot *pending, *next;
Chris Mason3063d292008-01-08 15:46:30 -05001353 struct list_head *head = &trans->transaction->pending_snapshots;
Miao Xieaec80302013-03-04 09:44:29 +00001354 int ret = 0;
Chris Mason3de45862008-11-17 21:02:50 -05001355
Miao Xieaec80302013-03-04 09:44:29 +00001356 list_for_each_entry_safe(pending, next, head, list) {
1357 list_del(&pending->list);
1358 ret = create_pending_snapshot(trans, fs_info, pending);
1359 if (ret)
1360 break;
1361 }
1362 return ret;
Chris Mason3de45862008-11-17 21:02:50 -05001363}
1364
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001365static void update_super_roots(struct btrfs_root *root)
1366{
1367 struct btrfs_root_item *root_item;
1368 struct btrfs_super_block *super;
1369
David Sterba6c417612011-04-13 15:41:04 +02001370 super = root->fs_info->super_copy;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001371
1372 root_item = &root->fs_info->chunk_root->root_item;
1373 super->chunk_root = root_item->bytenr;
1374 super->chunk_root_generation = root_item->generation;
1375 super->chunk_root_level = root_item->level;
1376
1377 root_item = &root->fs_info->tree_root->root_item;
1378 super->root = root_item->bytenr;
1379 super->generation = root_item->generation;
1380 super->root_level = root_item->level;
Josef Bacik73bc1872011-10-03 14:07:49 -04001381 if (btrfs_test_opt(root, SPACE_CACHE))
Josef Bacik0af3d002010-06-21 14:48:16 -04001382 super->cache_generation = root_item->generation;
Stefan Behrens70f80172013-08-15 17:11:23 +02001383 if (root->fs_info->update_uuid_tree_gen)
1384 super->uuid_tree_generation = root_item->generation;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001385}
1386
Chris Masonf36f3042009-07-30 10:04:48 -04001387int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1388{
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001389 struct btrfs_transaction *trans;
Chris Masonf36f3042009-07-30 10:04:48 -04001390 int ret = 0;
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001391
Josef Bacika4abeea2011-04-11 17:25:13 -04001392 spin_lock(&info->trans_lock);
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001393 trans = info->running_transaction;
1394 if (trans)
1395 ret = (trans->state >= TRANS_STATE_COMMIT_START);
Josef Bacika4abeea2011-04-11 17:25:13 -04001396 spin_unlock(&info->trans_lock);
Chris Masonf36f3042009-07-30 10:04:48 -04001397 return ret;
1398}
1399
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001400int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1401{
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001402 struct btrfs_transaction *trans;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001403 int ret = 0;
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001404
Josef Bacika4abeea2011-04-11 17:25:13 -04001405 spin_lock(&info->trans_lock);
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001406 trans = info->running_transaction;
1407 if (trans)
1408 ret = is_transaction_blocked(trans);
Josef Bacika4abeea2011-04-11 17:25:13 -04001409 spin_unlock(&info->trans_lock);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001410 return ret;
1411}
1412
Sage Weilbb9c12c2010-10-29 15:37:34 -04001413/*
1414 * wait for the current transaction commit to start and block subsequent
1415 * transaction joins
1416 */
1417static void wait_current_trans_commit_start(struct btrfs_root *root,
1418 struct btrfs_transaction *trans)
1419{
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001420 wait_event(root->fs_info->transaction_blocked_wait,
Josef Bacik501407a2013-06-10 16:47:23 -04001421 trans->state >= TRANS_STATE_COMMIT_START ||
1422 trans->aborted);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001423}
1424
1425/*
1426 * wait for the current transaction to start and then become unblocked.
1427 * caller holds ref.
1428 */
1429static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1430 struct btrfs_transaction *trans)
1431{
Li Zefan72d63ed2011-07-14 03:17:00 +00001432 wait_event(root->fs_info->transaction_wait,
Josef Bacik501407a2013-06-10 16:47:23 -04001433 trans->state >= TRANS_STATE_UNBLOCKED ||
1434 trans->aborted);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001435}
1436
1437/*
1438 * commit transactions asynchronously. once btrfs_commit_transaction_async
1439 * returns, any subsequent transaction will not be allowed to join.
1440 */
1441struct btrfs_async_commit {
1442 struct btrfs_trans_handle *newtrans;
1443 struct btrfs_root *root;
Miao Xie7892b5a2012-11-15 08:14:47 +00001444 struct work_struct work;
Sage Weilbb9c12c2010-10-29 15:37:34 -04001445};
1446
1447static void do_async_commit(struct work_struct *work)
1448{
1449 struct btrfs_async_commit *ac =
Miao Xie7892b5a2012-11-15 08:14:47 +00001450 container_of(work, struct btrfs_async_commit, work);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001451
Sage Weil6fc4e35482012-08-30 16:26:15 -06001452 /*
1453 * We've got freeze protection passed with the transaction.
1454 * Tell lockdep about it.
1455 */
Miao Xieff7c1d32012-11-26 08:41:29 +00001456 if (ac->newtrans->type < TRANS_JOIN_NOLOCK)
1457 rwsem_acquire_read(
1458 &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1459 0, 1, _THIS_IP_);
Sage Weil6fc4e35482012-08-30 16:26:15 -06001460
Sage Weile209db72012-08-30 16:26:16 -06001461 current->journal_info = ac->newtrans;
1462
Sage Weilbb9c12c2010-10-29 15:37:34 -04001463 btrfs_commit_transaction(ac->newtrans, ac->root);
1464 kfree(ac);
1465}
1466
1467int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1468 struct btrfs_root *root,
1469 int wait_for_unblock)
1470{
1471 struct btrfs_async_commit *ac;
1472 struct btrfs_transaction *cur_trans;
1473
1474 ac = kmalloc(sizeof(*ac), GFP_NOFS);
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00001475 if (!ac)
1476 return -ENOMEM;
Sage Weilbb9c12c2010-10-29 15:37:34 -04001477
Miao Xie7892b5a2012-11-15 08:14:47 +00001478 INIT_WORK(&ac->work, do_async_commit);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001479 ac->root = root;
Josef Bacik7a7eaa42011-04-13 12:54:33 -04001480 ac->newtrans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00001481 if (IS_ERR(ac->newtrans)) {
1482 int err = PTR_ERR(ac->newtrans);
1483 kfree(ac);
1484 return err;
1485 }
Sage Weilbb9c12c2010-10-29 15:37:34 -04001486
1487 /* take transaction reference */
Sage Weilbb9c12c2010-10-29 15:37:34 -04001488 cur_trans = trans->transaction;
Josef Bacik13c5a932011-04-11 15:45:29 -04001489 atomic_inc(&cur_trans->use_count);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001490
1491 btrfs_end_transaction(trans, root);
Sage Weil6fc4e35482012-08-30 16:26:15 -06001492
1493 /*
1494 * Tell lockdep we've released the freeze rwsem, since the
1495 * async commit thread will be the one to unlock it.
1496 */
Miao Xieff7c1d32012-11-26 08:41:29 +00001497 if (trans->type < TRANS_JOIN_NOLOCK)
1498 rwsem_release(
1499 &root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1500 1, _THIS_IP_);
Sage Weil6fc4e35482012-08-30 16:26:15 -06001501
Miao Xie7892b5a2012-11-15 08:14:47 +00001502 schedule_work(&ac->work);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001503
1504 /* wait for transaction to start and unblock */
Sage Weilbb9c12c2010-10-29 15:37:34 -04001505 if (wait_for_unblock)
1506 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1507 else
1508 wait_current_trans_commit_start(root, cur_trans);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001509
Sage Weil38e88052011-06-10 18:43:13 +00001510 if (current->journal_info == trans)
1511 current->journal_info = NULL;
1512
1513 put_transaction(cur_trans);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001514 return 0;
1515}
1516
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001517
1518static void cleanup_transaction(struct btrfs_trans_handle *trans,
Josef Bacik7b8b92a2012-05-31 15:52:43 -04001519 struct btrfs_root *root, int err)
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001520{
1521 struct btrfs_transaction *cur_trans = trans->transaction;
Liu Bof094ac32013-02-27 13:28:25 +00001522 DEFINE_WAIT(wait);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001523
1524 WARN_ON(trans->use_count > 1);
1525
Josef Bacik7b8b92a2012-05-31 15:52:43 -04001526 btrfs_abort_transaction(trans, root, err);
1527
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001528 spin_lock(&root->fs_info->trans_lock);
Liu Bo66b61352013-03-04 16:25:41 +00001529
Miao Xie25d8c282013-05-15 07:48:26 +00001530 /*
1531 * If the transaction is removed from the list, it means this
1532 * transaction has been committed successfully, so it is impossible
1533 * to call the cleanup function.
1534 */
1535 BUG_ON(list_empty(&cur_trans->list));
Liu Bo66b61352013-03-04 16:25:41 +00001536
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001537 list_del_init(&cur_trans->list);
Josef Bacikd7096fc2012-05-31 15:49:57 -04001538 if (cur_trans == root->fs_info->running_transaction) {
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001539 cur_trans->state = TRANS_STATE_COMMIT_DOING;
Liu Bof094ac32013-02-27 13:28:25 +00001540 spin_unlock(&root->fs_info->trans_lock);
1541 wait_event(cur_trans->writer_wait,
1542 atomic_read(&cur_trans->num_writers) == 1);
1543
1544 spin_lock(&root->fs_info->trans_lock);
Josef Bacikd7096fc2012-05-31 15:49:57 -04001545 }
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001546 spin_unlock(&root->fs_info->trans_lock);
1547
1548 btrfs_cleanup_one_transaction(trans->transaction, root);
1549
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001550 spin_lock(&root->fs_info->trans_lock);
1551 if (cur_trans == root->fs_info->running_transaction)
1552 root->fs_info->running_transaction = NULL;
1553 spin_unlock(&root->fs_info->trans_lock);
1554
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001555 put_transaction(cur_trans);
1556 put_transaction(cur_trans);
1557
1558 trace_btrfs_transaction_commit(root);
1559
1560 btrfs_scrub_continue(root);
1561
1562 if (current->journal_info == trans)
1563 current->journal_info = NULL;
1564
1565 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1566}
1567
Miao Xieca469632012-11-01 07:33:14 +00001568static int btrfs_flush_all_pending_stuffs(struct btrfs_trans_handle *trans,
1569 struct btrfs_root *root)
1570{
Miao Xieca469632012-11-01 07:33:14 +00001571 int ret;
1572
Miao Xieca469632012-11-01 07:33:14 +00001573 ret = btrfs_run_delayed_items(trans, root);
1574 if (ret)
1575 return ret;
1576
1577 /*
1578 * running the delayed items may have added new refs. account
1579 * them now so that they hinder processing of more delayed refs
1580 * as little as possible.
1581 */
1582 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
1583
1584 /*
1585 * rename don't use btrfs_join_transaction, so, once we
1586 * set the transaction to blocked above, we aren't going
1587 * to get any new ordered operations. We can safely run
1588 * it here and no for sure that nothing new will be added
1589 * to the list
1590 */
Josef Bacik569e0f32013-02-13 11:09:14 -05001591 ret = btrfs_run_ordered_operations(trans, root, 1);
Miao Xieca469632012-11-01 07:33:14 +00001592
Miao Xieeebc6082013-01-22 10:50:01 +00001593 return ret;
Miao Xieca469632012-11-01 07:33:14 +00001594}
1595
Miao Xie82436612013-05-15 07:48:28 +00001596static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
1597{
1598 if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT))
1599 return btrfs_start_all_delalloc_inodes(fs_info, 1);
1600 return 0;
1601}
1602
1603static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
1604{
1605 if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT))
1606 btrfs_wait_all_ordered_extents(fs_info, 1);
1607}
1608
Chris Mason79154b12007-03-22 15:59:16 -04001609int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1610 struct btrfs_root *root)
1611{
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001612 struct btrfs_transaction *cur_trans = trans->transaction;
Chris Mason8fd17792007-04-19 21:01:03 -04001613 struct btrfs_transaction *prev_trans = NULL;
Miao Xie25287e02012-10-25 09:31:03 +00001614 int ret;
Chris Mason79154b12007-03-22 15:59:16 -04001615
Josef Bacik569e0f32013-02-13 11:09:14 -05001616 ret = btrfs_run_ordered_operations(trans, root, 0);
Miao Xie25287e02012-10-25 09:31:03 +00001617 if (ret) {
1618 btrfs_abort_transaction(trans, root, ret);
Josef Bacike4a2bca2013-02-06 16:55:41 -05001619 btrfs_end_transaction(trans, root);
1620 return ret;
Miao Xie25287e02012-10-25 09:31:03 +00001621 }
1622
Miao Xie8d25a082013-01-15 06:27:25 +00001623 /* Stop the commit early if ->aborted is set */
1624 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
Miao Xie25287e02012-10-25 09:31:03 +00001625 ret = cur_trans->aborted;
Josef Bacike4a2bca2013-02-06 16:55:41 -05001626 btrfs_end_transaction(trans, root);
1627 return ret;
Miao Xie25287e02012-10-25 09:31:03 +00001628 }
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001629
Chris Mason56bec292009-03-13 10:10:06 -04001630 /* make a pass through all the delayed refs we have so far
1631 * any runnings procs may add more while we are here
1632 */
1633 ret = btrfs_run_delayed_refs(trans, root, 0);
Josef Bacike4a2bca2013-02-06 16:55:41 -05001634 if (ret) {
1635 btrfs_end_transaction(trans, root);
1636 return ret;
1637 }
Chris Mason56bec292009-03-13 10:10:06 -04001638
Josef Bacik0e721102012-06-26 16:13:18 -04001639 btrfs_trans_release_metadata(trans, root);
1640 trans->block_rsv = NULL;
Miao Xie272d26d2013-02-20 14:16:39 +00001641 if (trans->qgroup_reserved) {
1642 btrfs_qgroup_free(root, trans->qgroup_reserved);
1643 trans->qgroup_reserved = 0;
1644 }
Josef Bacik0e721102012-06-26 16:13:18 -04001645
Chris Masonb7ec40d2009-03-12 20:12:45 -04001646 cur_trans = trans->transaction;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001647
Chris Mason56bec292009-03-13 10:10:06 -04001648 /*
1649 * set the flushing flag so procs in this transaction have to
1650 * start sending their work down.
1651 */
Chris Masonb7ec40d2009-03-12 20:12:45 -04001652 cur_trans->delayed_refs.flushing = 1;
Josef Bacik1be41b72013-06-12 13:56:06 -04001653 smp_wmb();
Chris Mason56bec292009-03-13 10:10:06 -04001654
Josef Bacikea658ba2012-09-11 16:57:25 -04001655 if (!list_empty(&trans->new_bgs))
1656 btrfs_create_pending_block_groups(trans, root);
1657
Chris Masonc3e69d52009-03-13 10:17:05 -04001658 ret = btrfs_run_delayed_refs(trans, root, 0);
Josef Bacike4a2bca2013-02-06 16:55:41 -05001659 if (ret) {
1660 btrfs_end_transaction(trans, root);
1661 return ret;
1662 }
Chris Mason56bec292009-03-13 10:10:06 -04001663
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001664 spin_lock(&root->fs_info->trans_lock);
1665 if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
1666 spin_unlock(&root->fs_info->trans_lock);
Josef Bacik13c5a932011-04-11 15:45:29 -04001667 atomic_inc(&cur_trans->use_count);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001668 ret = btrfs_end_transaction(trans, root);
Chris Masonccd467d2007-06-28 15:57:36 -04001669
Li Zefanb9c83002011-07-14 03:17:14 +00001670 wait_for_commit(root, cur_trans);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001671
Chris Mason79154b12007-03-22 15:59:16 -04001672 put_transaction(cur_trans);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001673
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001674 return ret;
Chris Mason79154b12007-03-22 15:59:16 -04001675 }
Chris Mason4313b392008-01-03 09:08:48 -05001676
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001677 cur_trans->state = TRANS_STATE_COMMIT_START;
Sage Weilbb9c12c2010-10-29 15:37:34 -04001678 wake_up(&root->fs_info->transaction_blocked_wait);
1679
Chris Masonccd467d2007-06-28 15:57:36 -04001680 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1681 prev_trans = list_entry(cur_trans->list.prev,
1682 struct btrfs_transaction, list);
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001683 if (prev_trans->state != TRANS_STATE_COMPLETED) {
Josef Bacik13c5a932011-04-11 15:45:29 -04001684 atomic_inc(&prev_trans->use_count);
Josef Bacika4abeea2011-04-11 17:25:13 -04001685 spin_unlock(&root->fs_info->trans_lock);
Chris Masonccd467d2007-06-28 15:57:36 -04001686
1687 wait_for_commit(root, prev_trans);
Chris Masonccd467d2007-06-28 15:57:36 -04001688
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001689 put_transaction(prev_trans);
Josef Bacika4abeea2011-04-11 17:25:13 -04001690 } else {
1691 spin_unlock(&root->fs_info->trans_lock);
Chris Masonccd467d2007-06-28 15:57:36 -04001692 }
Josef Bacika4abeea2011-04-11 17:25:13 -04001693 } else {
1694 spin_unlock(&root->fs_info->trans_lock);
Chris Masonccd467d2007-06-28 15:57:36 -04001695 }
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001696
Miao Xie0860adf2013-05-15 07:48:27 +00001697 extwriter_counter_dec(cur_trans, trans->type);
1698
Miao Xie82436612013-05-15 07:48:28 +00001699 ret = btrfs_start_delalloc_flush(root->fs_info);
1700 if (ret)
1701 goto cleanup_transaction;
1702
Miao Xie581227d2013-05-15 07:48:30 +00001703 ret = btrfs_flush_all_pending_stuffs(trans, root);
1704 if (ret)
1705 goto cleanup_transaction;
Chris Mason89573b92009-03-12 20:12:45 -04001706
Miao Xie581227d2013-05-15 07:48:30 +00001707 wait_event(cur_trans->writer_wait,
1708 extwriter_counter_read(cur_trans) == 0);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001709
Miao Xie581227d2013-05-15 07:48:30 +00001710 /* some pending stuffs might be added after the previous flush. */
Miao Xieca469632012-11-01 07:33:14 +00001711 ret = btrfs_flush_all_pending_stuffs(trans, root);
1712 if (ret)
1713 goto cleanup_transaction;
1714
Miao Xie82436612013-05-15 07:48:28 +00001715 btrfs_wait_delalloc_flush(root->fs_info);
Chris Mason75857172011-06-13 20:00:16 -04001716 /*
Josef Baciked0ca142011-06-14 16:22:15 -04001717 * Ok now we need to make sure to block out any other joins while we
1718 * commit the transaction. We could have started a join before setting
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001719 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
Josef Baciked0ca142011-06-14 16:22:15 -04001720 */
1721 spin_lock(&root->fs_info->trans_lock);
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001722 cur_trans->state = TRANS_STATE_COMMIT_DOING;
Josef Baciked0ca142011-06-14 16:22:15 -04001723 spin_unlock(&root->fs_info->trans_lock);
1724 wait_event(cur_trans->writer_wait,
1725 atomic_read(&cur_trans->num_writers) == 1);
1726
Miao Xie2cba30f2013-01-15 06:29:12 +00001727 /* ->aborted might be set after the previous check, so check it */
1728 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1729 ret = cur_trans->aborted;
1730 goto cleanup_transaction;
1731 }
Chris Masone038dca2011-06-17 14:16:13 -04001732 /*
Chris Mason75857172011-06-13 20:00:16 -04001733 * the reloc mutex makes sure that we stop
1734 * the balancing code from coming in and moving
1735 * extents around in the middle of the commit
1736 */
1737 mutex_lock(&root->fs_info->reloc_mutex);
1738
Miao Xie42874b32012-09-06 04:03:32 -06001739 /*
1740 * We needn't worry about the delayed items because we will
1741 * deal with them in create_pending_snapshot(), which is the
1742 * core function of the snapshot creation.
1743 */
1744 ret = create_pending_snapshots(trans, root->fs_info);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001745 if (ret) {
1746 mutex_unlock(&root->fs_info->reloc_mutex);
1747 goto cleanup_transaction;
1748 }
Chris Mason3063d292008-01-08 15:46:30 -05001749
Miao Xie42874b32012-09-06 04:03:32 -06001750 /*
1751 * We insert the dir indexes of the snapshots and update the inode
1752 * of the snapshots' parents after the snapshot creation, so there
1753 * are some delayed items which are not dealt with. Now deal with
1754 * them.
1755 *
1756 * We needn't worry that this operation will corrupt the snapshots,
1757 * because all the tree which are snapshoted will be forced to COW
1758 * the nodes and leaves.
1759 */
1760 ret = btrfs_run_delayed_items(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001761 if (ret) {
1762 mutex_unlock(&root->fs_info->reloc_mutex);
1763 goto cleanup_transaction;
1764 }
Miao Xie16cdcec2011-04-22 18:12:22 +08001765
Chris Mason56bec292009-03-13 10:10:06 -04001766 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001767 if (ret) {
1768 mutex_unlock(&root->fs_info->reloc_mutex);
1769 goto cleanup_transaction;
1770 }
Chris Mason56bec292009-03-13 10:10:06 -04001771
Chris Masone9993762011-06-17 16:14:09 -04001772 /*
1773 * make sure none of the code above managed to slip in a
1774 * delayed item
1775 */
1776 btrfs_assert_delayed_root_empty(root);
1777
Chris Mason2c90e5d2007-04-02 10:50:19 -04001778 WARN_ON(cur_trans != trans->transaction);
Chris Masondc17ff82008-01-08 15:46:30 -05001779
Arne Jansena2de7332011-03-08 14:14:00 +01001780 btrfs_scrub_pause(root);
Chris Masone02119d2008-09-05 16:13:11 -04001781 /* btrfs_commit_tree_roots is responsible for getting the
1782 * various roots consistent with each other. Every pointer
1783 * in the tree of tree roots has to point to the most up to date
1784 * root for every subvolume and other tree. So, we have to keep
1785 * the tree logging code from jumping in and changing any
1786 * of the trees.
1787 *
1788 * At this point in the commit, there can't be any tree-log
1789 * writers, but a little lower down we drop the trans mutex
1790 * and let new people in. By holding the tree_log_mutex
1791 * from now until after the super is written, we avoid races
1792 * with the tree-log code.
1793 */
1794 mutex_lock(&root->fs_info->tree_log_mutex);
Zheng Yan1a40e232008-09-26 10:09:34 -04001795
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001796 ret = commit_fs_roots(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001797 if (ret) {
1798 mutex_unlock(&root->fs_info->tree_log_mutex);
David Sterba871383b2012-04-02 18:31:37 +02001799 mutex_unlock(&root->fs_info->reloc_mutex);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001800 goto cleanup_transaction;
1801 }
Chris Mason54aa1f42007-06-22 14:16:25 -04001802
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001803 /* commit_fs_roots gets rid of all the tree log roots, it is now
Chris Masone02119d2008-09-05 16:13:11 -04001804 * safe to free the root of tree log roots
1805 */
1806 btrfs_free_log_root_tree(trans, root->fs_info);
1807
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001808 ret = commit_cowonly_roots(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001809 if (ret) {
1810 mutex_unlock(&root->fs_info->tree_log_mutex);
David Sterba871383b2012-04-02 18:31:37 +02001811 mutex_unlock(&root->fs_info->reloc_mutex);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001812 goto cleanup_transaction;
1813 }
Chris Mason54aa1f42007-06-22 14:16:25 -04001814
Miao Xie2cba30f2013-01-15 06:29:12 +00001815 /*
1816 * The tasks which save the space cache and inode cache may also
1817 * update ->aborted, check it.
1818 */
1819 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1820 ret = cur_trans->aborted;
1821 mutex_unlock(&root->fs_info->tree_log_mutex);
1822 mutex_unlock(&root->fs_info->reloc_mutex);
1823 goto cleanup_transaction;
1824 }
1825
Yan Zheng11833d62009-09-11 16:11:19 -04001826 btrfs_prepare_extent_commit(trans, root);
1827
Chris Mason78fae272007-03-25 11:35:08 -04001828 cur_trans = root->fs_info->running_transaction;
Chris Mason5f39d392007-10-15 16:14:19 -04001829
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001830 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1831 root->fs_info->tree_root->node);
Josef Bacik817d52f2009-07-13 21:29:25 -04001832 switch_commit_root(root->fs_info->tree_root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001833
1834 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1835 root->fs_info->chunk_root->node);
Josef Bacik817d52f2009-07-13 21:29:25 -04001836 switch_commit_root(root->fs_info->chunk_root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001837
Jan Schmidtedf39272012-06-28 18:04:55 +02001838 assert_qgroups_uptodate(trans);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001839 update_super_roots(root);
Chris Masone02119d2008-09-05 16:13:11 -04001840
1841 if (!root->fs_info->log_root_recovering) {
David Sterba6c417612011-04-13 15:41:04 +02001842 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
1843 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
Chris Masone02119d2008-09-05 16:13:11 -04001844 }
1845
David Sterba6c417612011-04-13 15:41:04 +02001846 memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
1847 sizeof(*root->fs_info->super_copy));
Chris Masonccd467d2007-06-28 15:57:36 -04001848
Josef Bacika4abeea2011-04-11 17:25:13 -04001849 spin_lock(&root->fs_info->trans_lock);
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001850 cur_trans->state = TRANS_STATE_UNBLOCKED;
Josef Bacika4abeea2011-04-11 17:25:13 -04001851 root->fs_info->running_transaction = NULL;
Josef Bacika4abeea2011-04-11 17:25:13 -04001852 spin_unlock(&root->fs_info->trans_lock);
Chris Mason75857172011-06-13 20:00:16 -04001853 mutex_unlock(&root->fs_info->reloc_mutex);
Chris Masonb7ec40d2009-03-12 20:12:45 -04001854
Chris Masonf9295742008-07-17 12:54:14 -04001855 wake_up(&root->fs_info->transaction_wait);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001856
Chris Mason79154b12007-03-22 15:59:16 -04001857 ret = btrfs_write_and_wait_transaction(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001858 if (ret) {
1859 btrfs_error(root->fs_info, ret,
David Sterba08748812013-03-12 14:46:08 +00001860 "Error while writing out transaction");
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001861 mutex_unlock(&root->fs_info->tree_log_mutex);
1862 goto cleanup_transaction;
1863 }
1864
1865 ret = write_ctree_super(trans, root, 0);
1866 if (ret) {
1867 mutex_unlock(&root->fs_info->tree_log_mutex);
1868 goto cleanup_transaction;
1869 }
Chris Mason4313b392008-01-03 09:08:48 -05001870
Chris Masone02119d2008-09-05 16:13:11 -04001871 /*
1872 * the super is written, we can safely allow the tree-loggers
1873 * to go about their business
1874 */
1875 mutex_unlock(&root->fs_info->tree_log_mutex);
1876
Yan Zheng11833d62009-09-11 16:11:19 -04001877 btrfs_finish_extent_commit(trans, root);
Chris Mason4313b392008-01-03 09:08:48 -05001878
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001879 root->fs_info->last_trans_committed = cur_trans->transid;
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001880 /*
1881 * We needn't acquire the lock here because there is no other task
1882 * which can change it.
1883 */
1884 cur_trans->state = TRANS_STATE_COMPLETED;
Chris Mason2c90e5d2007-04-02 10:50:19 -04001885 wake_up(&cur_trans->commit_wait);
Chris Mason3de45862008-11-17 21:02:50 -05001886
Josef Bacika4abeea2011-04-11 17:25:13 -04001887 spin_lock(&root->fs_info->trans_lock);
Josef Bacik13c5a932011-04-11 15:45:29 -04001888 list_del_init(&cur_trans->list);
Josef Bacika4abeea2011-04-11 17:25:13 -04001889 spin_unlock(&root->fs_info->trans_lock);
1890
Chris Mason79154b12007-03-22 15:59:16 -04001891 put_transaction(cur_trans);
Chris Mason78fae272007-03-25 11:35:08 -04001892 put_transaction(cur_trans);
Josef Bacik58176a92007-08-29 15:47:34 -04001893
Miao Xie0860adf2013-05-15 07:48:27 +00001894 if (trans->type & __TRANS_FREEZABLE)
Miao Xie354aa0f2012-09-20 01:54:00 -06001895 sb_end_intwrite(root->fs_info->sb);
Jan Karab2b5ef52012-06-12 16:20:45 +02001896
liubo1abe9b82011-03-24 11:18:59 +00001897 trace_btrfs_transaction_commit(root);
1898
Arne Jansena2de7332011-03-08 14:14:00 +01001899 btrfs_scrub_continue(root);
1900
Josef Bacik9ed74f22009-09-11 16:12:44 -04001901 if (current->journal_info == trans)
1902 current->journal_info = NULL;
1903
Chris Mason2c90e5d2007-04-02 10:50:19 -04001904 kmem_cache_free(btrfs_trans_handle_cachep, trans);
Yan, Zheng24bbcf02009-11-12 09:36:34 +00001905
1906 if (current != root->fs_info->transaction_kthread)
1907 btrfs_run_delayed_iputs(root);
1908
Chris Mason79154b12007-03-22 15:59:16 -04001909 return ret;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001910
1911cleanup_transaction:
Josef Bacik0e721102012-06-26 16:13:18 -04001912 btrfs_trans_release_metadata(trans, root);
1913 trans->block_rsv = NULL;
Miao Xie272d26d2013-02-20 14:16:39 +00001914 if (trans->qgroup_reserved) {
1915 btrfs_qgroup_free(root, trans->qgroup_reserved);
1916 trans->qgroup_reserved = 0;
1917 }
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00001918 btrfs_warn(root->fs_info, "Skipping commit of aborted transaction.");
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001919 if (current->journal_info == trans)
1920 current->journal_info = NULL;
Josef Bacik7b8b92a2012-05-31 15:52:43 -04001921 cleanup_transaction(trans, root, ret);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001922
1923 return ret;
Chris Mason79154b12007-03-22 15:59:16 -04001924}
1925
Chris Masond352ac62008-09-29 15:18:18 -04001926/*
David Sterba9d1a2a32013-03-12 15:13:28 +00001927 * return < 0 if error
1928 * 0 if there are no more dead_roots at the time of call
1929 * 1 there are more to be processed, call me again
1930 *
1931 * The return value indicates there are certainly more snapshots to delete, but
1932 * if there comes a new one during processing, it may return 0. We don't mind,
1933 * because btrfs_commit_super will poke cleaner thread and it will process it a
1934 * few seconds later.
Chris Masond352ac62008-09-29 15:18:18 -04001935 */
David Sterba9d1a2a32013-03-12 15:13:28 +00001936int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root)
Chris Masone9d0b132007-08-10 14:06:19 -04001937{
David Sterba9d1a2a32013-03-12 15:13:28 +00001938 int ret;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001939 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone9d0b132007-08-10 14:06:19 -04001940
Josef Bacika4abeea2011-04-11 17:25:13 -04001941 spin_lock(&fs_info->trans_lock);
David Sterba9d1a2a32013-03-12 15:13:28 +00001942 if (list_empty(&fs_info->dead_roots)) {
1943 spin_unlock(&fs_info->trans_lock);
1944 return 0;
1945 }
1946 root = list_first_entry(&fs_info->dead_roots,
1947 struct btrfs_root, root_list);
Josef Bacikcfad3922013-07-25 15:11:47 -04001948 list_del_init(&root->root_list);
Josef Bacika4abeea2011-04-11 17:25:13 -04001949 spin_unlock(&fs_info->trans_lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001950
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +02001951 pr_debug("btrfs: cleaner removing %llu\n", root->objectid);
Jeff Mahoney2c536792011-10-03 23:22:41 -04001952
David Sterba9d1a2a32013-03-12 15:13:28 +00001953 btrfs_kill_all_delayed_nodes(root);
Yan, Zheng76dda932009-09-21 16:00:26 -04001954
David Sterba9d1a2a32013-03-12 15:13:28 +00001955 if (btrfs_header_backref_rev(root->node) <
1956 BTRFS_MIXED_BACKREF_REV)
1957 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
1958 else
1959 ret = btrfs_drop_snapshot(root, NULL, 1, 0);
1960 /*
1961 * If we encounter a transaction abort during snapshot cleaning, we
1962 * don't want to crash here
1963 */
Josef Bacik6596a9282013-07-31 10:28:05 -04001964 return (ret < 0) ? 0 : 1;
Chris Masone9d0b132007-08-10 14:06:19 -04001965}