blob: 9517de0e668cc9f52664a00964912eec39ddef19 [file] [log] [blame]
Chris Mason6cbd5572007-06-12 09:07:21 -04001/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
Chris Mason79154b12007-03-22 15:59:16 -040019#include <linux/fs.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090020#include <linux/slab.h>
Chris Mason34088782007-06-12 11:36:58 -040021#include <linux/sched.h>
Chris Masond3c2fdcf2007-09-17 10:58:06 -040022#include <linux/writeback.h>
Chris Mason5f39d392007-10-15 16:14:19 -040023#include <linux/pagemap.h>
Chris Mason5f2cc082008-11-07 18:22:45 -050024#include <linux/blkdev.h>
Alexander Block8ea05e32012-07-25 17:35:53 +020025#include <linux/uuid.h>
Chris Mason79154b12007-03-22 15:59:16 -040026#include "ctree.h"
27#include "disk-io.h"
28#include "transaction.h"
Chris Mason925baed2008-06-25 16:01:30 -040029#include "locking.h"
Chris Masone02119d2008-09-05 16:13:11 -040030#include "tree-log.h"
Li Zefan581bb052011-04-20 10:06:11 +080031#include "inode-map.h"
Stefan Behrens733f4fb2012-05-25 16:06:10 +020032#include "volumes.h"
Stefan Behrens8dabb742012-11-06 13:15:27 +010033#include "dev-replace.h"
Josef Bacikfcebe452014-05-13 17:30:47 -070034#include "qgroup.h"
Chris Mason79154b12007-03-22 15:59:16 -040035
Chris Mason0f7d52f2007-04-09 10:42:37 -040036#define BTRFS_ROOT_TRANS_TAG 0
37
David Sterbae8c9f182015-01-02 18:23:10 +010038static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = {
Miao Xie4a9d8bd2013-05-17 03:53:43 +000039 [TRANS_STATE_RUNNING] = 0U,
40 [TRANS_STATE_BLOCKED] = (__TRANS_USERSPACE |
41 __TRANS_START),
42 [TRANS_STATE_COMMIT_START] = (__TRANS_USERSPACE |
43 __TRANS_START |
44 __TRANS_ATTACH),
45 [TRANS_STATE_COMMIT_DOING] = (__TRANS_USERSPACE |
46 __TRANS_START |
47 __TRANS_ATTACH |
48 __TRANS_JOIN),
49 [TRANS_STATE_UNBLOCKED] = (__TRANS_USERSPACE |
50 __TRANS_START |
51 __TRANS_ATTACH |
52 __TRANS_JOIN |
53 __TRANS_JOIN_NOLOCK),
54 [TRANS_STATE_COMPLETED] = (__TRANS_USERSPACE |
55 __TRANS_START |
56 __TRANS_ATTACH |
57 __TRANS_JOIN |
58 __TRANS_JOIN_NOLOCK),
59};
60
Josef Bacik724e2312013-09-30 11:36:38 -040061void btrfs_put_transaction(struct btrfs_transaction *transaction)
Chris Mason79154b12007-03-22 15:59:16 -040062{
Josef Bacik13c5a932011-04-11 15:45:29 -040063 WARN_ON(atomic_read(&transaction->use_count) == 0);
64 if (atomic_dec_and_test(&transaction->use_count)) {
Josef Bacika4abeea2011-04-11 17:25:13 -040065 BUG_ON(!list_empty(&transaction->list));
Liu Boc46effa2013-10-14 12:59:45 +080066 WARN_ON(!RB_EMPTY_ROOT(&transaction->delayed_refs.href_root));
Josef Bacik12621332015-02-03 07:50:16 -080067 if (transaction->delayed_refs.pending_csums)
Jeff Mahoneyab8d0fc2016-09-20 10:05:02 -040068 btrfs_err(transaction->fs_info,
69 "pending csums is %llu",
70 transaction->delayed_refs.pending_csums);
Josef Bacik6df9a952013-06-27 13:22:46 -040071 while (!list_empty(&transaction->pending_chunks)) {
72 struct extent_map *em;
73
74 em = list_first_entry(&transaction->pending_chunks,
75 struct extent_map, list);
76 list_del_init(&em->list);
77 free_extent_map(em);
78 }
Filipe Manana7785a662015-11-27 16:12:00 +000079 /*
80 * If any block groups are found in ->deleted_bgs then it's
81 * because the transaction was aborted and a commit did not
82 * happen (things failed before writing the new superblock
83 * and calling btrfs_finish_extent_commit()), so we can not
84 * discard the physical locations of the block groups.
85 */
86 while (!list_empty(&transaction->deleted_bgs)) {
87 struct btrfs_block_group_cache *cache;
88
89 cache = list_first_entry(&transaction->deleted_bgs,
90 struct btrfs_block_group_cache,
91 bg_list);
92 list_del_init(&cache->bg_list);
93 btrfs_put_block_group_trimming(cache);
94 btrfs_put_block_group(cache);
95 }
Chris Mason2c90e5d2007-04-02 10:50:19 -040096 kmem_cache_free(btrfs_transaction_cachep, transaction);
Chris Mason78fae272007-03-25 11:35:08 -040097 }
Chris Mason79154b12007-03-22 15:59:16 -040098}
99
Filipe Manana663dfbb2014-10-13 12:28:37 +0100100static void clear_btree_io_tree(struct extent_io_tree *tree)
101{
102 spin_lock(&tree->lock);
David Sterbab666a9c2015-10-10 18:24:48 +0200103 /*
104 * Do a single barrier for the waitqueue_active check here, the state
105 * of the waitqueue should not change once clear_btree_io_tree is
106 * called.
107 */
108 smp_mb();
Filipe Manana663dfbb2014-10-13 12:28:37 +0100109 while (!RB_EMPTY_ROOT(&tree->state)) {
110 struct rb_node *node;
111 struct extent_state *state;
112
113 node = rb_first(&tree->state);
114 state = rb_entry(node, struct extent_state, rb_node);
115 rb_erase(&state->rb_node, &tree->state);
116 RB_CLEAR_NODE(&state->rb_node);
117 /*
118 * btree io trees aren't supposed to have tasks waiting for
119 * changes in the flags of extent states ever.
120 */
121 ASSERT(!waitqueue_active(&state->wq));
122 free_extent_state(state);
David Sterba351810c2015-01-08 15:20:54 +0100123
124 cond_resched_lock(&tree->lock);
Filipe Manana663dfbb2014-10-13 12:28:37 +0100125 }
126 spin_unlock(&tree->lock);
127}
128
Josef Bacik9e351cc2014-03-13 15:42:13 -0400129static noinline void switch_commit_roots(struct btrfs_transaction *trans,
130 struct btrfs_fs_info *fs_info)
Josef Bacik817d52f2009-07-13 21:29:25 -0400131{
Josef Bacik9e351cc2014-03-13 15:42:13 -0400132 struct btrfs_root *root, *tmp;
133
134 down_write(&fs_info->commit_root_sem);
135 list_for_each_entry_safe(root, tmp, &trans->switch_commits,
136 dirty_list) {
137 list_del_init(&root->dirty_list);
138 free_extent_buffer(root->commit_root);
139 root->commit_root = btrfs_root_node(root);
140 if (is_fstree(root->objectid))
141 btrfs_unpin_free_ino(root);
Filipe Manana663dfbb2014-10-13 12:28:37 +0100142 clear_btree_io_tree(&root->dirty_log_pages);
Josef Bacik9e351cc2014-03-13 15:42:13 -0400143 }
Josef Bacik2b9dbef2015-09-15 10:07:04 -0400144
145 /* We can free old roots now. */
146 spin_lock(&trans->dropped_roots_lock);
147 while (!list_empty(&trans->dropped_roots)) {
148 root = list_first_entry(&trans->dropped_roots,
149 struct btrfs_root, root_list);
150 list_del_init(&root->root_list);
151 spin_unlock(&trans->dropped_roots_lock);
152 btrfs_drop_and_free_fs_root(fs_info, root);
153 spin_lock(&trans->dropped_roots_lock);
154 }
155 spin_unlock(&trans->dropped_roots_lock);
Josef Bacik9e351cc2014-03-13 15:42:13 -0400156 up_write(&fs_info->commit_root_sem);
Josef Bacik817d52f2009-07-13 21:29:25 -0400157}
158
Miao Xie0860adf2013-05-15 07:48:27 +0000159static inline void extwriter_counter_inc(struct btrfs_transaction *trans,
160 unsigned int type)
161{
162 if (type & TRANS_EXTWRITERS)
163 atomic_inc(&trans->num_extwriters);
164}
165
166static inline void extwriter_counter_dec(struct btrfs_transaction *trans,
167 unsigned int type)
168{
169 if (type & TRANS_EXTWRITERS)
170 atomic_dec(&trans->num_extwriters);
171}
172
173static inline void extwriter_counter_init(struct btrfs_transaction *trans,
174 unsigned int type)
175{
176 atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0));
177}
178
179static inline int extwriter_counter_read(struct btrfs_transaction *trans)
180{
181 return atomic_read(&trans->num_extwriters);
Miao Xie178260b2013-02-20 09:16:24 +0000182}
183
Chris Masond352ac62008-09-29 15:18:18 -0400184/*
185 * either allocate a new transaction or hop into the existing one
186 */
Miao Xie0860adf2013-05-15 07:48:27 +0000187static noinline int join_transaction(struct btrfs_root *root, unsigned int type)
Chris Mason79154b12007-03-22 15:59:16 -0400188{
189 struct btrfs_transaction *cur_trans;
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200190 struct btrfs_fs_info *fs_info = root->fs_info;
Josef Bacika4abeea2011-04-11 17:25:13 -0400191
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200192 spin_lock(&fs_info->trans_lock);
Chris Masond43317d2011-11-06 03:26:19 -0500193loop:
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100194 /* The file system has been taken offline. No new transactions. */
Miao Xie87533c42013-01-29 10:14:48 +0000195 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200196 spin_unlock(&fs_info->trans_lock);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100197 return -EROFS;
198 }
199
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200200 cur_trans = fs_info->running_transaction;
Josef Bacika4abeea2011-04-11 17:25:13 -0400201 if (cur_trans) {
David Sterba871383b2012-04-02 18:31:37 +0200202 if (cur_trans->aborted) {
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200203 spin_unlock(&fs_info->trans_lock);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100204 return cur_trans->aborted;
David Sterba871383b2012-04-02 18:31:37 +0200205 }
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000206 if (btrfs_blocked_trans_types[cur_trans->state] & type) {
Miao Xie178260b2013-02-20 09:16:24 +0000207 spin_unlock(&fs_info->trans_lock);
208 return -EBUSY;
209 }
Josef Bacika4abeea2011-04-11 17:25:13 -0400210 atomic_inc(&cur_trans->use_count);
Josef Bacik13c5a932011-04-11 15:45:29 -0400211 atomic_inc(&cur_trans->num_writers);
Miao Xie0860adf2013-05-15 07:48:27 +0000212 extwriter_counter_inc(cur_trans, type);
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200213 spin_unlock(&fs_info->trans_lock);
Josef Bacika4abeea2011-04-11 17:25:13 -0400214 return 0;
Chris Mason79154b12007-03-22 15:59:16 -0400215 }
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200216 spin_unlock(&fs_info->trans_lock);
Josef Bacika4abeea2011-04-11 17:25:13 -0400217
Miao Xie354aa0f2012-09-20 01:54:00 -0600218 /*
219 * If we are ATTACH, we just want to catch the current transaction,
220 * and commit it. If there is no transaction, just return ENOENT.
221 */
222 if (type == TRANS_ATTACH)
223 return -ENOENT;
224
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000225 /*
226 * JOIN_NOLOCK only happens during the transaction commit, so
227 * it is impossible that ->running_transaction is NULL
228 */
229 BUG_ON(type == TRANS_JOIN_NOLOCK);
230
Josef Bacika4abeea2011-04-11 17:25:13 -0400231 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS);
232 if (!cur_trans)
233 return -ENOMEM;
Chris Masond43317d2011-11-06 03:26:19 -0500234
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200235 spin_lock(&fs_info->trans_lock);
236 if (fs_info->running_transaction) {
Chris Masond43317d2011-11-06 03:26:19 -0500237 /*
238 * someone started a transaction after we unlocked. Make sure
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000239 * to redo the checks above
Chris Masond43317d2011-11-06 03:26:19 -0500240 */
Josef Bacika4abeea2011-04-11 17:25:13 -0400241 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
Chris Masond43317d2011-11-06 03:26:19 -0500242 goto loop;
Miao Xie87533c42013-01-29 10:14:48 +0000243 } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
Dan Carpentere4b50e12012-06-19 13:30:11 +0300244 spin_unlock(&fs_info->trans_lock);
Josef Bacik7b8b92a2012-05-31 15:52:43 -0400245 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
246 return -EROFS;
Josef Bacika4abeea2011-04-11 17:25:13 -0400247 }
Chris Masond43317d2011-11-06 03:26:19 -0500248
Jeff Mahoneyab8d0fc2016-09-20 10:05:02 -0400249 cur_trans->fs_info = fs_info;
Josef Bacika4abeea2011-04-11 17:25:13 -0400250 atomic_set(&cur_trans->num_writers, 1);
Miao Xie0860adf2013-05-15 07:48:27 +0000251 extwriter_counter_init(cur_trans, type);
Josef Bacika4abeea2011-04-11 17:25:13 -0400252 init_waitqueue_head(&cur_trans->writer_wait);
253 init_waitqueue_head(&cur_trans->commit_wait);
Josef Bacik161c35492015-09-24 16:17:39 -0400254 init_waitqueue_head(&cur_trans->pending_wait);
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000255 cur_trans->state = TRANS_STATE_RUNNING;
Josef Bacika4abeea2011-04-11 17:25:13 -0400256 /*
257 * One for this trans handle, one so it will live on until we
258 * commit the transaction.
259 */
260 atomic_set(&cur_trans->use_count, 2);
Josef Bacik161c35492015-09-24 16:17:39 -0400261 atomic_set(&cur_trans->pending_ordered, 0);
Josef Bacik3204d332015-09-24 10:46:10 -0400262 cur_trans->flags = 0;
Josef Bacika4abeea2011-04-11 17:25:13 -0400263 cur_trans->start_time = get_seconds();
264
Alexandru Moisea099d0f2015-09-07 17:24:37 +0300265 memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs));
266
Liu Boc46effa2013-10-14 12:59:45 +0800267 cur_trans->delayed_refs.href_root = RB_ROOT;
Qu Wenruo3368d002015-04-16 14:34:17 +0800268 cur_trans->delayed_refs.dirty_extent_root = RB_ROOT;
Josef Bacikd7df2c72014-01-23 09:21:38 -0500269 atomic_set(&cur_trans->delayed_refs.num_entries, 0);
Jan Schmidt20b297d2012-05-20 15:43:53 +0200270
271 /*
272 * although the tree mod log is per file system and not per transaction,
273 * the log must never go across transaction boundaries.
274 */
275 smp_mb();
Julia Lawall31b1a2b2012-11-03 10:58:34 +0000276 if (!list_empty(&fs_info->tree_mod_seq_list))
Jeff Mahoney5d163e02016-09-20 10:05:00 -0400277 WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n");
Julia Lawall31b1a2b2012-11-03 10:58:34 +0000278 if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log))
Jeff Mahoney5d163e02016-09-20 10:05:00 -0400279 WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n");
Jan Schmidtfc36ed7e2013-04-24 16:57:33 +0000280 atomic64_set(&fs_info->tree_mod_seq, 0);
Jan Schmidt20b297d2012-05-20 15:43:53 +0200281
Josef Bacika4abeea2011-04-11 17:25:13 -0400282 spin_lock_init(&cur_trans->delayed_refs.lock);
283
284 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
Josef Bacik6df9a952013-06-27 13:22:46 -0400285 INIT_LIST_HEAD(&cur_trans->pending_chunks);
Josef Bacik9e351cc2014-03-13 15:42:13 -0400286 INIT_LIST_HEAD(&cur_trans->switch_commits);
Josef Bacikce93ec52014-11-17 15:45:48 -0500287 INIT_LIST_HEAD(&cur_trans->dirty_bgs);
Chris Mason1bbc6212015-04-06 12:46:08 -0700288 INIT_LIST_HEAD(&cur_trans->io_bgs);
Josef Bacik2b9dbef2015-09-15 10:07:04 -0400289 INIT_LIST_HEAD(&cur_trans->dropped_roots);
Chris Mason1bbc6212015-04-06 12:46:08 -0700290 mutex_init(&cur_trans->cache_write_mutex);
Josef Bacikcb723e42015-02-18 08:06:57 -0800291 cur_trans->num_dirty_bgs = 0;
Josef Bacikce93ec52014-11-17 15:45:48 -0500292 spin_lock_init(&cur_trans->dirty_bgs_lock);
Jeff Mahoneye33e17e2015-06-15 09:41:19 -0400293 INIT_LIST_HEAD(&cur_trans->deleted_bgs);
Josef Bacik2b9dbef2015-09-15 10:07:04 -0400294 spin_lock_init(&cur_trans->dropped_roots_lock);
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200295 list_add_tail(&cur_trans->list, &fs_info->trans_list);
Josef Bacika4abeea2011-04-11 17:25:13 -0400296 extent_io_tree_init(&cur_trans->dirty_pages,
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200297 fs_info->btree_inode->i_mapping);
298 fs_info->generation++;
299 cur_trans->transid = fs_info->generation;
300 fs_info->running_transaction = cur_trans;
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100301 cur_trans->aborted = 0;
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200302 spin_unlock(&fs_info->trans_lock);
Josef Bacik15ee9bc2007-08-10 16:22:09 -0400303
Chris Mason79154b12007-03-22 15:59:16 -0400304 return 0;
305}
306
Chris Masond352ac62008-09-29 15:18:18 -0400307/*
Chris Masond3977122009-01-05 21:25:51 -0500308 * this does all the record keeping required to make sure that a reference
309 * counted root is properly recorded in a given transaction. This is required
310 * to make sure the old root from before we joined the transaction is deleted
311 * when the transaction commits
Chris Masond352ac62008-09-29 15:18:18 -0400312 */
Chris Mason75857172011-06-13 20:00:16 -0400313static int record_root_in_trans(struct btrfs_trans_handle *trans,
Qu Wenruo6426c7a2016-05-11 12:53:52 -0700314 struct btrfs_root *root,
315 int force)
Chris Mason6702ed42007-08-07 16:15:09 -0400316{
Qu Wenruo6426c7a2016-05-11 12:53:52 -0700317 if ((test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
318 root->last_trans < trans->transid) || force) {
Chris Mason6702ed42007-08-07 16:15:09 -0400319 WARN_ON(root == root->fs_info->extent_root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400320 WARN_ON(root->commit_root != root->node);
Yan Zheng31153d82008-07-28 15:32:19 -0400321
Chris Mason75857172011-06-13 20:00:16 -0400322 /*
Miao Xie27cdeb72014-04-02 19:51:05 +0800323 * see below for IN_TRANS_SETUP usage rules
Chris Mason75857172011-06-13 20:00:16 -0400324 * we have the reloc mutex held now, so there
325 * is only one writer in this function
326 */
Miao Xie27cdeb72014-04-02 19:51:05 +0800327 set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
Chris Mason75857172011-06-13 20:00:16 -0400328
Miao Xie27cdeb72014-04-02 19:51:05 +0800329 /* make sure readers find IN_TRANS_SETUP before
Chris Mason75857172011-06-13 20:00:16 -0400330 * they find our root->last_trans update
331 */
332 smp_wmb();
333
Josef Bacika4abeea2011-04-11 17:25:13 -0400334 spin_lock(&root->fs_info->fs_roots_radix_lock);
Qu Wenruo6426c7a2016-05-11 12:53:52 -0700335 if (root->last_trans == trans->transid && !force) {
Josef Bacika4abeea2011-04-11 17:25:13 -0400336 spin_unlock(&root->fs_info->fs_roots_radix_lock);
337 return 0;
338 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400339 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
340 (unsigned long)root->root_key.objectid,
341 BTRFS_ROOT_TRANS_TAG);
Josef Bacika4abeea2011-04-11 17:25:13 -0400342 spin_unlock(&root->fs_info->fs_roots_radix_lock);
Chris Mason75857172011-06-13 20:00:16 -0400343 root->last_trans = trans->transid;
344
345 /* this is pretty tricky. We don't want to
346 * take the relocation lock in btrfs_record_root_in_trans
347 * unless we're really doing the first setup for this root in
348 * this transaction.
349 *
350 * Normally we'd use root->last_trans as a flag to decide
351 * if we want to take the expensive mutex.
352 *
353 * But, we have to set root->last_trans before we
354 * init the relocation root, otherwise, we trip over warnings
355 * in ctree.c. The solution used here is to flag ourselves
Miao Xie27cdeb72014-04-02 19:51:05 +0800356 * with root IN_TRANS_SETUP. When this is 1, we're still
Chris Mason75857172011-06-13 20:00:16 -0400357 * fixing up the reloc trees and everyone must wait.
358 *
359 * When this is zero, they can trust root->last_trans and fly
360 * through btrfs_record_root_in_trans without having to take the
361 * lock. smp_wmb() makes sure that all the writes above are
362 * done before we pop in the zero below
363 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400364 btrfs_init_reloc_root(trans, root);
Chris Masonc7548af2014-06-10 13:06:56 -0700365 smp_mb__before_atomic();
Miao Xie27cdeb72014-04-02 19:51:05 +0800366 clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
Chris Mason6702ed42007-08-07 16:15:09 -0400367 }
368 return 0;
369}
370
Chris Mason75857172011-06-13 20:00:16 -0400371
Josef Bacik2b9dbef2015-09-15 10:07:04 -0400372void btrfs_add_dropped_root(struct btrfs_trans_handle *trans,
373 struct btrfs_root *root)
374{
375 struct btrfs_transaction *cur_trans = trans->transaction;
376
377 /* Add ourselves to the transaction dropped list */
378 spin_lock(&cur_trans->dropped_roots_lock);
379 list_add_tail(&root->root_list, &cur_trans->dropped_roots);
380 spin_unlock(&cur_trans->dropped_roots_lock);
381
382 /* Make sure we don't try to update the root at commit time */
383 spin_lock(&root->fs_info->fs_roots_radix_lock);
384 radix_tree_tag_clear(&root->fs_info->fs_roots_radix,
385 (unsigned long)root->root_key.objectid,
386 BTRFS_ROOT_TRANS_TAG);
387 spin_unlock(&root->fs_info->fs_roots_radix_lock);
388}
389
Chris Mason75857172011-06-13 20:00:16 -0400390int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
391 struct btrfs_root *root)
392{
Miao Xie27cdeb72014-04-02 19:51:05 +0800393 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
Chris Mason75857172011-06-13 20:00:16 -0400394 return 0;
395
396 /*
Miao Xie27cdeb72014-04-02 19:51:05 +0800397 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
Chris Mason75857172011-06-13 20:00:16 -0400398 * and barriers
399 */
400 smp_rmb();
401 if (root->last_trans == trans->transid &&
Miao Xie27cdeb72014-04-02 19:51:05 +0800402 !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state))
Chris Mason75857172011-06-13 20:00:16 -0400403 return 0;
404
405 mutex_lock(&root->fs_info->reloc_mutex);
Qu Wenruo6426c7a2016-05-11 12:53:52 -0700406 record_root_in_trans(trans, root, 0);
Chris Mason75857172011-06-13 20:00:16 -0400407 mutex_unlock(&root->fs_info->reloc_mutex);
408
409 return 0;
410}
411
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000412static inline int is_transaction_blocked(struct btrfs_transaction *trans)
413{
414 return (trans->state >= TRANS_STATE_BLOCKED &&
Josef Bacik501407a2013-06-10 16:47:23 -0400415 trans->state < TRANS_STATE_UNBLOCKED &&
416 !trans->aborted);
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000417}
418
Chris Masond352ac62008-09-29 15:18:18 -0400419/* wait for commit against the current transaction to become unblocked
420 * when this is done, it is safe to start a new transaction, but the current
421 * transaction might not be fully on disk.
422 */
Chris Mason37d1aee2008-07-31 10:48:37 -0400423static void wait_current_trans(struct btrfs_root *root)
Chris Mason79154b12007-03-22 15:59:16 -0400424{
Chris Masonf9295742008-07-17 12:54:14 -0400425 struct btrfs_transaction *cur_trans;
Chris Mason79154b12007-03-22 15:59:16 -0400426
Josef Bacika4abeea2011-04-11 17:25:13 -0400427 spin_lock(&root->fs_info->trans_lock);
Chris Masonf9295742008-07-17 12:54:14 -0400428 cur_trans = root->fs_info->running_transaction;
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000429 if (cur_trans && is_transaction_blocked(cur_trans)) {
Josef Bacik13c5a932011-04-11 15:45:29 -0400430 atomic_inc(&cur_trans->use_count);
Josef Bacika4abeea2011-04-11 17:25:13 -0400431 spin_unlock(&root->fs_info->trans_lock);
Li Zefan72d63ed2011-07-14 03:17:00 +0000432
433 wait_event(root->fs_info->transaction_wait,
Josef Bacik501407a2013-06-10 16:47:23 -0400434 cur_trans->state >= TRANS_STATE_UNBLOCKED ||
435 cur_trans->aborted);
Josef Bacik724e2312013-09-30 11:36:38 -0400436 btrfs_put_transaction(cur_trans);
Josef Bacika4abeea2011-04-11 17:25:13 -0400437 } else {
438 spin_unlock(&root->fs_info->trans_lock);
Chris Masonf9295742008-07-17 12:54:14 -0400439 }
Chris Mason37d1aee2008-07-31 10:48:37 -0400440}
441
Yan, Zhenga22285a2010-05-16 10:48:46 -0400442static int may_wait_transaction(struct btrfs_root *root, int type)
Chris Mason37d1aee2008-07-31 10:48:37 -0400443{
Josef Bacikafcdd122016-09-02 15:40:02 -0400444 if (test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags))
Josef Bacika4abeea2011-04-11 17:25:13 -0400445 return 0;
446
447 if (type == TRANS_USERSPACE)
Yan, Zhenga22285a2010-05-16 10:48:46 -0400448 return 1;
Josef Bacika4abeea2011-04-11 17:25:13 -0400449
450 if (type == TRANS_START &&
451 !atomic_read(&root->fs_info->open_ioctl_trans))
452 return 1;
453
Yan, Zhenga22285a2010-05-16 10:48:46 -0400454 return 0;
455}
456
Miao Xie20dd2cb2013-09-25 21:47:45 +0800457static inline bool need_reserve_reloc_root(struct btrfs_root *root)
458{
459 if (!root->fs_info->reloc_ctl ||
Miao Xie27cdeb72014-04-02 19:51:05 +0800460 !test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
Miao Xie20dd2cb2013-09-25 21:47:45 +0800461 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
462 root->reloc_root)
463 return false;
464
465 return true;
466}
467
Miao Xie08e007d2012-10-16 11:33:38 +0000468static struct btrfs_trans_handle *
Alexandru Moise5aed1dd2015-09-22 20:59:15 +0000469start_transaction(struct btrfs_root *root, unsigned int num_items,
470 unsigned int type, enum btrfs_reserve_flush_enum flush)
Yan, Zhenga22285a2010-05-16 10:48:46 -0400471{
472 struct btrfs_trans_handle *h;
473 struct btrfs_transaction *cur_trans;
Josef Bacikb5009942011-06-07 15:07:51 -0400474 u64 num_bytes = 0;
Arne Jansenc5567232011-09-14 15:44:05 +0200475 u64 qgroup_reserved = 0;
Miao Xie20dd2cb2013-09-25 21:47:45 +0800476 bool reloc_reserved = false;
477 int ret;
liuboacce9522011-01-06 19:30:25 +0800478
Filipe Manana46c4e712014-06-24 17:48:28 +0100479 /* Send isn't supposed to start transactions. */
David Sterba2755a0d2014-07-31 00:43:18 +0200480 ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB);
Filipe Manana46c4e712014-06-24 17:48:28 +0100481
Miao Xie87533c42013-01-29 10:14:48 +0000482 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
liuboacce9522011-01-06 19:30:25 +0800483 return ERR_PTR(-EROFS);
Josef Bacik2a1eb462011-04-13 15:15:59 -0400484
Filipe Manana46c4e712014-06-24 17:48:28 +0100485 if (current->journal_info) {
Miao Xie0860adf2013-05-15 07:48:27 +0000486 WARN_ON(type & TRANS_EXTWRITERS);
Josef Bacik2a1eb462011-04-13 15:15:59 -0400487 h = current->journal_info;
488 h->use_count++;
Miao Xieb7d5b0a2012-11-01 07:32:18 +0000489 WARN_ON(h->use_count > 2);
Josef Bacik2a1eb462011-04-13 15:15:59 -0400490 h->orig_rsv = h->block_rsv;
491 h->block_rsv = NULL;
492 goto got_it;
493 }
Josef Bacikb5009942011-06-07 15:07:51 -0400494
495 /*
496 * Do the reservation before we join the transaction so we can do all
497 * the appropriate flushing if need be.
498 */
499 if (num_items > 0 && root != root->fs_info->chunk_root) {
Qu Wenruo71741092015-09-08 17:22:41 +0800500 qgroup_reserved = num_items * root->nodesize;
501 ret = btrfs_qgroup_reserve_meta(root, qgroup_reserved);
502 if (ret)
503 return ERR_PTR(ret);
Arne Jansenc5567232011-09-14 15:44:05 +0200504
Josef Bacikb5009942011-06-07 15:07:51 -0400505 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
Miao Xie20dd2cb2013-09-25 21:47:45 +0800506 /*
507 * Do the reservation for the relocation root creation
508 */
David Sterbaee39b432014-09-30 01:33:33 +0200509 if (need_reserve_reloc_root(root)) {
Miao Xie20dd2cb2013-09-25 21:47:45 +0800510 num_bytes += root->nodesize;
511 reloc_reserved = true;
512 }
513
Miao Xie08e007d2012-10-16 11:33:38 +0000514 ret = btrfs_block_rsv_add(root,
515 &root->fs_info->trans_block_rsv,
516 num_bytes, flush);
Josef Bacikb5009942011-06-07 15:07:51 -0400517 if (ret)
Miao Xie843fcf32013-01-28 12:36:22 +0000518 goto reserve_fail;
Josef Bacikb5009942011-06-07 15:07:51 -0400519 }
Yan, Zhenga22285a2010-05-16 10:48:46 -0400520again:
Alexandru Moisef2f767e2015-08-27 23:53:45 +0000521 h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS);
Miao Xie843fcf32013-01-28 12:36:22 +0000522 if (!h) {
523 ret = -ENOMEM;
524 goto alloc_fail;
525 }
Yan, Zhenga22285a2010-05-16 10:48:46 -0400526
Josef Bacik98114652012-09-14 11:22:38 -0400527 /*
528 * If we are JOIN_NOLOCK we're already committing a transaction and
529 * waiting on this guy, so we don't need to do the sb_start_intwrite
530 * because we're already holding a ref. We need this because we could
531 * have raced in and did an fsync() on a file which can kick a commit
532 * and then we deadlock with somebody doing a freeze.
Miao Xie354aa0f2012-09-20 01:54:00 -0600533 *
534 * If we are ATTACH, it means we just want to catch the current
535 * transaction and commit it, so we needn't do sb_start_intwrite().
Josef Bacik98114652012-09-14 11:22:38 -0400536 */
Miao Xie0860adf2013-05-15 07:48:27 +0000537 if (type & __TRANS_FREEZABLE)
Josef Bacik60376ce2012-09-14 10:34:40 -0400538 sb_start_intwrite(root->fs_info->sb);
Jan Karab2b5ef52012-06-12 16:20:45 +0200539
Yan, Zhenga22285a2010-05-16 10:48:46 -0400540 if (may_wait_transaction(root, type))
Chris Mason37d1aee2008-07-31 10:48:37 -0400541 wait_current_trans(root);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400542
Josef Bacika4abeea2011-04-11 17:25:13 -0400543 do {
Miao Xie354aa0f2012-09-20 01:54:00 -0600544 ret = join_transaction(root, type);
Miao Xie178260b2013-02-20 09:16:24 +0000545 if (ret == -EBUSY) {
Josef Bacika4abeea2011-04-11 17:25:13 -0400546 wait_current_trans(root);
Miao Xie178260b2013-02-20 09:16:24 +0000547 if (unlikely(type == TRANS_ATTACH))
548 ret = -ENOENT;
549 }
Josef Bacika4abeea2011-04-11 17:25:13 -0400550 } while (ret == -EBUSY);
551
Liu Boa43f7f82016-09-13 19:15:48 -0700552 if (ret < 0)
Miao Xie843fcf32013-01-28 12:36:22 +0000553 goto join_fail;
Chris Mason0f7d52f2007-04-09 10:42:37 -0400554
Yan, Zhenga22285a2010-05-16 10:48:46 -0400555 cur_trans = root->fs_info->running_transaction;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400556
557 h->transid = cur_trans->transid;
558 h->transaction = cur_trans;
Arne Jansend13603e2011-09-13 11:40:09 +0200559 h->root = root;
Josef Bacik2a1eb462011-04-13 15:15:59 -0400560 h->use_count = 1;
Jeff Mahoney64b63582016-06-20 17:23:41 -0400561 h->fs_info = root->fs_info;
Qu Wenruo71741092015-09-08 17:22:41 +0800562
Miao Xiea698d0752012-09-20 01:51:59 -0600563 h->type = type;
Filipe Mananad9a05402015-10-03 13:13:13 +0100564 h->can_flush_pending_bgs = true;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200565 INIT_LIST_HEAD(&h->qgroup_ref_list);
Josef Bacikea658ba2012-09-11 16:57:25 -0400566 INIT_LIST_HEAD(&h->new_bgs);
Chris Masonb7ec40d2009-03-12 20:12:45 -0400567
Yan, Zhenga22285a2010-05-16 10:48:46 -0400568 smp_mb();
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000569 if (cur_trans->state >= TRANS_STATE_BLOCKED &&
570 may_wait_transaction(root, type)) {
Filipe Mananaabdd2e82014-06-24 17:46:58 +0100571 current->journal_info = h;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400572 btrfs_commit_transaction(h, root);
573 goto again;
574 }
Josef Bacik9ed74f22009-09-11 16:12:44 -0400575
Josef Bacikb5009942011-06-07 15:07:51 -0400576 if (num_bytes) {
Josef Bacik8c2a3ca2012-01-10 10:31:31 -0500577 trace_btrfs_space_reservation(root->fs_info, "transaction",
Liu Bo2bcc0322012-03-29 09:57:44 -0400578 h->transid, num_bytes, 1);
Josef Bacikb5009942011-06-07 15:07:51 -0400579 h->block_rsv = &root->fs_info->trans_block_rsv;
580 h->bytes_reserved = num_bytes;
Miao Xie20dd2cb2013-09-25 21:47:45 +0800581 h->reloc_reserved = reloc_reserved;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400582 }
583
Josef Bacik2a1eb462011-04-13 15:15:59 -0400584got_it:
Josef Bacika4abeea2011-04-11 17:25:13 -0400585 btrfs_record_root_in_trans(h, root);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400586
587 if (!current->journal_info && type != TRANS_USERSPACE)
588 current->journal_info = h;
Chris Mason79154b12007-03-22 15:59:16 -0400589 return h;
Miao Xie843fcf32013-01-28 12:36:22 +0000590
591join_fail:
Miao Xie0860adf2013-05-15 07:48:27 +0000592 if (type & __TRANS_FREEZABLE)
Miao Xie843fcf32013-01-28 12:36:22 +0000593 sb_end_intwrite(root->fs_info->sb);
594 kmem_cache_free(btrfs_trans_handle_cachep, h);
595alloc_fail:
596 if (num_bytes)
597 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
598 num_bytes);
599reserve_fail:
Qu Wenruo71741092015-09-08 17:22:41 +0800600 btrfs_qgroup_free_meta(root, qgroup_reserved);
Miao Xie843fcf32013-01-28 12:36:22 +0000601 return ERR_PTR(ret);
Chris Mason79154b12007-03-22 15:59:16 -0400602}
603
Chris Masonf9295742008-07-17 12:54:14 -0400604struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
Alexandru Moise5aed1dd2015-09-22 20:59:15 +0000605 unsigned int num_items)
Chris Masonf9295742008-07-17 12:54:14 -0400606{
Miao Xie08e007d2012-10-16 11:33:38 +0000607 return start_transaction(root, num_items, TRANS_START,
608 BTRFS_RESERVE_FLUSH_ALL);
Chris Masonf9295742008-07-17 12:54:14 -0400609}
Filipe Manana8eab77f2015-11-13 23:57:16 +0000610struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv(
611 struct btrfs_root *root,
612 unsigned int num_items,
613 int min_factor)
614{
615 struct btrfs_trans_handle *trans;
616 u64 num_bytes;
617 int ret;
618
619 trans = btrfs_start_transaction(root, num_items);
620 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
621 return trans;
622
623 trans = btrfs_start_transaction(root, 0);
624 if (IS_ERR(trans))
625 return trans;
626
627 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
628 ret = btrfs_cond_migrate_bytes(root->fs_info,
629 &root->fs_info->trans_block_rsv,
630 num_bytes,
631 min_factor);
632 if (ret) {
633 btrfs_end_transaction(trans, root);
634 return ERR_PTR(ret);
635 }
636
637 trans->block_rsv = &root->fs_info->trans_block_rsv;
638 trans->bytes_reserved = num_bytes;
Josef Bacik88d3a5a2016-01-13 13:21:20 -0500639 trace_btrfs_space_reservation(root->fs_info, "transaction",
640 trans->transid, num_bytes, 1);
Filipe Manana8eab77f2015-11-13 23:57:16 +0000641
642 return trans;
643}
Miao Xie8407aa42012-09-07 01:43:32 -0600644
Miao Xie08e007d2012-10-16 11:33:38 +0000645struct btrfs_trans_handle *btrfs_start_transaction_lflush(
Alexandru Moise5aed1dd2015-09-22 20:59:15 +0000646 struct btrfs_root *root,
647 unsigned int num_items)
Miao Xie8407aa42012-09-07 01:43:32 -0600648{
Miao Xie08e007d2012-10-16 11:33:38 +0000649 return start_transaction(root, num_items, TRANS_START,
650 BTRFS_RESERVE_FLUSH_LIMIT);
Miao Xie8407aa42012-09-07 01:43:32 -0600651}
652
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400653struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
Chris Masonf9295742008-07-17 12:54:14 -0400654{
Alexandru Moise575a75d2015-10-25 19:35:44 +0000655 return start_transaction(root, 0, TRANS_JOIN,
656 BTRFS_RESERVE_NO_FLUSH);
Chris Masonf9295742008-07-17 12:54:14 -0400657}
658
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400659struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
Josef Bacik0af3d002010-06-21 14:48:16 -0400660{
Alexandru Moise575a75d2015-10-25 19:35:44 +0000661 return start_transaction(root, 0, TRANS_JOIN_NOLOCK,
662 BTRFS_RESERVE_NO_FLUSH);
Josef Bacik0af3d002010-06-21 14:48:16 -0400663}
664
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400665struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
Sage Weil9ca9ee02008-08-04 10:41:27 -0400666{
Alexandru Moise575a75d2015-10-25 19:35:44 +0000667 return start_transaction(root, 0, TRANS_USERSPACE,
668 BTRFS_RESERVE_NO_FLUSH);
Sage Weil9ca9ee02008-08-04 10:41:27 -0400669}
670
Miao Xied4edf392013-02-20 09:17:06 +0000671/*
672 * btrfs_attach_transaction() - catch the running transaction
673 *
674 * It is used when we want to commit the current the transaction, but
675 * don't want to start a new one.
676 *
677 * Note: If this function return -ENOENT, it just means there is no
678 * running transaction. But it is possible that the inactive transaction
679 * is still in the memory, not fully on disk. If you hope there is no
680 * inactive transaction in the fs when -ENOENT is returned, you should
681 * invoke
682 * btrfs_attach_transaction_barrier()
683 */
Miao Xie354aa0f2012-09-20 01:54:00 -0600684struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
Josef Bacik60376ce2012-09-14 10:34:40 -0400685{
Alexandru Moise575a75d2015-10-25 19:35:44 +0000686 return start_transaction(root, 0, TRANS_ATTACH,
687 BTRFS_RESERVE_NO_FLUSH);
Josef Bacik60376ce2012-09-14 10:34:40 -0400688}
689
Miao Xied4edf392013-02-20 09:17:06 +0000690/*
Wang Sheng-Hui90b6d282013-06-14 16:21:24 +0800691 * btrfs_attach_transaction_barrier() - catch the running transaction
Miao Xied4edf392013-02-20 09:17:06 +0000692 *
693 * It is similar to the above function, the differentia is this one
694 * will wait for all the inactive transactions until they fully
695 * complete.
696 */
697struct btrfs_trans_handle *
698btrfs_attach_transaction_barrier(struct btrfs_root *root)
699{
700 struct btrfs_trans_handle *trans;
701
Alexandru Moise575a75d2015-10-25 19:35:44 +0000702 trans = start_transaction(root, 0, TRANS_ATTACH,
703 BTRFS_RESERVE_NO_FLUSH);
Miao Xied4edf392013-02-20 09:17:06 +0000704 if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT)
705 btrfs_wait_for_commit(root, 0);
706
707 return trans;
708}
709
Chris Masond352ac62008-09-29 15:18:18 -0400710/* wait for a transaction commit to be fully complete */
Li Zefanb9c83002011-07-14 03:17:14 +0000711static noinline void wait_for_commit(struct btrfs_root *root,
Chris Mason89ce8a62008-06-25 16:01:31 -0400712 struct btrfs_transaction *commit)
713{
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000714 wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED);
Chris Mason89ce8a62008-06-25 16:01:31 -0400715}
716
Sage Weil46204592010-10-29 15:41:32 -0400717int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
718{
719 struct btrfs_transaction *cur_trans = NULL, *t;
Miao Xie8cd28072012-11-26 08:42:07 +0000720 int ret = 0;
Sage Weil46204592010-10-29 15:41:32 -0400721
Sage Weil46204592010-10-29 15:41:32 -0400722 if (transid) {
723 if (transid <= root->fs_info->last_trans_committed)
Josef Bacika4abeea2011-04-11 17:25:13 -0400724 goto out;
Sage Weil46204592010-10-29 15:41:32 -0400725
726 /* find specified transaction */
Josef Bacika4abeea2011-04-11 17:25:13 -0400727 spin_lock(&root->fs_info->trans_lock);
Sage Weil46204592010-10-29 15:41:32 -0400728 list_for_each_entry(t, &root->fs_info->trans_list, list) {
729 if (t->transid == transid) {
730 cur_trans = t;
Josef Bacika4abeea2011-04-11 17:25:13 -0400731 atomic_inc(&cur_trans->use_count);
Miao Xie8cd28072012-11-26 08:42:07 +0000732 ret = 0;
Sage Weil46204592010-10-29 15:41:32 -0400733 break;
734 }
Miao Xie8cd28072012-11-26 08:42:07 +0000735 if (t->transid > transid) {
736 ret = 0;
Sage Weil46204592010-10-29 15:41:32 -0400737 break;
Miao Xie8cd28072012-11-26 08:42:07 +0000738 }
Sage Weil46204592010-10-29 15:41:32 -0400739 }
Josef Bacika4abeea2011-04-11 17:25:13 -0400740 spin_unlock(&root->fs_info->trans_lock);
Sage Weil42383022014-09-26 08:30:06 -0700741
742 /*
743 * The specified transaction doesn't exist, or we
744 * raced with btrfs_commit_transaction
745 */
746 if (!cur_trans) {
747 if (transid > root->fs_info->last_trans_committed)
748 ret = -EINVAL;
Miao Xie8cd28072012-11-26 08:42:07 +0000749 goto out;
Sage Weil42383022014-09-26 08:30:06 -0700750 }
Sage Weil46204592010-10-29 15:41:32 -0400751 } else {
752 /* find newest transaction that is committing | committed */
Josef Bacika4abeea2011-04-11 17:25:13 -0400753 spin_lock(&root->fs_info->trans_lock);
Sage Weil46204592010-10-29 15:41:32 -0400754 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
755 list) {
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000756 if (t->state >= TRANS_STATE_COMMIT_START) {
757 if (t->state == TRANS_STATE_COMPLETED)
Josef Bacik3473f3c2011-06-09 10:15:17 -0400758 break;
Sage Weil46204592010-10-29 15:41:32 -0400759 cur_trans = t;
Josef Bacika4abeea2011-04-11 17:25:13 -0400760 atomic_inc(&cur_trans->use_count);
Sage Weil46204592010-10-29 15:41:32 -0400761 break;
762 }
763 }
Josef Bacika4abeea2011-04-11 17:25:13 -0400764 spin_unlock(&root->fs_info->trans_lock);
Sage Weil46204592010-10-29 15:41:32 -0400765 if (!cur_trans)
Josef Bacika4abeea2011-04-11 17:25:13 -0400766 goto out; /* nothing committing|committed */
Sage Weil46204592010-10-29 15:41:32 -0400767 }
768
Sage Weil46204592010-10-29 15:41:32 -0400769 wait_for_commit(root, cur_trans);
Josef Bacik724e2312013-09-30 11:36:38 -0400770 btrfs_put_transaction(cur_trans);
Josef Bacika4abeea2011-04-11 17:25:13 -0400771out:
Sage Weil46204592010-10-29 15:41:32 -0400772 return ret;
773}
774
Chris Mason37d1aee2008-07-31 10:48:37 -0400775void btrfs_throttle(struct btrfs_root *root)
776{
Josef Bacika4abeea2011-04-11 17:25:13 -0400777 if (!atomic_read(&root->fs_info->open_ioctl_trans))
Sage Weil9ca9ee02008-08-04 10:41:27 -0400778 wait_current_trans(root);
Chris Mason37d1aee2008-07-31 10:48:37 -0400779}
780
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400781static int should_end_transaction(struct btrfs_trans_handle *trans,
782 struct btrfs_root *root)
783{
Josef Bacik1be41b72013-06-12 13:56:06 -0400784 if (root->fs_info->global_block_rsv.space_info->full &&
Josef Bacik0a2b2a82014-01-23 10:54:11 -0500785 btrfs_check_space_for_delayed_refs(trans, root))
Josef Bacik1be41b72013-06-12 13:56:06 -0400786 return 1;
Josef Bacik36ba0222011-10-18 12:15:48 -0400787
Josef Bacik1be41b72013-06-12 13:56:06 -0400788 return !!btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400789}
790
791int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
792 struct btrfs_root *root)
793{
794 struct btrfs_transaction *cur_trans = trans->transaction;
795 int updates;
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100796 int err;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400797
Josef Bacika4abeea2011-04-11 17:25:13 -0400798 smp_mb();
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000799 if (cur_trans->state >= TRANS_STATE_BLOCKED ||
800 cur_trans->delayed_refs.flushing)
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400801 return 1;
802
803 updates = trans->delayed_ref_updates;
804 trans->delayed_ref_updates = 0;
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100805 if (updates) {
Chris Mason28ed1342014-12-17 09:41:04 -0800806 err = btrfs_run_delayed_refs(trans, root, updates * 2);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100807 if (err) /* Error code will also eval true */
808 return err;
809 }
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400810
811 return should_end_transaction(trans, root);
812}
813
Chris Mason89ce8a62008-06-25 16:01:31 -0400814static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
Miao Xiea698d0752012-09-20 01:51:59 -0600815 struct btrfs_root *root, int throttle)
Chris Mason79154b12007-03-22 15:59:16 -0400816{
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400817 struct btrfs_transaction *cur_trans = trans->transaction;
Chris Masonab78c842008-07-29 16:15:18 -0400818 struct btrfs_fs_info *info = root->fs_info;
Josef Bacik31b96552016-04-11 17:37:40 -0400819 u64 transid = trans->transid;
Josef Bacik1be41b72013-06-12 13:56:06 -0400820 unsigned long cur = trans->delayed_ref_updates;
Miao Xiea698d0752012-09-20 01:51:59 -0600821 int lock = (trans->type != TRANS_JOIN_NOLOCK);
Dave Jones4edc2ca2012-04-12 16:03:56 -0400822 int err = 0;
Chris Masona79b7d42014-05-22 16:18:52 -0700823 int must_run_delayed_refs = 0;
Chris Masond6e4a422007-04-06 15:37:36 -0400824
Josef Bacik3bbb24b2014-03-06 19:01:07 -0500825 if (trans->use_count > 1) {
826 trans->use_count--;
Josef Bacik2a1eb462011-04-13 15:15:59 -0400827 trans->block_rsv = trans->orig_rsv;
828 return 0;
829 }
830
Josef Bacikb24e03d2011-10-14 14:40:17 -0400831 btrfs_trans_release_metadata(trans, root);
Josef Bacik4c13d752011-08-30 11:31:29 -0400832 trans->block_rsv = NULL;
Arne Jansenc5567232011-09-14 15:44:05 +0200833
Josef Bacikea658ba2012-09-11 16:57:25 -0400834 if (!list_empty(&trans->new_bgs))
835 btrfs_create_pending_block_groups(trans, root);
836
Josef Bacik1be41b72013-06-12 13:56:06 -0400837 trans->delayed_ref_updates = 0;
Chris Masona79b7d42014-05-22 16:18:52 -0700838 if (!trans->sync) {
839 must_run_delayed_refs =
840 btrfs_should_throttle_delayed_refs(trans, root);
Josef Bacik0a2b2a82014-01-23 10:54:11 -0500841 cur = max_t(unsigned long, cur, 32);
Chris Masona79b7d42014-05-22 16:18:52 -0700842
843 /*
844 * don't make the caller wait if they are from a NOLOCK
845 * or ATTACH transaction, it will deadlock with commit
846 */
847 if (must_run_delayed_refs == 1 &&
848 (trans->type & (__TRANS_JOIN_NOLOCK | __TRANS_ATTACH)))
849 must_run_delayed_refs = 2;
Chris Mason56bec292009-03-13 10:10:06 -0400850 }
Chris Masonbb721702013-01-29 18:44:12 -0500851
Josef Bacik0e721102012-06-26 16:13:18 -0400852 btrfs_trans_release_metadata(trans, root);
853 trans->block_rsv = NULL;
Chris Mason56bec292009-03-13 10:10:06 -0400854
Josef Bacikea658ba2012-09-11 16:57:25 -0400855 if (!list_empty(&trans->new_bgs))
856 btrfs_create_pending_block_groups(trans, root);
857
Filipe Manana4fbcdf62015-05-20 14:01:54 +0100858 btrfs_trans_release_chunk_metadata(trans);
859
Josef Bacika4abeea2011-04-11 17:25:13 -0400860 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000861 should_end_transaction(trans, root) &&
862 ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) {
863 spin_lock(&info->trans_lock);
864 if (cur_trans->state == TRANS_STATE_RUNNING)
865 cur_trans->state = TRANS_STATE_BLOCKED;
866 spin_unlock(&info->trans_lock);
Josef Bacika4abeea2011-04-11 17:25:13 -0400867 }
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400868
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000869 if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) {
Josef Bacik3bbb24b2014-03-06 19:01:07 -0500870 if (throttle)
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400871 return btrfs_commit_transaction(trans, root);
Josef Bacik3bbb24b2014-03-06 19:01:07 -0500872 else
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400873 wake_up_process(info->transaction_kthread);
874 }
875
Miao Xie0860adf2013-05-15 07:48:27 +0000876 if (trans->type & __TRANS_FREEZABLE)
Josef Bacik98114652012-09-14 11:22:38 -0400877 sb_end_intwrite(root->fs_info->sb);
Josef Bacik6df78812012-09-05 08:08:30 -0600878
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400879 WARN_ON(cur_trans != info->running_transaction);
Josef Bacik13c5a932011-04-11 15:45:29 -0400880 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
881 atomic_dec(&cur_trans->num_writers);
Miao Xie0860adf2013-05-15 07:48:27 +0000882 extwriter_counter_dec(cur_trans, trans->type);
Chris Mason89ce8a62008-06-25 16:01:31 -0400883
David Sterbaa83342a2015-02-16 19:36:47 +0100884 /*
885 * Make sure counter is updated before we wake up waiters.
886 */
Sage Weil99d16cb2010-10-29 15:37:34 -0400887 smp_mb();
Chris Mason79154b12007-03-22 15:59:16 -0400888 if (waitqueue_active(&cur_trans->writer_wait))
889 wake_up(&cur_trans->writer_wait);
Josef Bacik724e2312013-09-30 11:36:38 -0400890 btrfs_put_transaction(cur_trans);
Josef Bacik9ed74f22009-09-11 16:12:44 -0400891
892 if (current->journal_info == trans)
893 current->journal_info = NULL;
Chris Masonab78c842008-07-29 16:15:18 -0400894
Yan, Zheng24bbcf02009-11-12 09:36:34 +0000895 if (throttle)
896 btrfs_run_delayed_iputs(root);
897
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100898 if (trans->aborted ||
Josef Bacik4e121c02013-09-27 16:32:39 -0400899 test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
900 wake_up_process(info->transaction_kthread);
Dave Jones4edc2ca2012-04-12 16:03:56 -0400901 err = -EIO;
Josef Bacik4e121c02013-09-27 16:32:39 -0400902 }
Jan Schmidtedf39272012-06-28 18:04:55 +0200903 assert_qgroups_uptodate(trans);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100904
Dave Jones4edc2ca2012-04-12 16:03:56 -0400905 kmem_cache_free(btrfs_trans_handle_cachep, trans);
Chris Masona79b7d42014-05-22 16:18:52 -0700906 if (must_run_delayed_refs) {
Josef Bacik31b96552016-04-11 17:37:40 -0400907 btrfs_async_run_delayed_refs(root, cur, transid,
Chris Masona79b7d42014-05-22 16:18:52 -0700908 must_run_delayed_refs == 1);
909 }
Dave Jones4edc2ca2012-04-12 16:03:56 -0400910 return err;
Chris Mason79154b12007-03-22 15:59:16 -0400911}
912
Chris Mason89ce8a62008-06-25 16:01:31 -0400913int btrfs_end_transaction(struct btrfs_trans_handle *trans,
914 struct btrfs_root *root)
915{
Wang Shilong98ad43be2013-04-14 14:08:49 +0000916 return __btrfs_end_transaction(trans, root, 0);
Chris Mason89ce8a62008-06-25 16:01:31 -0400917}
918
919int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
920 struct btrfs_root *root)
921{
Wang Shilong98ad43be2013-04-14 14:08:49 +0000922 return __btrfs_end_transaction(trans, root, 1);
Miao Xie16cdcec2011-04-22 18:12:22 +0800923}
924
Chris Masond352ac62008-09-29 15:18:18 -0400925/*
926 * when btree blocks are allocated, they have some corresponding bits set for
927 * them in one of two extent_io trees. This is used to make sure all of
Chris Mason690587d2009-10-13 13:29:19 -0400928 * those extents are sent to disk but does not wait on them
Chris Masond352ac62008-09-29 15:18:18 -0400929 */
Chris Mason690587d2009-10-13 13:29:19 -0400930int btrfs_write_marked_extents(struct btrfs_root *root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000931 struct extent_io_tree *dirty_pages, int mark)
Chris Mason79154b12007-03-22 15:59:16 -0400932{
Chris Mason777e6bd2008-08-15 15:34:15 -0400933 int err = 0;
Chris Mason7c4452b2007-04-28 09:29:35 -0400934 int werr = 0;
Josef Bacik17283662011-09-26 13:58:47 -0400935 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
Josef Bacike6138872012-09-27 17:07:30 -0400936 struct extent_state *cached_state = NULL;
Chris Mason777e6bd2008-08-15 15:34:15 -0400937 u64 start = 0;
Chris Mason5f39d392007-10-15 16:14:19 -0400938 u64 end;
Chris Mason7c4452b2007-04-28 09:29:35 -0400939
Josef Bacik17283662011-09-26 13:58:47 -0400940 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
Josef Bacike6138872012-09-27 17:07:30 -0400941 mark, &cached_state)) {
Filipe Manana663dfbb2014-10-13 12:28:37 +0100942 bool wait_writeback = false;
943
944 err = convert_extent_bit(dirty_pages, start, end,
945 EXTENT_NEED_WAIT,
David Sterba210aa272016-04-26 23:54:39 +0200946 mark, &cached_state);
Filipe Manana663dfbb2014-10-13 12:28:37 +0100947 /*
948 * convert_extent_bit can return -ENOMEM, which is most of the
949 * time a temporary error. So when it happens, ignore the error
950 * and wait for writeback of this range to finish - because we
951 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
952 * to btrfs_wait_marked_extents() would not know that writeback
953 * for this range started and therefore wouldn't wait for it to
954 * finish - we don't want to commit a superblock that points to
955 * btree nodes/leafs for which writeback hasn't finished yet
956 * (and without errors).
957 * We cleanup any entries left in the io tree when committing
958 * the transaction (through clear_btree_io_tree()).
959 */
960 if (err == -ENOMEM) {
961 err = 0;
962 wait_writeback = true;
963 }
964 if (!err)
965 err = filemap_fdatawrite_range(mapping, start, end);
Josef Bacik17283662011-09-26 13:58:47 -0400966 if (err)
967 werr = err;
Filipe Manana663dfbb2014-10-13 12:28:37 +0100968 else if (wait_writeback)
969 werr = filemap_fdatawait_range(mapping, start, end);
Filipe Mananae38e2ed2014-10-13 12:28:38 +0100970 free_extent_state(cached_state);
Filipe Manana663dfbb2014-10-13 12:28:37 +0100971 cached_state = NULL;
Josef Bacik17283662011-09-26 13:58:47 -0400972 cond_resched();
973 start = end + 1;
Chris Mason7c4452b2007-04-28 09:29:35 -0400974 }
Chris Mason690587d2009-10-13 13:29:19 -0400975 return werr;
976}
977
978/*
979 * when btree blocks are allocated, they have some corresponding bits set for
980 * them in one of two extent_io trees. This is used to make sure all of
981 * those extents are on disk for transaction or log commit. We wait
982 * on all the pages and clear them from the dirty pages state tree
983 */
984int btrfs_wait_marked_extents(struct btrfs_root *root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000985 struct extent_io_tree *dirty_pages, int mark)
Chris Mason690587d2009-10-13 13:29:19 -0400986{
Chris Mason690587d2009-10-13 13:29:19 -0400987 int err = 0;
988 int werr = 0;
Josef Bacik17283662011-09-26 13:58:47 -0400989 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
Josef Bacike6138872012-09-27 17:07:30 -0400990 struct extent_state *cached_state = NULL;
Chris Mason690587d2009-10-13 13:29:19 -0400991 u64 start = 0;
992 u64 end;
Filipe Manana656f30d2014-09-26 12:25:56 +0100993 bool errors = false;
Chris Mason690587d2009-10-13 13:29:19 -0400994
Josef Bacik17283662011-09-26 13:58:47 -0400995 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
Josef Bacike6138872012-09-27 17:07:30 -0400996 EXTENT_NEED_WAIT, &cached_state)) {
Filipe Manana663dfbb2014-10-13 12:28:37 +0100997 /*
998 * Ignore -ENOMEM errors returned by clear_extent_bit().
999 * When committing the transaction, we'll remove any entries
1000 * left in the io tree. For a log commit, we don't remove them
1001 * after committing the log because the tree can be accessed
1002 * concurrently - we do it only at transaction commit time when
1003 * it's safe to do it (through clear_btree_io_tree()).
1004 */
1005 err = clear_extent_bit(dirty_pages, start, end,
1006 EXTENT_NEED_WAIT,
1007 0, 0, &cached_state, GFP_NOFS);
1008 if (err == -ENOMEM)
1009 err = 0;
1010 if (!err)
1011 err = filemap_fdatawait_range(mapping, start, end);
Josef Bacik17283662011-09-26 13:58:47 -04001012 if (err)
1013 werr = err;
Filipe Mananae38e2ed2014-10-13 12:28:38 +01001014 free_extent_state(cached_state);
1015 cached_state = NULL;
Josef Bacik17283662011-09-26 13:58:47 -04001016 cond_resched();
1017 start = end + 1;
Chris Mason777e6bd2008-08-15 15:34:15 -04001018 }
Chris Mason7c4452b2007-04-28 09:29:35 -04001019 if (err)
1020 werr = err;
Filipe Manana656f30d2014-09-26 12:25:56 +01001021
1022 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
1023 if ((mark & EXTENT_DIRTY) &&
Josef Bacikafcdd122016-09-02 15:40:02 -04001024 test_and_clear_bit(BTRFS_FS_LOG1_ERR,
1025 &root->fs_info->flags))
Filipe Manana656f30d2014-09-26 12:25:56 +01001026 errors = true;
1027
1028 if ((mark & EXTENT_NEW) &&
Josef Bacikafcdd122016-09-02 15:40:02 -04001029 test_and_clear_bit(BTRFS_FS_LOG2_ERR,
1030 &root->fs_info->flags))
Filipe Manana656f30d2014-09-26 12:25:56 +01001031 errors = true;
1032 } else {
Josef Bacikafcdd122016-09-02 15:40:02 -04001033 if (test_and_clear_bit(BTRFS_FS_BTREE_ERR,
1034 &root->fs_info->flags))
Filipe Manana656f30d2014-09-26 12:25:56 +01001035 errors = true;
1036 }
1037
1038 if (errors && !werr)
1039 werr = -EIO;
1040
Chris Mason7c4452b2007-04-28 09:29:35 -04001041 return werr;
Chris Mason79154b12007-03-22 15:59:16 -04001042}
1043
Chris Mason690587d2009-10-13 13:29:19 -04001044/*
1045 * when btree blocks are allocated, they have some corresponding bits set for
1046 * them in one of two extent_io trees. This is used to make sure all of
1047 * those extents are on disk for transaction or log commit
1048 */
Sergei Trofimovich171170c2013-08-14 23:27:46 +03001049static int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +00001050 struct extent_io_tree *dirty_pages, int mark)
Chris Mason690587d2009-10-13 13:29:19 -04001051{
1052 int ret;
1053 int ret2;
Miao Xiec6adc9c2013-05-28 10:05:39 +00001054 struct blk_plug plug;
Chris Mason690587d2009-10-13 13:29:19 -04001055
Miao Xiec6adc9c2013-05-28 10:05:39 +00001056 blk_start_plug(&plug);
Yan, Zheng8cef4e12009-11-12 09:33:26 +00001057 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
Miao Xiec6adc9c2013-05-28 10:05:39 +00001058 blk_finish_plug(&plug);
Yan, Zheng8cef4e12009-11-12 09:33:26 +00001059 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
Chris Masonbf0da8c2011-11-04 12:29:37 -04001060
1061 if (ret)
1062 return ret;
1063 if (ret2)
1064 return ret2;
1065 return 0;
Chris Mason690587d2009-10-13 13:29:19 -04001066}
1067
Filipe Manana663dfbb2014-10-13 12:28:37 +01001068static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
Chris Masond0c803c2008-09-11 16:17:57 -04001069 struct btrfs_root *root)
1070{
Filipe Manana663dfbb2014-10-13 12:28:37 +01001071 int ret;
1072
1073 ret = btrfs_write_and_wait_marked_extents(root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +00001074 &trans->transaction->dirty_pages,
1075 EXTENT_DIRTY);
Filipe Manana663dfbb2014-10-13 12:28:37 +01001076 clear_btree_io_tree(&trans->transaction->dirty_pages);
1077
1078 return ret;
Chris Masond0c803c2008-09-11 16:17:57 -04001079}
1080
Chris Masond352ac62008-09-29 15:18:18 -04001081/*
1082 * this is used to update the root pointer in the tree of tree roots.
1083 *
1084 * But, in the case of the extent allocation tree, updating the root
1085 * pointer may allocate blocks which may change the root of the extent
1086 * allocation tree.
1087 *
1088 * So, this loops and repeats and makes sure the cowonly root didn't
1089 * change while the root pointer was being updated in the metadata.
1090 */
Chris Mason0b86a832008-03-24 15:01:56 -04001091static int update_cowonly_root(struct btrfs_trans_handle *trans,
1092 struct btrfs_root *root)
1093{
1094 int ret;
1095 u64 old_root_bytenr;
Yan, Zheng86b9f2e2009-11-12 09:36:50 +00001096 u64 old_root_used;
Chris Mason0b86a832008-03-24 15:01:56 -04001097 struct btrfs_root *tree_root = root->fs_info->tree_root;
1098
Yan, Zheng86b9f2e2009-11-12 09:36:50 +00001099 old_root_used = btrfs_root_used(&root->root_item);
Chris Mason56bec292009-03-13 10:10:06 -04001100
Chris Masond3977122009-01-05 21:25:51 -05001101 while (1) {
Chris Mason0b86a832008-03-24 15:01:56 -04001102 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
Yan, Zheng86b9f2e2009-11-12 09:36:50 +00001103 if (old_root_bytenr == root->node->start &&
Josef Bacikea526d12015-03-13 16:40:45 -04001104 old_root_used == btrfs_root_used(&root->root_item))
Chris Mason0b86a832008-03-24 15:01:56 -04001105 break;
Chris Mason87ef2bb2008-10-30 11:23:27 -04001106
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001107 btrfs_set_root_node(&root->root_item, root->node);
Chris Mason0b86a832008-03-24 15:01:56 -04001108 ret = btrfs_update_root(trans, tree_root,
1109 &root->root_key,
1110 &root->root_item);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001111 if (ret)
1112 return ret;
Chris Mason56bec292009-03-13 10:10:06 -04001113
Yan, Zheng86b9f2e2009-11-12 09:36:50 +00001114 old_root_used = btrfs_root_used(&root->root_item);
Chris Mason0b86a832008-03-24 15:01:56 -04001115 }
Yan Zheng276e6802009-07-30 09:40:40 -04001116
Chris Mason0b86a832008-03-24 15:01:56 -04001117 return 0;
1118}
1119
Chris Masond352ac62008-09-29 15:18:18 -04001120/*
1121 * update all the cowonly tree roots on disk
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001122 *
1123 * The error handling in this function may not be obvious. Any of the
1124 * failures will cause the file system to go offline. We still need
1125 * to clean up the delayed refs.
Chris Masond352ac62008-09-29 15:18:18 -04001126 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001127static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
1128 struct btrfs_root *root)
Chris Mason79154b12007-03-22 15:59:16 -04001129{
Chris Mason79154b12007-03-22 15:59:16 -04001130 struct btrfs_fs_info *fs_info = root->fs_info;
Josef Bacikea526d12015-03-13 16:40:45 -04001131 struct list_head *dirty_bgs = &trans->transaction->dirty_bgs;
Chris Mason1bbc6212015-04-06 12:46:08 -07001132 struct list_head *io_bgs = &trans->transaction->io_bgs;
Chris Mason0b86a832008-03-24 15:01:56 -04001133 struct list_head *next;
Yan Zheng84234f32008-10-29 14:49:05 -04001134 struct extent_buffer *eb;
Chris Mason56bec292009-03-13 10:10:06 -04001135 int ret;
Yan Zheng84234f32008-10-29 14:49:05 -04001136
1137 eb = btrfs_lock_root_node(fs_info->tree_root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001138 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
1139 0, &eb);
Yan Zheng84234f32008-10-29 14:49:05 -04001140 btrfs_tree_unlock(eb);
1141 free_extent_buffer(eb);
Chris Mason79154b12007-03-22 15:59:16 -04001142
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001143 if (ret)
1144 return ret;
1145
Chris Mason56bec292009-03-13 10:10:06 -04001146 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001147 if (ret)
1148 return ret;
Chris Mason87ef2bb2008-10-30 11:23:27 -04001149
Stefan Behrens733f4fb2012-05-25 16:06:10 +02001150 ret = btrfs_run_dev_stats(trans, root->fs_info);
Josef Bacikc16ce192013-09-27 16:38:20 -04001151 if (ret)
1152 return ret;
Stefan Behrens8dabb742012-11-06 13:15:27 +01001153 ret = btrfs_run_dev_replace(trans, root->fs_info);
Josef Bacikc16ce192013-09-27 16:38:20 -04001154 if (ret)
1155 return ret;
Jan Schmidt546adb02012-06-14 16:37:44 +02001156 ret = btrfs_run_qgroups(trans, root->fs_info);
Josef Bacikc16ce192013-09-27 16:38:20 -04001157 if (ret)
1158 return ret;
Jan Schmidt546adb02012-06-14 16:37:44 +02001159
Josef Bacikdcdf7f62015-03-02 16:37:31 -05001160 ret = btrfs_setup_space_cache(trans, root);
1161 if (ret)
1162 return ret;
1163
Jan Schmidt546adb02012-06-14 16:37:44 +02001164 /* run_qgroups might have added some more refs */
1165 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
Josef Bacikc16ce192013-09-27 16:38:20 -04001166 if (ret)
1167 return ret;
Josef Bacikea526d12015-03-13 16:40:45 -04001168again:
Chris Masond3977122009-01-05 21:25:51 -05001169 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
Chris Mason0b86a832008-03-24 15:01:56 -04001170 next = fs_info->dirty_cowonly_roots.next;
1171 list_del_init(next);
1172 root = list_entry(next, struct btrfs_root, dirty_list);
Josef Bacike7070be2014-12-16 08:54:43 -08001173 clear_bit(BTRFS_ROOT_DIRTY, &root->state);
Chris Mason87ef2bb2008-10-30 11:23:27 -04001174
Josef Bacik9e351cc2014-03-13 15:42:13 -04001175 if (root != fs_info->extent_root)
1176 list_add_tail(&root->dirty_list,
1177 &trans->transaction->switch_commits);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001178 ret = update_cowonly_root(trans, root);
1179 if (ret)
1180 return ret;
Josef Bacikea526d12015-03-13 16:40:45 -04001181 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1182 if (ret)
1183 return ret;
Chris Mason79154b12007-03-22 15:59:16 -04001184 }
Yan Zheng276e6802009-07-30 09:40:40 -04001185
Chris Mason1bbc6212015-04-06 12:46:08 -07001186 while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) {
Josef Bacikea526d12015-03-13 16:40:45 -04001187 ret = btrfs_write_dirty_block_groups(trans, root);
1188 if (ret)
1189 return ret;
1190 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1191 if (ret)
1192 return ret;
1193 }
1194
1195 if (!list_empty(&fs_info->dirty_cowonly_roots))
1196 goto again;
1197
Josef Bacik9e351cc2014-03-13 15:42:13 -04001198 list_add_tail(&fs_info->extent_root->dirty_list,
1199 &trans->transaction->switch_commits);
Stefan Behrens8dabb742012-11-06 13:15:27 +01001200 btrfs_after_dev_replace_commit(fs_info);
1201
Chris Mason79154b12007-03-22 15:59:16 -04001202 return 0;
1203}
1204
Chris Masond352ac62008-09-29 15:18:18 -04001205/*
1206 * dead roots are old snapshots that need to be deleted. This allocates
1207 * a dirty root struct and adds it into the list of dead roots that need to
1208 * be deleted
1209 */
Josef Bacikcfad3922013-07-25 15:11:47 -04001210void btrfs_add_dead_root(struct btrfs_root *root)
Chris Mason5eda7b52007-06-22 14:16:25 -04001211{
Josef Bacika4abeea2011-04-11 17:25:13 -04001212 spin_lock(&root->fs_info->trans_lock);
Josef Bacikcfad3922013-07-25 15:11:47 -04001213 if (list_empty(&root->root_list))
1214 list_add_tail(&root->root_list, &root->fs_info->dead_roots);
Josef Bacika4abeea2011-04-11 17:25:13 -04001215 spin_unlock(&root->fs_info->trans_lock);
Chris Mason5eda7b52007-06-22 14:16:25 -04001216}
1217
Chris Masond352ac62008-09-29 15:18:18 -04001218/*
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001219 * update all the cowonly tree roots on disk
Chris Masond352ac62008-09-29 15:18:18 -04001220 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001221static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
1222 struct btrfs_root *root)
Chris Mason0f7d52f2007-04-09 10:42:37 -04001223{
Chris Mason0f7d52f2007-04-09 10:42:37 -04001224 struct btrfs_root *gang[8];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001225 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Mason0f7d52f2007-04-09 10:42:37 -04001226 int i;
1227 int ret;
Chris Mason54aa1f42007-06-22 14:16:25 -04001228 int err = 0;
1229
Josef Bacika4abeea2011-04-11 17:25:13 -04001230 spin_lock(&fs_info->fs_roots_radix_lock);
Chris Masond3977122009-01-05 21:25:51 -05001231 while (1) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001232 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1233 (void **)gang, 0,
Chris Mason0f7d52f2007-04-09 10:42:37 -04001234 ARRAY_SIZE(gang),
1235 BTRFS_ROOT_TRANS_TAG);
1236 if (ret == 0)
1237 break;
1238 for (i = 0; i < ret; i++) {
1239 root = gang[i];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001240 radix_tree_tag_clear(&fs_info->fs_roots_radix,
1241 (unsigned long)root->root_key.objectid,
1242 BTRFS_ROOT_TRANS_TAG);
Josef Bacika4abeea2011-04-11 17:25:13 -04001243 spin_unlock(&fs_info->fs_roots_radix_lock);
Yan Zheng31153d82008-07-28 15:32:19 -04001244
Chris Masone02119d2008-09-05 16:13:11 -04001245 btrfs_free_log(trans, root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001246 btrfs_update_reloc_root(trans, root);
Yan, Zhengd68fc572010-05-16 10:49:58 -04001247 btrfs_orphan_commit_root(trans, root);
Chris Masone02119d2008-09-05 16:13:11 -04001248
Li Zefan82d59022011-04-20 10:33:24 +08001249 btrfs_save_ino_cache(root, trans);
1250
Liu Bof1ebcc72011-11-14 20:48:06 -05001251 /* see comments in should_cow_block() */
Miao Xie27cdeb72014-04-02 19:51:05 +08001252 clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
Chris Masonc7548af2014-06-10 13:06:56 -07001253 smp_mb__after_atomic();
Liu Bof1ebcc72011-11-14 20:48:06 -05001254
Yan Zheng978d9102009-06-15 20:01:02 -04001255 if (root->commit_root != root->node) {
Josef Bacik9e351cc2014-03-13 15:42:13 -04001256 list_add_tail(&root->dirty_list,
1257 &trans->transaction->switch_commits);
Yan Zheng978d9102009-06-15 20:01:02 -04001258 btrfs_set_root_node(&root->root_item,
1259 root->node);
1260 }
Chris Mason9f3a7422007-08-07 15:52:19 -04001261
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001262 err = btrfs_update_root(trans, fs_info->tree_root,
Chris Mason0f7d52f2007-04-09 10:42:37 -04001263 &root->root_key,
1264 &root->root_item);
Josef Bacika4abeea2011-04-11 17:25:13 -04001265 spin_lock(&fs_info->fs_roots_radix_lock);
Chris Mason54aa1f42007-06-22 14:16:25 -04001266 if (err)
1267 break;
Qu Wenruo71741092015-09-08 17:22:41 +08001268 btrfs_qgroup_free_meta_all(root);
Chris Mason0f7d52f2007-04-09 10:42:37 -04001269 }
1270 }
Josef Bacika4abeea2011-04-11 17:25:13 -04001271 spin_unlock(&fs_info->fs_roots_radix_lock);
Chris Mason54aa1f42007-06-22 14:16:25 -04001272 return err;
Chris Mason0f7d52f2007-04-09 10:42:37 -04001273}
1274
Chris Masond352ac62008-09-29 15:18:18 -04001275/*
Eric Sandeende78b512013-01-31 18:21:12 +00001276 * defrag a given btree.
1277 * Every leaf in the btree is read and defragged.
Chris Masond352ac62008-09-29 15:18:18 -04001278 */
Eric Sandeende78b512013-01-31 18:21:12 +00001279int btrfs_defrag_root(struct btrfs_root *root)
Chris Masone9d0b132007-08-10 14:06:19 -04001280{
1281 struct btrfs_fs_info *info = root->fs_info;
Chris Masone9d0b132007-08-10 14:06:19 -04001282 struct btrfs_trans_handle *trans;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001283 int ret;
Chris Masone9d0b132007-08-10 14:06:19 -04001284
Miao Xie27cdeb72014-04-02 19:51:05 +08001285 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state))
Chris Masone9d0b132007-08-10 14:06:19 -04001286 return 0;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001287
Chris Mason6b800532007-10-15 16:17:34 -04001288 while (1) {
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001289 trans = btrfs_start_transaction(root, 0);
1290 if (IS_ERR(trans))
1291 return PTR_ERR(trans);
1292
Eric Sandeende78b512013-01-31 18:21:12 +00001293 ret = btrfs_defrag_leaves(trans, root);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001294
Chris Masone9d0b132007-08-10 14:06:19 -04001295 btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00001296 btrfs_btree_balance_dirty(info->tree_root);
Chris Masone9d0b132007-08-10 14:06:19 -04001297 cond_resched();
1298
Jeff Mahoneyab8d0fc2016-09-20 10:05:02 -04001299 if (btrfs_fs_closing(info) || ret != -EAGAIN)
Chris Masone9d0b132007-08-10 14:06:19 -04001300 break;
David Sterba210549e2013-02-09 23:38:06 +00001301
Jeff Mahoneyab8d0fc2016-09-20 10:05:02 -04001302 if (btrfs_defrag_cancelled(info)) {
1303 btrfs_debug(info, "defrag_root cancelled");
David Sterba210549e2013-02-09 23:38:06 +00001304 ret = -EAGAIN;
1305 break;
1306 }
Chris Masone9d0b132007-08-10 14:06:19 -04001307 }
Miao Xie27cdeb72014-04-02 19:51:05 +08001308 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001309 return ret;
Chris Masone9d0b132007-08-10 14:06:19 -04001310}
1311
Chris Masond352ac62008-09-29 15:18:18 -04001312/*
Qu Wenruo6426c7a2016-05-11 12:53:52 -07001313 * Do all special snapshot related qgroup dirty hack.
1314 *
1315 * Will do all needed qgroup inherit and dirty hack like switch commit
1316 * roots inside one transaction and write all btree into disk, to make
1317 * qgroup works.
1318 */
1319static int qgroup_account_snapshot(struct btrfs_trans_handle *trans,
1320 struct btrfs_root *src,
1321 struct btrfs_root *parent,
1322 struct btrfs_qgroup_inherit *inherit,
1323 u64 dst_objectid)
1324{
1325 struct btrfs_fs_info *fs_info = src->fs_info;
1326 int ret;
1327
1328 /*
1329 * Save some performance in the case that qgroups are not
1330 * enabled. If this check races with the ioctl, rescan will
1331 * kick in anyway.
1332 */
1333 mutex_lock(&fs_info->qgroup_ioctl_lock);
Josef Bacikafcdd122016-09-02 15:40:02 -04001334 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
Qu Wenruo6426c7a2016-05-11 12:53:52 -07001335 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1336 return 0;
1337 }
1338 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1339
1340 /*
1341 * We are going to commit transaction, see btrfs_commit_transaction()
1342 * comment for reason locking tree_log_mutex
1343 */
1344 mutex_lock(&fs_info->tree_log_mutex);
1345
1346 ret = commit_fs_roots(trans, src);
1347 if (ret)
1348 goto out;
1349 ret = btrfs_qgroup_prepare_account_extents(trans, fs_info);
1350 if (ret < 0)
1351 goto out;
1352 ret = btrfs_qgroup_account_extents(trans, fs_info);
1353 if (ret < 0)
1354 goto out;
1355
1356 /* Now qgroup are all updated, we can inherit it to new qgroups */
1357 ret = btrfs_qgroup_inherit(trans, fs_info,
1358 src->root_key.objectid, dst_objectid,
1359 inherit);
1360 if (ret < 0)
1361 goto out;
1362
1363 /*
1364 * Now we do a simplified commit transaction, which will:
1365 * 1) commit all subvolume and extent tree
1366 * To ensure all subvolume and extent tree have a valid
1367 * commit_root to accounting later insert_dir_item()
1368 * 2) write all btree blocks onto disk
1369 * This is to make sure later btree modification will be cowed
1370 * Or commit_root can be populated and cause wrong qgroup numbers
1371 * In this simplified commit, we don't really care about other trees
1372 * like chunk and root tree, as they won't affect qgroup.
1373 * And we don't write super to avoid half committed status.
1374 */
1375 ret = commit_cowonly_roots(trans, src);
1376 if (ret)
1377 goto out;
1378 switch_commit_roots(trans->transaction, fs_info);
1379 ret = btrfs_write_and_wait_transaction(trans, src);
1380 if (ret)
David Sterbaf7af3932016-06-17 18:15:25 +02001381 btrfs_handle_fs_error(fs_info, ret,
Qu Wenruo6426c7a2016-05-11 12:53:52 -07001382 "Error while writing out transaction for qgroup");
1383
1384out:
1385 mutex_unlock(&fs_info->tree_log_mutex);
1386
1387 /*
1388 * Force parent root to be updated, as we recorded it before so its
1389 * last_trans == cur_transid.
1390 * Or it won't be committed again onto disk after later
1391 * insert_dir_item()
1392 */
1393 if (!ret)
1394 record_root_in_trans(trans, parent, 1);
1395 return ret;
1396}
1397
1398/*
Chris Masond352ac62008-09-29 15:18:18 -04001399 * new snapshots need to be created at a very specific time in the
Miao Xieaec80302013-03-04 09:44:29 +00001400 * transaction commit. This does the actual creation.
1401 *
1402 * Note:
1403 * If the error which may affect the commitment of the current transaction
1404 * happens, we should return the error number. If the error which just affect
1405 * the creation of the pending snapshots, just return 0.
Chris Masond352ac62008-09-29 15:18:18 -04001406 */
Chris Mason80b67942008-02-01 16:35:04 -05001407static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
Chris Mason3063d292008-01-08 15:46:30 -05001408 struct btrfs_fs_info *fs_info,
1409 struct btrfs_pending_snapshot *pending)
1410{
1411 struct btrfs_key key;
Chris Mason80b67942008-02-01 16:35:04 -05001412 struct btrfs_root_item *new_root_item;
Chris Mason3063d292008-01-08 15:46:30 -05001413 struct btrfs_root *tree_root = fs_info->tree_root;
1414 struct btrfs_root *root = pending->root;
Sage Weil6bdb72d2010-03-15 17:27:13 +00001415 struct btrfs_root *parent_root;
Liu Bo98c99422011-09-11 10:52:24 -04001416 struct btrfs_block_rsv *rsv;
Sage Weil6bdb72d2010-03-15 17:27:13 +00001417 struct inode *parent_inode;
Miao Xie42874b32012-09-06 04:03:32 -06001418 struct btrfs_path *path;
1419 struct btrfs_dir_item *dir_item;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001420 struct dentry *dentry;
Chris Mason3063d292008-01-08 15:46:30 -05001421 struct extent_buffer *tmp;
Chris Mason925baed2008-06-25 16:01:30 -04001422 struct extent_buffer *old;
Deepa Dinamani04b285f2016-02-06 23:57:21 -08001423 struct timespec cur_time;
Miao Xieaec80302013-03-04 09:44:29 +00001424 int ret = 0;
Yan, Zhengd68fc572010-05-16 10:49:58 -04001425 u64 to_reserve = 0;
Sage Weil6bdb72d2010-03-15 17:27:13 +00001426 u64 index = 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001427 u64 objectid;
Li Zefanb83cc962010-12-20 16:04:08 +08001428 u64 root_flags;
Alexander Block8ea05e32012-07-25 17:35:53 +02001429 uuid_le new_uuid;
Chris Mason3063d292008-01-08 15:46:30 -05001430
David Sterba8546b572015-11-10 18:54:03 +01001431 ASSERT(pending->path);
1432 path = pending->path;
Miao Xie42874b32012-09-06 04:03:32 -06001433
David Sterbab0c0ea62015-11-10 18:54:00 +01001434 ASSERT(pending->root_item);
1435 new_root_item = pending->root_item;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001436
Miao Xieaec80302013-03-04 09:44:29 +00001437 pending->error = btrfs_find_free_objectid(tree_root, &objectid);
1438 if (pending->error)
Miao Xie6fa97002012-09-06 04:00:32 -06001439 goto no_free_objectid;
Chris Mason3063d292008-01-08 15:46:30 -05001440
Qu Wenruod6726332015-04-20 10:09:06 +08001441 /*
1442 * Make qgroup to skip current new snapshot's qgroupid, as it is
1443 * accounted by later btrfs_qgroup_inherit().
1444 */
1445 btrfs_set_skip_qgroup(trans, objectid);
1446
Zhaolei147d2562015-08-06 20:58:11 +08001447 btrfs_reloc_pre_snapshot(pending, &to_reserve);
Yan, Zhengd68fc572010-05-16 10:49:58 -04001448
1449 if (to_reserve > 0) {
Miao Xieaec80302013-03-04 09:44:29 +00001450 pending->error = btrfs_block_rsv_add(root,
1451 &pending->block_rsv,
1452 to_reserve,
1453 BTRFS_RESERVE_NO_FLUSH);
1454 if (pending->error)
Qu Wenruod6726332015-04-20 10:09:06 +08001455 goto clear_skip_qgroup;
Yan, Zhengd68fc572010-05-16 10:49:58 -04001456 }
1457
Chris Mason3063d292008-01-08 15:46:30 -05001458 key.objectid = objectid;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001459 key.offset = (u64)-1;
1460 key.type = BTRFS_ROOT_ITEM_KEY;
Chris Mason3063d292008-01-08 15:46:30 -05001461
Miao Xie6fa97002012-09-06 04:00:32 -06001462 rsv = trans->block_rsv;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001463 trans->block_rsv = &pending->block_rsv;
Liu Bo2382c5c2013-02-22 04:33:36 +00001464 trans->bytes_reserved = trans->block_rsv->reserved;
Josef Bacik88d3a5a2016-01-13 13:21:20 -05001465 trace_btrfs_space_reservation(root->fs_info, "transaction",
1466 trans->transid,
1467 trans->bytes_reserved, 1);
Yan, Zhenga22285a2010-05-16 10:48:46 -04001468 dentry = pending->dentry;
Miao Xiee9662f72013-02-28 10:01:15 +00001469 parent_inode = pending->dir;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001470 parent_root = BTRFS_I(parent_inode)->root;
Qu Wenruo6426c7a2016-05-11 12:53:52 -07001471 record_root_in_trans(trans, parent_root, 0);
Yan, Zhenga22285a2010-05-16 10:48:46 -04001472
Deepa Dinamanic2050a42016-09-14 07:48:06 -07001473 cur_time = current_time(parent_inode);
Deepa Dinamani04b285f2016-02-06 23:57:21 -08001474
Sage Weil6bdb72d2010-03-15 17:27:13 +00001475 /*
1476 * insert the directory item
1477 */
Sage Weil6bdb72d2010-03-15 17:27:13 +00001478 ret = btrfs_set_inode_index(parent_inode, &index);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001479 BUG_ON(ret); /* -ENOMEM */
Miao Xie42874b32012-09-06 04:03:32 -06001480
1481 /* check if there is a file/dir which has the same name. */
1482 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1483 btrfs_ino(parent_inode),
1484 dentry->d_name.name,
1485 dentry->d_name.len, 0);
1486 if (dir_item != NULL && !IS_ERR(dir_item)) {
Chris Masonfe66a052012-02-20 08:40:56 -05001487 pending->error = -EEXIST;
Miao Xieaec80302013-03-04 09:44:29 +00001488 goto dir_item_existed;
Miao Xie42874b32012-09-06 04:03:32 -06001489 } else if (IS_ERR(dir_item)) {
1490 ret = PTR_ERR(dir_item);
Jeff Mahoney66642832016-06-10 18:19:25 -04001491 btrfs_abort_transaction(trans, ret);
Miao Xie8732d442012-09-17 23:52:38 -06001492 goto fail;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001493 }
Miao Xie42874b32012-09-06 04:03:32 -06001494 btrfs_release_path(path);
Sage Weil6bdb72d2010-03-15 17:27:13 +00001495
Chris Masone9993762011-06-17 16:14:09 -04001496 /*
1497 * pull in the delayed directory update
1498 * and the delayed inode item
1499 * otherwise we corrupt the FS during
1500 * snapshot
1501 */
1502 ret = btrfs_run_delayed_items(trans, root);
Miao Xie8732d442012-09-17 23:52:38 -06001503 if (ret) { /* Transaction aborted */
Jeff Mahoney66642832016-06-10 18:19:25 -04001504 btrfs_abort_transaction(trans, ret);
Miao Xie8732d442012-09-17 23:52:38 -06001505 goto fail;
1506 }
Chris Masone9993762011-06-17 16:14:09 -04001507
Qu Wenruo6426c7a2016-05-11 12:53:52 -07001508 record_root_in_trans(trans, root, 0);
Sage Weil6bdb72d2010-03-15 17:27:13 +00001509 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1510 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
Li Zefan08fe4db2011-03-28 02:01:25 +00001511 btrfs_check_and_init_root_item(new_root_item);
Sage Weil6bdb72d2010-03-15 17:27:13 +00001512
Li Zefanb83cc962010-12-20 16:04:08 +08001513 root_flags = btrfs_root_flags(new_root_item);
1514 if (pending->readonly)
1515 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1516 else
1517 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1518 btrfs_set_root_flags(new_root_item, root_flags);
1519
Alexander Block8ea05e32012-07-25 17:35:53 +02001520 btrfs_set_root_generation_v2(new_root_item,
1521 trans->transid);
1522 uuid_le_gen(&new_uuid);
1523 memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
1524 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1525 BTRFS_UUID_SIZE);
Stefan Behrens70023da2013-04-17 09:11:47 +00001526 if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1527 memset(new_root_item->received_uuid, 0,
1528 sizeof(new_root_item->received_uuid));
1529 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1530 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1531 btrfs_set_root_stransid(new_root_item, 0);
1532 btrfs_set_root_rtransid(new_root_item, 0);
1533 }
Qu Wenruo3cae2102013-07-16 11:19:18 +08001534 btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1535 btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
Alexander Block8ea05e32012-07-25 17:35:53 +02001536 btrfs_set_root_otransid(new_root_item, trans->transid);
Alexander Block8ea05e32012-07-25 17:35:53 +02001537
Chris Mason925baed2008-06-25 16:01:30 -04001538 old = btrfs_lock_root_node(root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001539 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001540 if (ret) {
1541 btrfs_tree_unlock(old);
1542 free_extent_buffer(old);
Jeff Mahoney66642832016-06-10 18:19:25 -04001543 btrfs_abort_transaction(trans, ret);
Miao Xie8732d442012-09-17 23:52:38 -06001544 goto fail;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001545 }
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001546
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001547 btrfs_set_lock_blocking(old);
Chris Mason3063d292008-01-08 15:46:30 -05001548
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001549 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001550 /* clean up in any case */
Chris Mason925baed2008-06-25 16:01:30 -04001551 btrfs_tree_unlock(old);
1552 free_extent_buffer(old);
Miao Xie8732d442012-09-17 23:52:38 -06001553 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04001554 btrfs_abort_transaction(trans, ret);
Miao Xie8732d442012-09-17 23:52:38 -06001555 goto fail;
1556 }
Liu Bof1ebcc72011-11-14 20:48:06 -05001557 /* see comments in should_cow_block() */
Miao Xie27cdeb72014-04-02 19:51:05 +08001558 set_bit(BTRFS_ROOT_FORCE_COW, &root->state);
Liu Bof1ebcc72011-11-14 20:48:06 -05001559 smp_wmb();
1560
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001561 btrfs_set_root_node(new_root_item, tmp);
Yan, Zhenga22285a2010-05-16 10:48:46 -04001562 /* record when the snapshot was created in key.offset */
1563 key.offset = trans->transid;
1564 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
Chris Mason925baed2008-06-25 16:01:30 -04001565 btrfs_tree_unlock(tmp);
Chris Mason3063d292008-01-08 15:46:30 -05001566 free_extent_buffer(tmp);
Miao Xie8732d442012-09-17 23:52:38 -06001567 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04001568 btrfs_abort_transaction(trans, ret);
Miao Xie8732d442012-09-17 23:52:38 -06001569 goto fail;
1570 }
Chris Mason0660b5a2008-11-17 20:37:39 -05001571
Yan, Zhenga22285a2010-05-16 10:48:46 -04001572 /*
1573 * insert root back/forward references
1574 */
1575 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1576 parent_root->root_key.objectid,
Li Zefan33345d012011-04-20 10:31:50 +08001577 btrfs_ino(parent_inode), index,
Yan, Zhenga22285a2010-05-16 10:48:46 -04001578 dentry->d_name.name, dentry->d_name.len);
Miao Xie8732d442012-09-17 23:52:38 -06001579 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04001580 btrfs_abort_transaction(trans, ret);
Miao Xie8732d442012-09-17 23:52:38 -06001581 goto fail;
1582 }
Yan, Zhenga22285a2010-05-16 10:48:46 -04001583
1584 key.offset = (u64)-1;
1585 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001586 if (IS_ERR(pending->snap)) {
1587 ret = PTR_ERR(pending->snap);
Jeff Mahoney66642832016-06-10 18:19:25 -04001588 btrfs_abort_transaction(trans, ret);
Miao Xie8732d442012-09-17 23:52:38 -06001589 goto fail;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001590 }
Yan, Zhengd68fc572010-05-16 10:49:58 -04001591
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001592 ret = btrfs_reloc_post_snapshot(trans, pending);
Miao Xie8732d442012-09-17 23:52:38 -06001593 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04001594 btrfs_abort_transaction(trans, ret);
Miao Xie8732d442012-09-17 23:52:38 -06001595 goto fail;
1596 }
Miao Xie361048f2012-09-06 04:00:57 -06001597
1598 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
Miao Xie8732d442012-09-17 23:52:38 -06001599 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04001600 btrfs_abort_transaction(trans, ret);
Miao Xie8732d442012-09-17 23:52:38 -06001601 goto fail;
1602 }
Miao Xie42874b32012-09-06 04:03:32 -06001603
Qu Wenruo6426c7a2016-05-11 12:53:52 -07001604 /*
1605 * Do special qgroup accounting for snapshot, as we do some qgroup
1606 * snapshot hack to do fast snapshot.
1607 * To co-operate with that hack, we do hack again.
1608 * Or snapshot will be greatly slowed down by a subtree qgroup rescan
1609 */
1610 ret = qgroup_account_snapshot(trans, root, parent_root,
1611 pending->inherit, objectid);
1612 if (ret < 0)
1613 goto fail;
1614
Miao Xie42874b32012-09-06 04:03:32 -06001615 ret = btrfs_insert_dir_item(trans, parent_root,
1616 dentry->d_name.name, dentry->d_name.len,
1617 parent_inode, &key,
1618 BTRFS_FT_DIR, index);
1619 /* We have check then name at the beginning, so it is impossible. */
Chris Mason9c520572012-12-17 14:26:57 -05001620 BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
Miao Xie8732d442012-09-17 23:52:38 -06001621 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04001622 btrfs_abort_transaction(trans, ret);
Miao Xie8732d442012-09-17 23:52:38 -06001623 goto fail;
1624 }
Miao Xie42874b32012-09-06 04:03:32 -06001625
1626 btrfs_i_size_write(parent_inode, parent_inode->i_size +
1627 dentry->d_name.len * 2);
Deepa Dinamani04b285f2016-02-06 23:57:21 -08001628 parent_inode->i_mtime = parent_inode->i_ctime =
Deepa Dinamanic2050a42016-09-14 07:48:06 -07001629 current_time(parent_inode);
Josef Bacikbe6aef62012-10-22 15:43:12 -04001630 ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode);
Stefan Behrensdd5f9612013-08-15 17:11:20 +02001631 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04001632 btrfs_abort_transaction(trans, ret);
Stefan Behrensdd5f9612013-08-15 17:11:20 +02001633 goto fail;
1634 }
1635 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, new_uuid.b,
1636 BTRFS_UUID_KEY_SUBVOL, objectid);
1637 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04001638 btrfs_abort_transaction(trans, ret);
Stefan Behrensdd5f9612013-08-15 17:11:20 +02001639 goto fail;
1640 }
1641 if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1642 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
1643 new_root_item->received_uuid,
1644 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1645 objectid);
1646 if (ret && ret != -EEXIST) {
Jeff Mahoney66642832016-06-10 18:19:25 -04001647 btrfs_abort_transaction(trans, ret);
Stefan Behrensdd5f9612013-08-15 17:11:20 +02001648 goto fail;
1649 }
1650 }
Qu Wenruod6726332015-04-20 10:09:06 +08001651
1652 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1653 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04001654 btrfs_abort_transaction(trans, ret);
Qu Wenruod6726332015-04-20 10:09:06 +08001655 goto fail;
1656 }
1657
Chris Mason3063d292008-01-08 15:46:30 -05001658fail:
Miao Xieaec80302013-03-04 09:44:29 +00001659 pending->error = ret;
1660dir_item_existed:
Liu Bo98c99422011-09-11 10:52:24 -04001661 trans->block_rsv = rsv;
Liu Bo2382c5c2013-02-22 04:33:36 +00001662 trans->bytes_reserved = 0;
Qu Wenruod6726332015-04-20 10:09:06 +08001663clear_skip_qgroup:
1664 btrfs_clear_skip_qgroup(trans);
Miao Xie6fa97002012-09-06 04:00:32 -06001665no_free_objectid:
1666 kfree(new_root_item);
David Sterbab0c0ea62015-11-10 18:54:00 +01001667 pending->root_item = NULL;
Miao Xie42874b32012-09-06 04:03:32 -06001668 btrfs_free_path(path);
David Sterba8546b572015-11-10 18:54:03 +01001669 pending->path = NULL;
1670
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001671 return ret;
Chris Mason3063d292008-01-08 15:46:30 -05001672}
1673
Chris Masond352ac62008-09-29 15:18:18 -04001674/*
1675 * create all the snapshots we've scheduled for creation
1676 */
Chris Mason80b67942008-02-01 16:35:04 -05001677static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1678 struct btrfs_fs_info *fs_info)
Chris Mason3063d292008-01-08 15:46:30 -05001679{
Miao Xieaec80302013-03-04 09:44:29 +00001680 struct btrfs_pending_snapshot *pending, *next;
Chris Mason3063d292008-01-08 15:46:30 -05001681 struct list_head *head = &trans->transaction->pending_snapshots;
Miao Xieaec80302013-03-04 09:44:29 +00001682 int ret = 0;
Chris Mason3de45862008-11-17 21:02:50 -05001683
Miao Xieaec80302013-03-04 09:44:29 +00001684 list_for_each_entry_safe(pending, next, head, list) {
1685 list_del(&pending->list);
1686 ret = create_pending_snapshot(trans, fs_info, pending);
1687 if (ret)
1688 break;
1689 }
1690 return ret;
Chris Mason3de45862008-11-17 21:02:50 -05001691}
1692
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001693static void update_super_roots(struct btrfs_root *root)
1694{
1695 struct btrfs_root_item *root_item;
1696 struct btrfs_super_block *super;
1697
David Sterba6c417612011-04-13 15:41:04 +02001698 super = root->fs_info->super_copy;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001699
1700 root_item = &root->fs_info->chunk_root->root_item;
1701 super->chunk_root = root_item->bytenr;
1702 super->chunk_root_generation = root_item->generation;
1703 super->chunk_root_level = root_item->level;
1704
1705 root_item = &root->fs_info->tree_root->root_item;
1706 super->root = root_item->bytenr;
1707 super->generation = root_item->generation;
1708 super->root_level = root_item->level;
Jeff Mahoney3cdde222016-06-09 21:38:35 -04001709 if (btrfs_test_opt(root->fs_info, SPACE_CACHE))
Josef Bacik0af3d002010-06-21 14:48:16 -04001710 super->cache_generation = root_item->generation;
Josef Bacikafcdd122016-09-02 15:40:02 -04001711 if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &root->fs_info->flags))
Stefan Behrens70f80172013-08-15 17:11:23 +02001712 super->uuid_tree_generation = root_item->generation;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001713}
1714
Chris Masonf36f3042009-07-30 10:04:48 -04001715int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1716{
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001717 struct btrfs_transaction *trans;
Chris Masonf36f3042009-07-30 10:04:48 -04001718 int ret = 0;
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001719
Josef Bacika4abeea2011-04-11 17:25:13 -04001720 spin_lock(&info->trans_lock);
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001721 trans = info->running_transaction;
1722 if (trans)
1723 ret = (trans->state >= TRANS_STATE_COMMIT_START);
Josef Bacika4abeea2011-04-11 17:25:13 -04001724 spin_unlock(&info->trans_lock);
Chris Masonf36f3042009-07-30 10:04:48 -04001725 return ret;
1726}
1727
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001728int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1729{
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001730 struct btrfs_transaction *trans;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001731 int ret = 0;
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001732
Josef Bacika4abeea2011-04-11 17:25:13 -04001733 spin_lock(&info->trans_lock);
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001734 trans = info->running_transaction;
1735 if (trans)
1736 ret = is_transaction_blocked(trans);
Josef Bacika4abeea2011-04-11 17:25:13 -04001737 spin_unlock(&info->trans_lock);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001738 return ret;
1739}
1740
Sage Weilbb9c12c2010-10-29 15:37:34 -04001741/*
1742 * wait for the current transaction commit to start and block subsequent
1743 * transaction joins
1744 */
1745static void wait_current_trans_commit_start(struct btrfs_root *root,
1746 struct btrfs_transaction *trans)
1747{
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001748 wait_event(root->fs_info->transaction_blocked_wait,
Josef Bacik501407a2013-06-10 16:47:23 -04001749 trans->state >= TRANS_STATE_COMMIT_START ||
1750 trans->aborted);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001751}
1752
1753/*
1754 * wait for the current transaction to start and then become unblocked.
1755 * caller holds ref.
1756 */
1757static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1758 struct btrfs_transaction *trans)
1759{
Li Zefan72d63ed2011-07-14 03:17:00 +00001760 wait_event(root->fs_info->transaction_wait,
Josef Bacik501407a2013-06-10 16:47:23 -04001761 trans->state >= TRANS_STATE_UNBLOCKED ||
1762 trans->aborted);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001763}
1764
1765/*
1766 * commit transactions asynchronously. once btrfs_commit_transaction_async
1767 * returns, any subsequent transaction will not be allowed to join.
1768 */
1769struct btrfs_async_commit {
1770 struct btrfs_trans_handle *newtrans;
1771 struct btrfs_root *root;
Miao Xie7892b5a2012-11-15 08:14:47 +00001772 struct work_struct work;
Sage Weilbb9c12c2010-10-29 15:37:34 -04001773};
1774
1775static void do_async_commit(struct work_struct *work)
1776{
1777 struct btrfs_async_commit *ac =
Miao Xie7892b5a2012-11-15 08:14:47 +00001778 container_of(work, struct btrfs_async_commit, work);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001779
Sage Weil6fc4e35482012-08-30 16:26:15 -06001780 /*
1781 * We've got freeze protection passed with the transaction.
1782 * Tell lockdep about it.
1783 */
Liu Bob1a06a42013-11-06 16:57:55 +08001784 if (ac->newtrans->type & __TRANS_FREEZABLE)
Oleg Nesterovbee91822015-07-19 23:48:20 +02001785 __sb_writers_acquired(ac->root->fs_info->sb, SB_FREEZE_FS);
Sage Weil6fc4e35482012-08-30 16:26:15 -06001786
Sage Weile209db72012-08-30 16:26:16 -06001787 current->journal_info = ac->newtrans;
1788
Sage Weilbb9c12c2010-10-29 15:37:34 -04001789 btrfs_commit_transaction(ac->newtrans, ac->root);
1790 kfree(ac);
1791}
1792
1793int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1794 struct btrfs_root *root,
1795 int wait_for_unblock)
1796{
1797 struct btrfs_async_commit *ac;
1798 struct btrfs_transaction *cur_trans;
1799
1800 ac = kmalloc(sizeof(*ac), GFP_NOFS);
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00001801 if (!ac)
1802 return -ENOMEM;
Sage Weilbb9c12c2010-10-29 15:37:34 -04001803
Miao Xie7892b5a2012-11-15 08:14:47 +00001804 INIT_WORK(&ac->work, do_async_commit);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001805 ac->root = root;
Josef Bacik7a7eaa42011-04-13 12:54:33 -04001806 ac->newtrans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00001807 if (IS_ERR(ac->newtrans)) {
1808 int err = PTR_ERR(ac->newtrans);
1809 kfree(ac);
1810 return err;
1811 }
Sage Weilbb9c12c2010-10-29 15:37:34 -04001812
1813 /* take transaction reference */
Sage Weilbb9c12c2010-10-29 15:37:34 -04001814 cur_trans = trans->transaction;
Josef Bacik13c5a932011-04-11 15:45:29 -04001815 atomic_inc(&cur_trans->use_count);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001816
1817 btrfs_end_transaction(trans, root);
Sage Weil6fc4e35482012-08-30 16:26:15 -06001818
1819 /*
1820 * Tell lockdep we've released the freeze rwsem, since the
1821 * async commit thread will be the one to unlock it.
1822 */
Liu Bob1a06a42013-11-06 16:57:55 +08001823 if (ac->newtrans->type & __TRANS_FREEZABLE)
Oleg Nesterovbee91822015-07-19 23:48:20 +02001824 __sb_writers_release(root->fs_info->sb, SB_FREEZE_FS);
Sage Weil6fc4e35482012-08-30 16:26:15 -06001825
Miao Xie7892b5a2012-11-15 08:14:47 +00001826 schedule_work(&ac->work);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001827
1828 /* wait for transaction to start and unblock */
Sage Weilbb9c12c2010-10-29 15:37:34 -04001829 if (wait_for_unblock)
1830 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1831 else
1832 wait_current_trans_commit_start(root, cur_trans);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001833
Sage Weil38e88052011-06-10 18:43:13 +00001834 if (current->journal_info == trans)
1835 current->journal_info = NULL;
1836
Josef Bacik724e2312013-09-30 11:36:38 -04001837 btrfs_put_transaction(cur_trans);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001838 return 0;
1839}
1840
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001841
1842static void cleanup_transaction(struct btrfs_trans_handle *trans,
Josef Bacik7b8b92a2012-05-31 15:52:43 -04001843 struct btrfs_root *root, int err)
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001844{
1845 struct btrfs_transaction *cur_trans = trans->transaction;
Liu Bof094ac32013-02-27 13:28:25 +00001846 DEFINE_WAIT(wait);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001847
1848 WARN_ON(trans->use_count > 1);
1849
Jeff Mahoney66642832016-06-10 18:19:25 -04001850 btrfs_abort_transaction(trans, err);
Josef Bacik7b8b92a2012-05-31 15:52:43 -04001851
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001852 spin_lock(&root->fs_info->trans_lock);
Liu Bo66b61352013-03-04 16:25:41 +00001853
Miao Xie25d8c282013-05-15 07:48:26 +00001854 /*
1855 * If the transaction is removed from the list, it means this
1856 * transaction has been committed successfully, so it is impossible
1857 * to call the cleanup function.
1858 */
1859 BUG_ON(list_empty(&cur_trans->list));
Liu Bo66b61352013-03-04 16:25:41 +00001860
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001861 list_del_init(&cur_trans->list);
Josef Bacikd7096fc2012-05-31 15:49:57 -04001862 if (cur_trans == root->fs_info->running_transaction) {
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001863 cur_trans->state = TRANS_STATE_COMMIT_DOING;
Liu Bof094ac32013-02-27 13:28:25 +00001864 spin_unlock(&root->fs_info->trans_lock);
1865 wait_event(cur_trans->writer_wait,
1866 atomic_read(&cur_trans->num_writers) == 1);
1867
1868 spin_lock(&root->fs_info->trans_lock);
Josef Bacikd7096fc2012-05-31 15:49:57 -04001869 }
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001870 spin_unlock(&root->fs_info->trans_lock);
1871
1872 btrfs_cleanup_one_transaction(trans->transaction, root);
1873
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001874 spin_lock(&root->fs_info->trans_lock);
1875 if (cur_trans == root->fs_info->running_transaction)
1876 root->fs_info->running_transaction = NULL;
1877 spin_unlock(&root->fs_info->trans_lock);
1878
Josef Bacike0228282013-09-20 22:26:29 -04001879 if (trans->type & __TRANS_FREEZABLE)
1880 sb_end_intwrite(root->fs_info->sb);
Josef Bacik724e2312013-09-30 11:36:38 -04001881 btrfs_put_transaction(cur_trans);
1882 btrfs_put_transaction(cur_trans);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001883
1884 trace_btrfs_transaction_commit(root);
1885
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001886 if (current->journal_info == trans)
1887 current->journal_info = NULL;
Wang Shilongc0af8f02014-02-19 19:24:18 +08001888 btrfs_scrub_cancel(root->fs_info);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001889
1890 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1891}
1892
Miao Xie82436612013-05-15 07:48:28 +00001893static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
1894{
Jeff Mahoney3cdde222016-06-09 21:38:35 -04001895 if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
Miao Xie6c255e62014-03-06 13:55:01 +08001896 return btrfs_start_delalloc_roots(fs_info, 1, -1);
Miao Xie82436612013-05-15 07:48:28 +00001897 return 0;
1898}
1899
1900static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
1901{
Jeff Mahoney3cdde222016-06-09 21:38:35 -04001902 if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
Filipe Manana578def72016-04-26 15:36:38 +01001903 btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1);
Miao Xie82436612013-05-15 07:48:28 +00001904}
1905
Josef Bacik50d9aa92014-11-21 14:52:38 -05001906static inline void
Josef Bacik161c35492015-09-24 16:17:39 -04001907btrfs_wait_pending_ordered(struct btrfs_transaction *cur_trans)
Josef Bacik50d9aa92014-11-21 14:52:38 -05001908{
Josef Bacik161c35492015-09-24 16:17:39 -04001909 wait_event(cur_trans->pending_wait,
1910 atomic_read(&cur_trans->pending_ordered) == 0);
Josef Bacik50d9aa92014-11-21 14:52:38 -05001911}
1912
Chris Mason79154b12007-03-22 15:59:16 -04001913int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1914 struct btrfs_root *root)
1915{
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001916 struct btrfs_transaction *cur_trans = trans->transaction;
Chris Mason8fd17792007-04-19 21:01:03 -04001917 struct btrfs_transaction *prev_trans = NULL;
Miao Xie25287e02012-10-25 09:31:03 +00001918 int ret;
Chris Mason79154b12007-03-22 15:59:16 -04001919
Miao Xie8d25a082013-01-15 06:27:25 +00001920 /* Stop the commit early if ->aborted is set */
1921 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
Miao Xie25287e02012-10-25 09:31:03 +00001922 ret = cur_trans->aborted;
Josef Bacike4a2bca2013-02-06 16:55:41 -05001923 btrfs_end_transaction(trans, root);
1924 return ret;
Miao Xie25287e02012-10-25 09:31:03 +00001925 }
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001926
Chris Mason56bec292009-03-13 10:10:06 -04001927 /* make a pass through all the delayed refs we have so far
1928 * any runnings procs may add more while we are here
1929 */
1930 ret = btrfs_run_delayed_refs(trans, root, 0);
Josef Bacike4a2bca2013-02-06 16:55:41 -05001931 if (ret) {
1932 btrfs_end_transaction(trans, root);
1933 return ret;
1934 }
Chris Mason56bec292009-03-13 10:10:06 -04001935
Josef Bacik0e721102012-06-26 16:13:18 -04001936 btrfs_trans_release_metadata(trans, root);
1937 trans->block_rsv = NULL;
1938
Chris Masonb7ec40d2009-03-12 20:12:45 -04001939 cur_trans = trans->transaction;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001940
Chris Mason56bec292009-03-13 10:10:06 -04001941 /*
1942 * set the flushing flag so procs in this transaction have to
1943 * start sending their work down.
1944 */
Chris Masonb7ec40d2009-03-12 20:12:45 -04001945 cur_trans->delayed_refs.flushing = 1;
Josef Bacik1be41b72013-06-12 13:56:06 -04001946 smp_wmb();
Chris Mason56bec292009-03-13 10:10:06 -04001947
Josef Bacikea658ba2012-09-11 16:57:25 -04001948 if (!list_empty(&trans->new_bgs))
1949 btrfs_create_pending_block_groups(trans, root);
1950
Chris Masonc3e69d52009-03-13 10:17:05 -04001951 ret = btrfs_run_delayed_refs(trans, root, 0);
Josef Bacike4a2bca2013-02-06 16:55:41 -05001952 if (ret) {
1953 btrfs_end_transaction(trans, root);
1954 return ret;
1955 }
Chris Mason56bec292009-03-13 10:10:06 -04001956
Josef Bacik3204d332015-09-24 10:46:10 -04001957 if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) {
Chris Mason1bbc6212015-04-06 12:46:08 -07001958 int run_it = 0;
1959
1960 /* this mutex is also taken before trying to set
1961 * block groups readonly. We need to make sure
1962 * that nobody has set a block group readonly
1963 * after a extents from that block group have been
1964 * allocated for cache files. btrfs_set_block_group_ro
1965 * will wait for the transaction to commit if it
Josef Bacik3204d332015-09-24 10:46:10 -04001966 * finds BTRFS_TRANS_DIRTY_BG_RUN set.
Chris Mason1bbc6212015-04-06 12:46:08 -07001967 *
Josef Bacik3204d332015-09-24 10:46:10 -04001968 * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure
1969 * only one process starts all the block group IO. It wouldn't
Chris Mason1bbc6212015-04-06 12:46:08 -07001970 * hurt to have more than one go through, but there's no
1971 * real advantage to it either.
1972 */
1973 mutex_lock(&root->fs_info->ro_block_group_mutex);
Josef Bacik3204d332015-09-24 10:46:10 -04001974 if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN,
1975 &cur_trans->flags))
Chris Mason1bbc6212015-04-06 12:46:08 -07001976 run_it = 1;
Chris Mason1bbc6212015-04-06 12:46:08 -07001977 mutex_unlock(&root->fs_info->ro_block_group_mutex);
1978
1979 if (run_it)
1980 ret = btrfs_start_dirty_block_groups(trans, root);
1981 }
1982 if (ret) {
1983 btrfs_end_transaction(trans, root);
1984 return ret;
1985 }
1986
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001987 spin_lock(&root->fs_info->trans_lock);
1988 if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
1989 spin_unlock(&root->fs_info->trans_lock);
Josef Bacik13c5a932011-04-11 15:45:29 -04001990 atomic_inc(&cur_trans->use_count);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001991 ret = btrfs_end_transaction(trans, root);
Chris Masonccd467d2007-06-28 15:57:36 -04001992
Li Zefanb9c83002011-07-14 03:17:14 +00001993 wait_for_commit(root, cur_trans);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001994
Liu Bob4924a02015-03-06 20:23:44 +08001995 if (unlikely(cur_trans->aborted))
1996 ret = cur_trans->aborted;
1997
Josef Bacik724e2312013-09-30 11:36:38 -04001998 btrfs_put_transaction(cur_trans);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001999
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002000 return ret;
Chris Mason79154b12007-03-22 15:59:16 -04002001 }
Chris Mason4313b392008-01-03 09:08:48 -05002002
Miao Xie4a9d8bd2013-05-17 03:53:43 +00002003 cur_trans->state = TRANS_STATE_COMMIT_START;
Sage Weilbb9c12c2010-10-29 15:37:34 -04002004 wake_up(&root->fs_info->transaction_blocked_wait);
2005
Chris Masonccd467d2007-06-28 15:57:36 -04002006 if (cur_trans->list.prev != &root->fs_info->trans_list) {
2007 prev_trans = list_entry(cur_trans->list.prev,
2008 struct btrfs_transaction, list);
Miao Xie4a9d8bd2013-05-17 03:53:43 +00002009 if (prev_trans->state != TRANS_STATE_COMPLETED) {
Josef Bacik13c5a932011-04-11 15:45:29 -04002010 atomic_inc(&prev_trans->use_count);
Josef Bacika4abeea2011-04-11 17:25:13 -04002011 spin_unlock(&root->fs_info->trans_lock);
Chris Masonccd467d2007-06-28 15:57:36 -04002012
2013 wait_for_commit(root, prev_trans);
Filipe Manana1f9b8c82015-08-12 11:54:35 +01002014 ret = prev_trans->aborted;
Chris Masonccd467d2007-06-28 15:57:36 -04002015
Josef Bacik724e2312013-09-30 11:36:38 -04002016 btrfs_put_transaction(prev_trans);
Filipe Manana1f9b8c82015-08-12 11:54:35 +01002017 if (ret)
2018 goto cleanup_transaction;
Josef Bacika4abeea2011-04-11 17:25:13 -04002019 } else {
2020 spin_unlock(&root->fs_info->trans_lock);
Chris Masonccd467d2007-06-28 15:57:36 -04002021 }
Josef Bacika4abeea2011-04-11 17:25:13 -04002022 } else {
2023 spin_unlock(&root->fs_info->trans_lock);
Chris Masonccd467d2007-06-28 15:57:36 -04002024 }
Josef Bacik15ee9bc2007-08-10 16:22:09 -04002025
Miao Xie0860adf2013-05-15 07:48:27 +00002026 extwriter_counter_dec(cur_trans, trans->type);
2027
Miao Xie82436612013-05-15 07:48:28 +00002028 ret = btrfs_start_delalloc_flush(root->fs_info);
2029 if (ret)
2030 goto cleanup_transaction;
2031
Chris Mason8d875f92014-08-12 10:47:42 -07002032 ret = btrfs_run_delayed_items(trans, root);
Miao Xie581227d2013-05-15 07:48:30 +00002033 if (ret)
2034 goto cleanup_transaction;
Chris Mason89573b92009-03-12 20:12:45 -04002035
Miao Xie581227d2013-05-15 07:48:30 +00002036 wait_event(cur_trans->writer_wait,
2037 extwriter_counter_read(cur_trans) == 0);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04002038
Miao Xie581227d2013-05-15 07:48:30 +00002039 /* some pending stuffs might be added after the previous flush. */
Chris Mason8d875f92014-08-12 10:47:42 -07002040 ret = btrfs_run_delayed_items(trans, root);
Miao Xieca469632012-11-01 07:33:14 +00002041 if (ret)
2042 goto cleanup_transaction;
2043
Miao Xie82436612013-05-15 07:48:28 +00002044 btrfs_wait_delalloc_flush(root->fs_info);
Wang Shilongcb7ab022013-12-04 21:16:53 +08002045
Josef Bacik161c35492015-09-24 16:17:39 -04002046 btrfs_wait_pending_ordered(cur_trans);
Josef Bacik50d9aa92014-11-21 14:52:38 -05002047
Wang Shilongcb7ab022013-12-04 21:16:53 +08002048 btrfs_scrub_pause(root);
Chris Mason75857172011-06-13 20:00:16 -04002049 /*
Josef Baciked0ca142011-06-14 16:22:15 -04002050 * Ok now we need to make sure to block out any other joins while we
2051 * commit the transaction. We could have started a join before setting
Miao Xie4a9d8bd2013-05-17 03:53:43 +00002052 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
Josef Baciked0ca142011-06-14 16:22:15 -04002053 */
2054 spin_lock(&root->fs_info->trans_lock);
Miao Xie4a9d8bd2013-05-17 03:53:43 +00002055 cur_trans->state = TRANS_STATE_COMMIT_DOING;
Josef Baciked0ca142011-06-14 16:22:15 -04002056 spin_unlock(&root->fs_info->trans_lock);
2057 wait_event(cur_trans->writer_wait,
2058 atomic_read(&cur_trans->num_writers) == 1);
2059
Miao Xie2cba30f2013-01-15 06:29:12 +00002060 /* ->aborted might be set after the previous check, so check it */
2061 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
2062 ret = cur_trans->aborted;
Wang Shilong6cf7f772014-02-19 19:24:16 +08002063 goto scrub_continue;
Miao Xie2cba30f2013-01-15 06:29:12 +00002064 }
Chris Masone038dca2011-06-17 14:16:13 -04002065 /*
Chris Mason75857172011-06-13 20:00:16 -04002066 * the reloc mutex makes sure that we stop
2067 * the balancing code from coming in and moving
2068 * extents around in the middle of the commit
2069 */
2070 mutex_lock(&root->fs_info->reloc_mutex);
2071
Miao Xie42874b32012-09-06 04:03:32 -06002072 /*
2073 * We needn't worry about the delayed items because we will
2074 * deal with them in create_pending_snapshot(), which is the
2075 * core function of the snapshot creation.
2076 */
2077 ret = create_pending_snapshots(trans, root->fs_info);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002078 if (ret) {
2079 mutex_unlock(&root->fs_info->reloc_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08002080 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002081 }
Chris Mason3063d292008-01-08 15:46:30 -05002082
Miao Xie42874b32012-09-06 04:03:32 -06002083 /*
2084 * We insert the dir indexes of the snapshots and update the inode
2085 * of the snapshots' parents after the snapshot creation, so there
2086 * are some delayed items which are not dealt with. Now deal with
2087 * them.
2088 *
2089 * We needn't worry that this operation will corrupt the snapshots,
2090 * because all the tree which are snapshoted will be forced to COW
2091 * the nodes and leaves.
2092 */
2093 ret = btrfs_run_delayed_items(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002094 if (ret) {
2095 mutex_unlock(&root->fs_info->reloc_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08002096 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002097 }
Miao Xie16cdcec2011-04-22 18:12:22 +08002098
Chris Mason56bec292009-03-13 10:10:06 -04002099 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002100 if (ret) {
2101 mutex_unlock(&root->fs_info->reloc_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08002102 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002103 }
Chris Mason56bec292009-03-13 10:10:06 -04002104
Qu Wenruo0ed47922015-04-16 16:55:08 +08002105 /* Reocrd old roots for later qgroup accounting */
2106 ret = btrfs_qgroup_prepare_account_extents(trans, root->fs_info);
2107 if (ret) {
2108 mutex_unlock(&root->fs_info->reloc_mutex);
2109 goto scrub_continue;
2110 }
2111
Chris Masone9993762011-06-17 16:14:09 -04002112 /*
2113 * make sure none of the code above managed to slip in a
2114 * delayed item
2115 */
2116 btrfs_assert_delayed_root_empty(root);
2117
Chris Mason2c90e5d2007-04-02 10:50:19 -04002118 WARN_ON(cur_trans != trans->transaction);
Chris Masondc17ff82008-01-08 15:46:30 -05002119
Chris Masone02119d2008-09-05 16:13:11 -04002120 /* btrfs_commit_tree_roots is responsible for getting the
2121 * various roots consistent with each other. Every pointer
2122 * in the tree of tree roots has to point to the most up to date
2123 * root for every subvolume and other tree. So, we have to keep
2124 * the tree logging code from jumping in and changing any
2125 * of the trees.
2126 *
2127 * At this point in the commit, there can't be any tree-log
2128 * writers, but a little lower down we drop the trans mutex
2129 * and let new people in. By holding the tree_log_mutex
2130 * from now until after the super is written, we avoid races
2131 * with the tree-log code.
2132 */
2133 mutex_lock(&root->fs_info->tree_log_mutex);
Zheng Yan1a40e232008-09-26 10:09:34 -04002134
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002135 ret = commit_fs_roots(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002136 if (ret) {
2137 mutex_unlock(&root->fs_info->tree_log_mutex);
David Sterba871383b2012-04-02 18:31:37 +02002138 mutex_unlock(&root->fs_info->reloc_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08002139 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002140 }
Chris Mason54aa1f42007-06-22 14:16:25 -04002141
Qu Wenruo3818aea2014-01-13 13:36:06 +08002142 /*
David Sterba7e1876a2014-02-05 15:26:17 +01002143 * Since the transaction is done, we can apply the pending changes
2144 * before the next transaction.
Qu Wenruo3818aea2014-01-13 13:36:06 +08002145 */
David Sterba572d9ab2014-02-05 15:26:17 +01002146 btrfs_apply_pending_changes(root->fs_info);
Qu Wenruo3818aea2014-01-13 13:36:06 +08002147
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002148 /* commit_fs_roots gets rid of all the tree log roots, it is now
Chris Masone02119d2008-09-05 16:13:11 -04002149 * safe to free the root of tree log roots
2150 */
2151 btrfs_free_log_root_tree(trans, root->fs_info);
2152
Qu Wenruo0ed47922015-04-16 16:55:08 +08002153 /*
2154 * Since fs roots are all committed, we can get a quite accurate
2155 * new_roots. So let's do quota accounting.
2156 */
2157 ret = btrfs_qgroup_account_extents(trans, root->fs_info);
2158 if (ret < 0) {
2159 mutex_unlock(&root->fs_info->tree_log_mutex);
2160 mutex_unlock(&root->fs_info->reloc_mutex);
2161 goto scrub_continue;
2162 }
2163
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002164 ret = commit_cowonly_roots(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002165 if (ret) {
2166 mutex_unlock(&root->fs_info->tree_log_mutex);
David Sterba871383b2012-04-02 18:31:37 +02002167 mutex_unlock(&root->fs_info->reloc_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08002168 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002169 }
Chris Mason54aa1f42007-06-22 14:16:25 -04002170
Miao Xie2cba30f2013-01-15 06:29:12 +00002171 /*
2172 * The tasks which save the space cache and inode cache may also
2173 * update ->aborted, check it.
2174 */
2175 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
2176 ret = cur_trans->aborted;
2177 mutex_unlock(&root->fs_info->tree_log_mutex);
2178 mutex_unlock(&root->fs_info->reloc_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08002179 goto scrub_continue;
Miao Xie2cba30f2013-01-15 06:29:12 +00002180 }
2181
Yan Zheng11833d62009-09-11 16:11:19 -04002182 btrfs_prepare_extent_commit(trans, root);
2183
Chris Mason78fae272007-03-25 11:35:08 -04002184 cur_trans = root->fs_info->running_transaction;
Chris Mason5f39d392007-10-15 16:14:19 -04002185
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002186 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
2187 root->fs_info->tree_root->node);
Josef Bacik9e351cc2014-03-13 15:42:13 -04002188 list_add_tail(&root->fs_info->tree_root->dirty_list,
2189 &cur_trans->switch_commits);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002190
2191 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
2192 root->fs_info->chunk_root->node);
Josef Bacik9e351cc2014-03-13 15:42:13 -04002193 list_add_tail(&root->fs_info->chunk_root->dirty_list,
2194 &cur_trans->switch_commits);
2195
2196 switch_commit_roots(cur_trans, root->fs_info);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002197
Jan Schmidtedf39272012-06-28 18:04:55 +02002198 assert_qgroups_uptodate(trans);
Josef Bacikce93ec52014-11-17 15:45:48 -05002199 ASSERT(list_empty(&cur_trans->dirty_bgs));
Chris Mason1bbc6212015-04-06 12:46:08 -07002200 ASSERT(list_empty(&cur_trans->io_bgs));
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002201 update_super_roots(root);
Chris Masone02119d2008-09-05 16:13:11 -04002202
Josef Bacik60e7cd32013-09-30 14:10:43 -04002203 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
2204 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
David Sterba6c417612011-04-13 15:41:04 +02002205 memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
2206 sizeof(*root->fs_info->super_copy));
Chris Masonccd467d2007-06-28 15:57:36 -04002207
Miao Xie935e5cc2014-09-03 21:35:33 +08002208 btrfs_update_commit_device_size(root->fs_info);
Miao Xiece7213c2014-09-03 21:35:34 +08002209 btrfs_update_commit_device_bytes_used(root, cur_trans);
Miao Xie935e5cc2014-09-03 21:35:33 +08002210
Josef Bacikafcdd122016-09-02 15:40:02 -04002211 clear_bit(BTRFS_FS_LOG1_ERR, &root->fs_info->flags);
2212 clear_bit(BTRFS_FS_LOG2_ERR, &root->fs_info->flags);
Filipe Manana656f30d2014-09-26 12:25:56 +01002213
Filipe Manana4fbcdf62015-05-20 14:01:54 +01002214 btrfs_trans_release_chunk_metadata(trans);
2215
Josef Bacika4abeea2011-04-11 17:25:13 -04002216 spin_lock(&root->fs_info->trans_lock);
Miao Xie4a9d8bd2013-05-17 03:53:43 +00002217 cur_trans->state = TRANS_STATE_UNBLOCKED;
Josef Bacika4abeea2011-04-11 17:25:13 -04002218 root->fs_info->running_transaction = NULL;
Josef Bacika4abeea2011-04-11 17:25:13 -04002219 spin_unlock(&root->fs_info->trans_lock);
Chris Mason75857172011-06-13 20:00:16 -04002220 mutex_unlock(&root->fs_info->reloc_mutex);
Chris Masonb7ec40d2009-03-12 20:12:45 -04002221
Chris Masonf9295742008-07-17 12:54:14 -04002222 wake_up(&root->fs_info->transaction_wait);
Chris Masone6dcd2d2008-07-17 12:53:50 -04002223
Chris Mason79154b12007-03-22 15:59:16 -04002224 ret = btrfs_write_and_wait_transaction(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002225 if (ret) {
Anand Jain34d97002016-03-16 16:43:06 +08002226 btrfs_handle_fs_error(root->fs_info, ret,
David Sterba08748812013-03-12 14:46:08 +00002227 "Error while writing out transaction");
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002228 mutex_unlock(&root->fs_info->tree_log_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08002229 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002230 }
2231
2232 ret = write_ctree_super(trans, root, 0);
2233 if (ret) {
2234 mutex_unlock(&root->fs_info->tree_log_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08002235 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002236 }
Chris Mason4313b392008-01-03 09:08:48 -05002237
Chris Masone02119d2008-09-05 16:13:11 -04002238 /*
2239 * the super is written, we can safely allow the tree-loggers
2240 * to go about their business
2241 */
2242 mutex_unlock(&root->fs_info->tree_log_mutex);
2243
Yan Zheng11833d62009-09-11 16:11:19 -04002244 btrfs_finish_extent_commit(trans, root);
Chris Mason4313b392008-01-03 09:08:48 -05002245
Josef Bacik3204d332015-09-24 10:46:10 -04002246 if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags))
Zhao Lei13212b52015-02-12 14:18:17 +08002247 btrfs_clear_space_info_full(root->fs_info);
2248
Josef Bacik15ee9bc2007-08-10 16:22:09 -04002249 root->fs_info->last_trans_committed = cur_trans->transid;
Miao Xie4a9d8bd2013-05-17 03:53:43 +00002250 /*
2251 * We needn't acquire the lock here because there is no other task
2252 * which can change it.
2253 */
2254 cur_trans->state = TRANS_STATE_COMPLETED;
Chris Mason2c90e5d2007-04-02 10:50:19 -04002255 wake_up(&cur_trans->commit_wait);
Chris Mason3de45862008-11-17 21:02:50 -05002256
Josef Bacika4abeea2011-04-11 17:25:13 -04002257 spin_lock(&root->fs_info->trans_lock);
Josef Bacik13c5a932011-04-11 15:45:29 -04002258 list_del_init(&cur_trans->list);
Josef Bacika4abeea2011-04-11 17:25:13 -04002259 spin_unlock(&root->fs_info->trans_lock);
2260
Josef Bacik724e2312013-09-30 11:36:38 -04002261 btrfs_put_transaction(cur_trans);
2262 btrfs_put_transaction(cur_trans);
Josef Bacik58176a92007-08-29 15:47:34 -04002263
Miao Xie0860adf2013-05-15 07:48:27 +00002264 if (trans->type & __TRANS_FREEZABLE)
Miao Xie354aa0f2012-09-20 01:54:00 -06002265 sb_end_intwrite(root->fs_info->sb);
Jan Karab2b5ef52012-06-12 16:20:45 +02002266
liubo1abe9b82011-03-24 11:18:59 +00002267 trace_btrfs_transaction_commit(root);
2268
Arne Jansena2de7332011-03-08 14:14:00 +01002269 btrfs_scrub_continue(root);
2270
Josef Bacik9ed74f22009-09-11 16:12:44 -04002271 if (current->journal_info == trans)
2272 current->journal_info = NULL;
2273
Chris Mason2c90e5d2007-04-02 10:50:19 -04002274 kmem_cache_free(btrfs_trans_handle_cachep, trans);
Yan, Zheng24bbcf02009-11-12 09:36:34 +00002275
Wang Xiaoguang9e7cc912016-08-01 13:28:08 +08002276 /*
2277 * If fs has been frozen, we can not handle delayed iputs, otherwise
2278 * it'll result in deadlock about SB_FREEZE_FS.
2279 */
Zhao Lei8a733012015-07-15 11:48:14 +08002280 if (current != root->fs_info->transaction_kthread &&
Wang Xiaoguang9e7cc912016-08-01 13:28:08 +08002281 current != root->fs_info->cleaner_kthread &&
2282 !root->fs_info->fs_frozen)
Yan, Zheng24bbcf02009-11-12 09:36:34 +00002283 btrfs_run_delayed_iputs(root);
2284
Chris Mason79154b12007-03-22 15:59:16 -04002285 return ret;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002286
Wang Shilong6cf7f772014-02-19 19:24:16 +08002287scrub_continue:
2288 btrfs_scrub_continue(root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002289cleanup_transaction:
Josef Bacik0e721102012-06-26 16:13:18 -04002290 btrfs_trans_release_metadata(trans, root);
Filipe Manana4fbcdf62015-05-20 14:01:54 +01002291 btrfs_trans_release_chunk_metadata(trans);
Josef Bacik0e721102012-06-26 16:13:18 -04002292 trans->block_rsv = NULL;
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00002293 btrfs_warn(root->fs_info, "Skipping commit of aborted transaction.");
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002294 if (current->journal_info == trans)
2295 current->journal_info = NULL;
Josef Bacik7b8b92a2012-05-31 15:52:43 -04002296 cleanup_transaction(trans, root, ret);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002297
2298 return ret;
Chris Mason79154b12007-03-22 15:59:16 -04002299}
2300
Chris Masond352ac62008-09-29 15:18:18 -04002301/*
David Sterba9d1a2a32013-03-12 15:13:28 +00002302 * return < 0 if error
2303 * 0 if there are no more dead_roots at the time of call
2304 * 1 there are more to be processed, call me again
2305 *
2306 * The return value indicates there are certainly more snapshots to delete, but
2307 * if there comes a new one during processing, it may return 0. We don't mind,
2308 * because btrfs_commit_super will poke cleaner thread and it will process it a
2309 * few seconds later.
Chris Masond352ac62008-09-29 15:18:18 -04002310 */
David Sterba9d1a2a32013-03-12 15:13:28 +00002311int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root)
Chris Masone9d0b132007-08-10 14:06:19 -04002312{
David Sterba9d1a2a32013-03-12 15:13:28 +00002313 int ret;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002314 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone9d0b132007-08-10 14:06:19 -04002315
Josef Bacika4abeea2011-04-11 17:25:13 -04002316 spin_lock(&fs_info->trans_lock);
David Sterba9d1a2a32013-03-12 15:13:28 +00002317 if (list_empty(&fs_info->dead_roots)) {
2318 spin_unlock(&fs_info->trans_lock);
2319 return 0;
2320 }
2321 root = list_first_entry(&fs_info->dead_roots,
2322 struct btrfs_root, root_list);
Josef Bacikcfad3922013-07-25 15:11:47 -04002323 list_del_init(&root->root_list);
Josef Bacika4abeea2011-04-11 17:25:13 -04002324 spin_unlock(&fs_info->trans_lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002325
Jeff Mahoneyab8d0fc2016-09-20 10:05:02 -04002326 btrfs_debug(fs_info, "cleaner removing %llu", root->objectid);
Jeff Mahoney2c536792011-10-03 23:22:41 -04002327
David Sterba9d1a2a32013-03-12 15:13:28 +00002328 btrfs_kill_all_delayed_nodes(root);
Yan, Zheng76dda932009-09-21 16:00:26 -04002329
David Sterba9d1a2a32013-03-12 15:13:28 +00002330 if (btrfs_header_backref_rev(root->node) <
2331 BTRFS_MIXED_BACKREF_REV)
2332 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
2333 else
2334 ret = btrfs_drop_snapshot(root, NULL, 1, 0);
David Sterba32471dc2014-02-05 02:03:47 +01002335
Josef Bacik6596a9282013-07-31 10:28:05 -04002336 return (ret < 0) ? 0 : 1;
Chris Masone9d0b132007-08-10 14:06:19 -04002337}
David Sterba572d9ab2014-02-05 15:26:17 +01002338
2339void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info)
2340{
2341 unsigned long prev;
2342 unsigned long bit;
2343
Qu Wenruo6c9fe142015-01-20 17:05:33 +08002344 prev = xchg(&fs_info->pending_changes, 0);
David Sterba572d9ab2014-02-05 15:26:17 +01002345 if (!prev)
2346 return;
2347
David Sterba7e1876a2014-02-05 15:26:17 +01002348 bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE;
2349 if (prev & bit)
2350 btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2351 prev &= ~bit;
2352
2353 bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE;
2354 if (prev & bit)
2355 btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2356 prev &= ~bit;
2357
David Sterbad51033d2014-11-12 14:24:35 +01002358 bit = 1 << BTRFS_PENDING_COMMIT;
2359 if (prev & bit)
2360 btrfs_debug(fs_info, "pending commit done");
2361 prev &= ~bit;
2362
David Sterba572d9ab2014-02-05 15:26:17 +01002363 if (prev)
2364 btrfs_warn(fs_info,
2365 "unknown pending changes left 0x%lx, ignoring", prev);
2366}