blob: f6e24cb423aed0c78316b983d3687d07bb2ae1ec [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)
68 printk(KERN_ERR "pending csums is %llu\n",
69 transaction->delayed_refs.pending_csums);
Josef Bacik6df9a952013-06-27 13:22:46 -040070 while (!list_empty(&transaction->pending_chunks)) {
71 struct extent_map *em;
72
73 em = list_first_entry(&transaction->pending_chunks,
74 struct extent_map, list);
75 list_del_init(&em->list);
76 free_extent_map(em);
77 }
Filipe Manana7785a662015-11-27 16:12:00 +000078 /*
79 * If any block groups are found in ->deleted_bgs then it's
80 * because the transaction was aborted and a commit did not
81 * happen (things failed before writing the new superblock
82 * and calling btrfs_finish_extent_commit()), so we can not
83 * discard the physical locations of the block groups.
84 */
85 while (!list_empty(&transaction->deleted_bgs)) {
86 struct btrfs_block_group_cache *cache;
87
88 cache = list_first_entry(&transaction->deleted_bgs,
89 struct btrfs_block_group_cache,
90 bg_list);
91 list_del_init(&cache->bg_list);
92 btrfs_put_block_group_trimming(cache);
93 btrfs_put_block_group(cache);
94 }
Chris Mason2c90e5d2007-04-02 10:50:19 -040095 kmem_cache_free(btrfs_transaction_cachep, transaction);
Chris Mason78fae272007-03-25 11:35:08 -040096 }
Chris Mason79154b12007-03-22 15:59:16 -040097}
98
Filipe Manana663dfbb2014-10-13 12:28:37 +010099static void clear_btree_io_tree(struct extent_io_tree *tree)
100{
101 spin_lock(&tree->lock);
David Sterbab666a9c2015-10-10 18:24:48 +0200102 /*
103 * Do a single barrier for the waitqueue_active check here, the state
104 * of the waitqueue should not change once clear_btree_io_tree is
105 * called.
106 */
107 smp_mb();
Filipe Manana663dfbb2014-10-13 12:28:37 +0100108 while (!RB_EMPTY_ROOT(&tree->state)) {
109 struct rb_node *node;
110 struct extent_state *state;
111
112 node = rb_first(&tree->state);
113 state = rb_entry(node, struct extent_state, rb_node);
114 rb_erase(&state->rb_node, &tree->state);
115 RB_CLEAR_NODE(&state->rb_node);
116 /*
117 * btree io trees aren't supposed to have tasks waiting for
118 * changes in the flags of extent states ever.
119 */
120 ASSERT(!waitqueue_active(&state->wq));
121 free_extent_state(state);
David Sterba351810c2015-01-08 15:20:54 +0100122
123 cond_resched_lock(&tree->lock);
Filipe Manana663dfbb2014-10-13 12:28:37 +0100124 }
125 spin_unlock(&tree->lock);
126}
127
Josef Bacik9e351cc2014-03-13 15:42:13 -0400128static noinline void switch_commit_roots(struct btrfs_transaction *trans,
129 struct btrfs_fs_info *fs_info)
Josef Bacik817d52f2009-07-13 21:29:25 -0400130{
Josef Bacik9e351cc2014-03-13 15:42:13 -0400131 struct btrfs_root *root, *tmp;
132
133 down_write(&fs_info->commit_root_sem);
134 list_for_each_entry_safe(root, tmp, &trans->switch_commits,
135 dirty_list) {
136 list_del_init(&root->dirty_list);
137 free_extent_buffer(root->commit_root);
138 root->commit_root = btrfs_root_node(root);
139 if (is_fstree(root->objectid))
140 btrfs_unpin_free_ino(root);
Filipe Manana663dfbb2014-10-13 12:28:37 +0100141 clear_btree_io_tree(&root->dirty_log_pages);
Josef Bacik9e351cc2014-03-13 15:42:13 -0400142 }
Josef Bacik2b9dbef2015-09-15 10:07:04 -0400143
144 /* We can free old roots now. */
145 spin_lock(&trans->dropped_roots_lock);
146 while (!list_empty(&trans->dropped_roots)) {
147 root = list_first_entry(&trans->dropped_roots,
148 struct btrfs_root, root_list);
149 list_del_init(&root->root_list);
150 spin_unlock(&trans->dropped_roots_lock);
151 btrfs_drop_and_free_fs_root(fs_info, root);
152 spin_lock(&trans->dropped_roots_lock);
153 }
154 spin_unlock(&trans->dropped_roots_lock);
Josef Bacik9e351cc2014-03-13 15:42:13 -0400155 up_write(&fs_info->commit_root_sem);
Josef Bacik817d52f2009-07-13 21:29:25 -0400156}
157
Miao Xie0860adf2013-05-15 07:48:27 +0000158static inline void extwriter_counter_inc(struct btrfs_transaction *trans,
159 unsigned int type)
160{
161 if (type & TRANS_EXTWRITERS)
162 atomic_inc(&trans->num_extwriters);
163}
164
165static inline void extwriter_counter_dec(struct btrfs_transaction *trans,
166 unsigned int type)
167{
168 if (type & TRANS_EXTWRITERS)
169 atomic_dec(&trans->num_extwriters);
170}
171
172static inline void extwriter_counter_init(struct btrfs_transaction *trans,
173 unsigned int type)
174{
175 atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0));
176}
177
178static inline int extwriter_counter_read(struct btrfs_transaction *trans)
179{
180 return atomic_read(&trans->num_extwriters);
Miao Xie178260b2013-02-20 09:16:24 +0000181}
182
Chris Masond352ac62008-09-29 15:18:18 -0400183/*
184 * either allocate a new transaction or hop into the existing one
185 */
Miao Xie0860adf2013-05-15 07:48:27 +0000186static noinline int join_transaction(struct btrfs_root *root, unsigned int type)
Chris Mason79154b12007-03-22 15:59:16 -0400187{
188 struct btrfs_transaction *cur_trans;
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200189 struct btrfs_fs_info *fs_info = root->fs_info;
Josef Bacika4abeea2011-04-11 17:25:13 -0400190
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200191 spin_lock(&fs_info->trans_lock);
Chris Masond43317d2011-11-06 03:26:19 -0500192loop:
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100193 /* The file system has been taken offline. No new transactions. */
Miao Xie87533c42013-01-29 10:14:48 +0000194 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200195 spin_unlock(&fs_info->trans_lock);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100196 return -EROFS;
197 }
198
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200199 cur_trans = fs_info->running_transaction;
Josef Bacika4abeea2011-04-11 17:25:13 -0400200 if (cur_trans) {
David Sterba871383b2012-04-02 18:31:37 +0200201 if (cur_trans->aborted) {
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200202 spin_unlock(&fs_info->trans_lock);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100203 return cur_trans->aborted;
David Sterba871383b2012-04-02 18:31:37 +0200204 }
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000205 if (btrfs_blocked_trans_types[cur_trans->state] & type) {
Miao Xie178260b2013-02-20 09:16:24 +0000206 spin_unlock(&fs_info->trans_lock);
207 return -EBUSY;
208 }
Josef Bacika4abeea2011-04-11 17:25:13 -0400209 atomic_inc(&cur_trans->use_count);
Josef Bacik13c5a932011-04-11 15:45:29 -0400210 atomic_inc(&cur_trans->num_writers);
Miao Xie0860adf2013-05-15 07:48:27 +0000211 extwriter_counter_inc(cur_trans, type);
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200212 spin_unlock(&fs_info->trans_lock);
Josef Bacika4abeea2011-04-11 17:25:13 -0400213 return 0;
Chris Mason79154b12007-03-22 15:59:16 -0400214 }
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200215 spin_unlock(&fs_info->trans_lock);
Josef Bacika4abeea2011-04-11 17:25:13 -0400216
Miao Xie354aa0f2012-09-20 01:54:00 -0600217 /*
218 * If we are ATTACH, we just want to catch the current transaction,
219 * and commit it. If there is no transaction, just return ENOENT.
220 */
221 if (type == TRANS_ATTACH)
222 return -ENOENT;
223
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000224 /*
225 * JOIN_NOLOCK only happens during the transaction commit, so
226 * it is impossible that ->running_transaction is NULL
227 */
228 BUG_ON(type == TRANS_JOIN_NOLOCK);
229
Josef Bacika4abeea2011-04-11 17:25:13 -0400230 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS);
231 if (!cur_trans)
232 return -ENOMEM;
Chris Masond43317d2011-11-06 03:26:19 -0500233
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200234 spin_lock(&fs_info->trans_lock);
235 if (fs_info->running_transaction) {
Chris Masond43317d2011-11-06 03:26:19 -0500236 /*
237 * someone started a transaction after we unlocked. Make sure
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000238 * to redo the checks above
Chris Masond43317d2011-11-06 03:26:19 -0500239 */
Josef Bacika4abeea2011-04-11 17:25:13 -0400240 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
Chris Masond43317d2011-11-06 03:26:19 -0500241 goto loop;
Miao Xie87533c42013-01-29 10:14:48 +0000242 } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
Dan Carpentere4b50e12012-06-19 13:30:11 +0300243 spin_unlock(&fs_info->trans_lock);
Josef Bacik7b8b92a2012-05-31 15:52:43 -0400244 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
245 return -EROFS;
Josef Bacika4abeea2011-04-11 17:25:13 -0400246 }
Chris Masond43317d2011-11-06 03:26:19 -0500247
Josef Bacika4abeea2011-04-11 17:25:13 -0400248 atomic_set(&cur_trans->num_writers, 1);
Miao Xie0860adf2013-05-15 07:48:27 +0000249 extwriter_counter_init(cur_trans, type);
Josef Bacika4abeea2011-04-11 17:25:13 -0400250 init_waitqueue_head(&cur_trans->writer_wait);
251 init_waitqueue_head(&cur_trans->commit_wait);
Josef Bacik161c35492015-09-24 16:17:39 -0400252 init_waitqueue_head(&cur_trans->pending_wait);
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000253 cur_trans->state = TRANS_STATE_RUNNING;
Josef Bacika4abeea2011-04-11 17:25:13 -0400254 /*
255 * One for this trans handle, one so it will live on until we
256 * commit the transaction.
257 */
258 atomic_set(&cur_trans->use_count, 2);
Josef Bacik161c35492015-09-24 16:17:39 -0400259 atomic_set(&cur_trans->pending_ordered, 0);
Josef Bacik3204d332015-09-24 10:46:10 -0400260 cur_trans->flags = 0;
Josef Bacika4abeea2011-04-11 17:25:13 -0400261 cur_trans->start_time = get_seconds();
262
Alexandru Moisea099d0f2015-09-07 17:24:37 +0300263 memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs));
264
Liu Boc46effa2013-10-14 12:59:45 +0800265 cur_trans->delayed_refs.href_root = RB_ROOT;
Qu Wenruo3368d002015-04-16 14:34:17 +0800266 cur_trans->delayed_refs.dirty_extent_root = RB_ROOT;
Josef Bacikd7df2c72014-01-23 09:21:38 -0500267 atomic_set(&cur_trans->delayed_refs.num_entries, 0);
Jan Schmidt20b297d2012-05-20 15:43:53 +0200268
269 /*
270 * although the tree mod log is per file system and not per transaction,
271 * the log must never go across transaction boundaries.
272 */
273 smp_mb();
Julia Lawall31b1a2b2012-11-03 10:58:34 +0000274 if (!list_empty(&fs_info->tree_mod_seq_list))
Frank Holtonefe120a2013-12-20 11:37:06 -0500275 WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when "
Jan Schmidt20b297d2012-05-20 15:43:53 +0200276 "creating a fresh transaction\n");
Julia Lawall31b1a2b2012-11-03 10:58:34 +0000277 if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log))
Frank Holtonefe120a2013-12-20 11:37:06 -0500278 WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when "
Jan Schmidt20b297d2012-05-20 15:43:53 +0200279 "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 Bacika4abeea2011-04-11 17:25:13 -0400444 if (root->fs_info->log_root_recovering)
445 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
Tsutomu Itohdb5b4932011-03-23 08:14:16 +0000552 if (ret < 0) {
Miao Xie354aa0f2012-09-20 01:54:00 -0600553 /* We must get the transaction if we are JOIN_NOLOCK. */
554 BUG_ON(type == TRANS_JOIN_NOLOCK);
Miao Xie843fcf32013-01-28 12:36:22 +0000555 goto join_fail;
Tsutomu Itohdb5b4932011-03-23 08:14:16 +0000556 }
Chris Mason0f7d52f2007-04-09 10:42:37 -0400557
Yan, Zhenga22285a2010-05-16 10:48:46 -0400558 cur_trans = root->fs_info->running_transaction;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400559
560 h->transid = cur_trans->transid;
561 h->transaction = cur_trans;
Arne Jansend13603e2011-09-13 11:40:09 +0200562 h->root = root;
Josef Bacik2a1eb462011-04-13 15:15:59 -0400563 h->use_count = 1;
Qu Wenruo71741092015-09-08 17:22:41 +0800564
Miao Xiea698d0752012-09-20 01:51:59 -0600565 h->type = type;
Filipe Mananad9a05402015-10-03 13:13:13 +0100566 h->can_flush_pending_bgs = true;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200567 INIT_LIST_HEAD(&h->qgroup_ref_list);
Josef Bacikea658ba2012-09-11 16:57:25 -0400568 INIT_LIST_HEAD(&h->new_bgs);
Chris Masonb7ec40d2009-03-12 20:12:45 -0400569
Yan, Zhenga22285a2010-05-16 10:48:46 -0400570 smp_mb();
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000571 if (cur_trans->state >= TRANS_STATE_BLOCKED &&
572 may_wait_transaction(root, type)) {
Filipe Mananaabdd2e802014-06-24 17:46:58 +0100573 current->journal_info = h;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400574 btrfs_commit_transaction(h, root);
575 goto again;
576 }
Josef Bacik9ed74f22009-09-11 16:12:44 -0400577
Josef Bacikb5009942011-06-07 15:07:51 -0400578 if (num_bytes) {
Josef Bacik8c2a3ca2012-01-10 10:31:31 -0500579 trace_btrfs_space_reservation(root->fs_info, "transaction",
Liu Bo2bcc0322012-03-29 09:57:44 -0400580 h->transid, num_bytes, 1);
Josef Bacikb5009942011-06-07 15:07:51 -0400581 h->block_rsv = &root->fs_info->trans_block_rsv;
582 h->bytes_reserved = num_bytes;
Miao Xie20dd2cb2013-09-25 21:47:45 +0800583 h->reloc_reserved = reloc_reserved;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400584 }
585
Josef Bacik2a1eb462011-04-13 15:15:59 -0400586got_it:
Josef Bacika4abeea2011-04-11 17:25:13 -0400587 btrfs_record_root_in_trans(h, root);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400588
589 if (!current->journal_info && type != TRANS_USERSPACE)
590 current->journal_info = h;
Chris Mason79154b12007-03-22 15:59:16 -0400591 return h;
Miao Xie843fcf32013-01-28 12:36:22 +0000592
593join_fail:
Miao Xie0860adf2013-05-15 07:48:27 +0000594 if (type & __TRANS_FREEZABLE)
Miao Xie843fcf32013-01-28 12:36:22 +0000595 sb_end_intwrite(root->fs_info->sb);
596 kmem_cache_free(btrfs_trans_handle_cachep, h);
597alloc_fail:
598 if (num_bytes)
599 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
600 num_bytes);
601reserve_fail:
Qu Wenruo71741092015-09-08 17:22:41 +0800602 btrfs_qgroup_free_meta(root, qgroup_reserved);
Miao Xie843fcf32013-01-28 12:36:22 +0000603 return ERR_PTR(ret);
Chris Mason79154b12007-03-22 15:59:16 -0400604}
605
Chris Masonf9295742008-07-17 12:54:14 -0400606struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
Alexandru Moise5aed1dd2015-09-22 20:59:15 +0000607 unsigned int num_items)
Chris Masonf9295742008-07-17 12:54:14 -0400608{
Miao Xie08e007d2012-10-16 11:33:38 +0000609 return start_transaction(root, num_items, TRANS_START,
610 BTRFS_RESERVE_FLUSH_ALL);
Chris Masonf9295742008-07-17 12:54:14 -0400611}
Filipe Manana8eab77f2015-11-13 23:57:16 +0000612struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv(
613 struct btrfs_root *root,
614 unsigned int num_items,
615 int min_factor)
616{
617 struct btrfs_trans_handle *trans;
618 u64 num_bytes;
619 int ret;
620
621 trans = btrfs_start_transaction(root, num_items);
622 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
623 return trans;
624
625 trans = btrfs_start_transaction(root, 0);
626 if (IS_ERR(trans))
627 return trans;
628
629 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
630 ret = btrfs_cond_migrate_bytes(root->fs_info,
631 &root->fs_info->trans_block_rsv,
632 num_bytes,
633 min_factor);
634 if (ret) {
635 btrfs_end_transaction(trans, root);
636 return ERR_PTR(ret);
637 }
638
639 trans->block_rsv = &root->fs_info->trans_block_rsv;
640 trans->bytes_reserved = num_bytes;
Josef Bacik88d3a5a2016-01-13 13:21:20 -0500641 trace_btrfs_space_reservation(root->fs_info, "transaction",
642 trans->transid, num_bytes, 1);
Filipe Manana8eab77f2015-11-13 23:57:16 +0000643
644 return trans;
645}
Miao Xie8407aa42012-09-07 01:43:32 -0600646
Miao Xie08e007d2012-10-16 11:33:38 +0000647struct btrfs_trans_handle *btrfs_start_transaction_lflush(
Alexandru Moise5aed1dd2015-09-22 20:59:15 +0000648 struct btrfs_root *root,
649 unsigned int num_items)
Miao Xie8407aa42012-09-07 01:43:32 -0600650{
Miao Xie08e007d2012-10-16 11:33:38 +0000651 return start_transaction(root, num_items, TRANS_START,
652 BTRFS_RESERVE_FLUSH_LIMIT);
Miao Xie8407aa42012-09-07 01:43:32 -0600653}
654
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400655struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
Chris Masonf9295742008-07-17 12:54:14 -0400656{
Alexandru Moise575a75d2015-10-25 19:35:44 +0000657 return start_transaction(root, 0, TRANS_JOIN,
658 BTRFS_RESERVE_NO_FLUSH);
Chris Masonf9295742008-07-17 12:54:14 -0400659}
660
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400661struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
Josef Bacik0af3d002010-06-21 14:48:16 -0400662{
Alexandru Moise575a75d2015-10-25 19:35:44 +0000663 return start_transaction(root, 0, TRANS_JOIN_NOLOCK,
664 BTRFS_RESERVE_NO_FLUSH);
Josef Bacik0af3d002010-06-21 14:48:16 -0400665}
666
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400667struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
Sage Weil9ca9ee02008-08-04 10:41:27 -0400668{
Alexandru Moise575a75d2015-10-25 19:35:44 +0000669 return start_transaction(root, 0, TRANS_USERSPACE,
670 BTRFS_RESERVE_NO_FLUSH);
Sage Weil9ca9ee02008-08-04 10:41:27 -0400671}
672
Miao Xied4edf392013-02-20 09:17:06 +0000673/*
674 * btrfs_attach_transaction() - catch the running transaction
675 *
676 * It is used when we want to commit the current the transaction, but
677 * don't want to start a new one.
678 *
679 * Note: If this function return -ENOENT, it just means there is no
680 * running transaction. But it is possible that the inactive transaction
681 * is still in the memory, not fully on disk. If you hope there is no
682 * inactive transaction in the fs when -ENOENT is returned, you should
683 * invoke
684 * btrfs_attach_transaction_barrier()
685 */
Miao Xie354aa0f2012-09-20 01:54:00 -0600686struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
Josef Bacik60376ce2012-09-14 10:34:40 -0400687{
Alexandru Moise575a75d2015-10-25 19:35:44 +0000688 return start_transaction(root, 0, TRANS_ATTACH,
689 BTRFS_RESERVE_NO_FLUSH);
Josef Bacik60376ce2012-09-14 10:34:40 -0400690}
691
Miao Xied4edf392013-02-20 09:17:06 +0000692/*
Wang Sheng-Hui90b6d282013-06-14 16:21:24 +0800693 * btrfs_attach_transaction_barrier() - catch the running transaction
Miao Xied4edf392013-02-20 09:17:06 +0000694 *
695 * It is similar to the above function, the differentia is this one
696 * will wait for all the inactive transactions until they fully
697 * complete.
698 */
699struct btrfs_trans_handle *
700btrfs_attach_transaction_barrier(struct btrfs_root *root)
701{
702 struct btrfs_trans_handle *trans;
703
Alexandru Moise575a75d2015-10-25 19:35:44 +0000704 trans = start_transaction(root, 0, TRANS_ATTACH,
705 BTRFS_RESERVE_NO_FLUSH);
Miao Xied4edf392013-02-20 09:17:06 +0000706 if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT)
707 btrfs_wait_for_commit(root, 0);
708
709 return trans;
710}
711
Chris Masond352ac62008-09-29 15:18:18 -0400712/* wait for a transaction commit to be fully complete */
Li Zefanb9c83002011-07-14 03:17:14 +0000713static noinline void wait_for_commit(struct btrfs_root *root,
Chris Mason89ce8a62008-06-25 16:01:31 -0400714 struct btrfs_transaction *commit)
715{
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000716 wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED);
Chris Mason89ce8a62008-06-25 16:01:31 -0400717}
718
Sage Weil46204592010-10-29 15:41:32 -0400719int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
720{
721 struct btrfs_transaction *cur_trans = NULL, *t;
Miao Xie8cd28072012-11-26 08:42:07 +0000722 int ret = 0;
Sage Weil46204592010-10-29 15:41:32 -0400723
Sage Weil46204592010-10-29 15:41:32 -0400724 if (transid) {
725 if (transid <= root->fs_info->last_trans_committed)
Josef Bacika4abeea2011-04-11 17:25:13 -0400726 goto out;
Sage Weil46204592010-10-29 15:41:32 -0400727
728 /* find specified transaction */
Josef Bacika4abeea2011-04-11 17:25:13 -0400729 spin_lock(&root->fs_info->trans_lock);
Sage Weil46204592010-10-29 15:41:32 -0400730 list_for_each_entry(t, &root->fs_info->trans_list, list) {
731 if (t->transid == transid) {
732 cur_trans = t;
Josef Bacika4abeea2011-04-11 17:25:13 -0400733 atomic_inc(&cur_trans->use_count);
Miao Xie8cd28072012-11-26 08:42:07 +0000734 ret = 0;
Sage Weil46204592010-10-29 15:41:32 -0400735 break;
736 }
Miao Xie8cd28072012-11-26 08:42:07 +0000737 if (t->transid > transid) {
738 ret = 0;
Sage Weil46204592010-10-29 15:41:32 -0400739 break;
Miao Xie8cd28072012-11-26 08:42:07 +0000740 }
Sage Weil46204592010-10-29 15:41:32 -0400741 }
Josef Bacika4abeea2011-04-11 17:25:13 -0400742 spin_unlock(&root->fs_info->trans_lock);
Sage Weil42383022014-09-26 08:30:06 -0700743
744 /*
745 * The specified transaction doesn't exist, or we
746 * raced with btrfs_commit_transaction
747 */
748 if (!cur_trans) {
749 if (transid > root->fs_info->last_trans_committed)
750 ret = -EINVAL;
Miao Xie8cd28072012-11-26 08:42:07 +0000751 goto out;
Sage Weil42383022014-09-26 08:30:06 -0700752 }
Sage Weil46204592010-10-29 15:41:32 -0400753 } else {
754 /* find newest transaction that is committing | committed */
Josef Bacika4abeea2011-04-11 17:25:13 -0400755 spin_lock(&root->fs_info->trans_lock);
Sage Weil46204592010-10-29 15:41:32 -0400756 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
757 list) {
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000758 if (t->state >= TRANS_STATE_COMMIT_START) {
759 if (t->state == TRANS_STATE_COMPLETED)
Josef Bacik3473f3c2011-06-09 10:15:17 -0400760 break;
Sage Weil46204592010-10-29 15:41:32 -0400761 cur_trans = t;
Josef Bacika4abeea2011-04-11 17:25:13 -0400762 atomic_inc(&cur_trans->use_count);
Sage Weil46204592010-10-29 15:41:32 -0400763 break;
764 }
765 }
Josef Bacika4abeea2011-04-11 17:25:13 -0400766 spin_unlock(&root->fs_info->trans_lock);
Sage Weil46204592010-10-29 15:41:32 -0400767 if (!cur_trans)
Josef Bacika4abeea2011-04-11 17:25:13 -0400768 goto out; /* nothing committing|committed */
Sage Weil46204592010-10-29 15:41:32 -0400769 }
770
Sage Weil46204592010-10-29 15:41:32 -0400771 wait_for_commit(root, cur_trans);
Josef Bacik724e2312013-09-30 11:36:38 -0400772 btrfs_put_transaction(cur_trans);
Josef Bacika4abeea2011-04-11 17:25:13 -0400773out:
Sage Weil46204592010-10-29 15:41:32 -0400774 return ret;
775}
776
Chris Mason37d1aee2008-07-31 10:48:37 -0400777void btrfs_throttle(struct btrfs_root *root)
778{
Josef Bacika4abeea2011-04-11 17:25:13 -0400779 if (!atomic_read(&root->fs_info->open_ioctl_trans))
Sage Weil9ca9ee02008-08-04 10:41:27 -0400780 wait_current_trans(root);
Chris Mason37d1aee2008-07-31 10:48:37 -0400781}
782
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400783static int should_end_transaction(struct btrfs_trans_handle *trans,
784 struct btrfs_root *root)
785{
Josef Bacik1be41b72013-06-12 13:56:06 -0400786 if (root->fs_info->global_block_rsv.space_info->full &&
Josef Bacik0a2b2a82014-01-23 10:54:11 -0500787 btrfs_check_space_for_delayed_refs(trans, root))
Josef Bacik1be41b72013-06-12 13:56:06 -0400788 return 1;
Josef Bacik36ba0222011-10-18 12:15:48 -0400789
Josef Bacik1be41b72013-06-12 13:56:06 -0400790 return !!btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400791}
792
793int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
794 struct btrfs_root *root)
795{
796 struct btrfs_transaction *cur_trans = trans->transaction;
797 int updates;
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100798 int err;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400799
Josef Bacika4abeea2011-04-11 17:25:13 -0400800 smp_mb();
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000801 if (cur_trans->state >= TRANS_STATE_BLOCKED ||
802 cur_trans->delayed_refs.flushing)
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400803 return 1;
804
805 updates = trans->delayed_ref_updates;
806 trans->delayed_ref_updates = 0;
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100807 if (updates) {
Chris Mason28ed1342014-12-17 09:41:04 -0800808 err = btrfs_run_delayed_refs(trans, root, updates * 2);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100809 if (err) /* Error code will also eval true */
810 return err;
811 }
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400812
813 return should_end_transaction(trans, root);
814}
815
Chris Mason89ce8a62008-06-25 16:01:31 -0400816static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
Miao Xiea698d0752012-09-20 01:51:59 -0600817 struct btrfs_root *root, int throttle)
Chris Mason79154b12007-03-22 15:59:16 -0400818{
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400819 struct btrfs_transaction *cur_trans = trans->transaction;
Chris Masonab78c842008-07-29 16:15:18 -0400820 struct btrfs_fs_info *info = root->fs_info;
Josef Bacik1be41b72013-06-12 13:56:06 -0400821 unsigned long cur = trans->delayed_ref_updates;
Miao Xiea698d0752012-09-20 01:51:59 -0600822 int lock = (trans->type != TRANS_JOIN_NOLOCK);
Dave Jones4edc2ca2012-04-12 16:03:56 -0400823 int err = 0;
Chris Masona79b7d42014-05-22 16:18:52 -0700824 int must_run_delayed_refs = 0;
Chris Masond6e4a422007-04-06 15:37:36 -0400825
Josef Bacik3bbb24b2014-03-06 19:01:07 -0500826 if (trans->use_count > 1) {
827 trans->use_count--;
Josef Bacik2a1eb462011-04-13 15:15:59 -0400828 trans->block_rsv = trans->orig_rsv;
829 return 0;
830 }
831
Josef Bacikb24e03d2011-10-14 14:40:17 -0400832 btrfs_trans_release_metadata(trans, root);
Josef Bacik4c13d752011-08-30 11:31:29 -0400833 trans->block_rsv = NULL;
Arne Jansenc5567232011-09-14 15:44:05 +0200834
Josef Bacikea658ba2012-09-11 16:57:25 -0400835 if (!list_empty(&trans->new_bgs))
836 btrfs_create_pending_block_groups(trans, root);
837
Josef Bacik1be41b72013-06-12 13:56:06 -0400838 trans->delayed_ref_updates = 0;
Chris Masona79b7d42014-05-22 16:18:52 -0700839 if (!trans->sync) {
840 must_run_delayed_refs =
841 btrfs_should_throttle_delayed_refs(trans, root);
Josef Bacik0a2b2a82014-01-23 10:54:11 -0500842 cur = max_t(unsigned long, cur, 32);
Chris Masona79b7d42014-05-22 16:18:52 -0700843
844 /*
845 * don't make the caller wait if they are from a NOLOCK
846 * or ATTACH transaction, it will deadlock with commit
847 */
848 if (must_run_delayed_refs == 1 &&
849 (trans->type & (__TRANS_JOIN_NOLOCK | __TRANS_ATTACH)))
850 must_run_delayed_refs = 2;
Chris Mason56bec292009-03-13 10:10:06 -0400851 }
Chris Masonbb721702013-01-29 18:44:12 -0500852
Josef Bacik0e721102012-06-26 16:13:18 -0400853 btrfs_trans_release_metadata(trans, root);
854 trans->block_rsv = NULL;
Chris Mason56bec292009-03-13 10:10:06 -0400855
Josef Bacikea658ba2012-09-11 16:57:25 -0400856 if (!list_empty(&trans->new_bgs))
857 btrfs_create_pending_block_groups(trans, root);
858
Filipe Manana4fbcdf62015-05-20 14:01:54 +0100859 btrfs_trans_release_chunk_metadata(trans);
860
Josef Bacika4abeea2011-04-11 17:25:13 -0400861 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000862 should_end_transaction(trans, root) &&
863 ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) {
864 spin_lock(&info->trans_lock);
865 if (cur_trans->state == TRANS_STATE_RUNNING)
866 cur_trans->state = TRANS_STATE_BLOCKED;
867 spin_unlock(&info->trans_lock);
Josef Bacika4abeea2011-04-11 17:25:13 -0400868 }
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400869
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000870 if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) {
Josef Bacik3bbb24b2014-03-06 19:01:07 -0500871 if (throttle)
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400872 return btrfs_commit_transaction(trans, root);
Josef Bacik3bbb24b2014-03-06 19:01:07 -0500873 else
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400874 wake_up_process(info->transaction_kthread);
875 }
876
Miao Xie0860adf2013-05-15 07:48:27 +0000877 if (trans->type & __TRANS_FREEZABLE)
Josef Bacik98114652012-09-14 11:22:38 -0400878 sb_end_intwrite(root->fs_info->sb);
Josef Bacik6df78812012-09-05 08:08:30 -0600879
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400880 WARN_ON(cur_trans != info->running_transaction);
Josef Bacik13c5a932011-04-11 15:45:29 -0400881 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
882 atomic_dec(&cur_trans->num_writers);
Miao Xie0860adf2013-05-15 07:48:27 +0000883 extwriter_counter_dec(cur_trans, trans->type);
Chris Mason89ce8a62008-06-25 16:01:31 -0400884
David Sterbaa83342a2015-02-16 19:36:47 +0100885 /*
886 * Make sure counter is updated before we wake up waiters.
887 */
Sage Weil99d16cb2010-10-29 15:37:34 -0400888 smp_mb();
Chris Mason79154b12007-03-22 15:59:16 -0400889 if (waitqueue_active(&cur_trans->writer_wait))
890 wake_up(&cur_trans->writer_wait);
Josef Bacik724e2312013-09-30 11:36:38 -0400891 btrfs_put_transaction(cur_trans);
Josef Bacik9ed74f22009-09-11 16:12:44 -0400892
893 if (current->journal_info == trans)
894 current->journal_info = NULL;
Chris Masonab78c842008-07-29 16:15:18 -0400895
Yan, Zheng24bbcf02009-11-12 09:36:34 +0000896 if (throttle)
897 btrfs_run_delayed_iputs(root);
898
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100899 if (trans->aborted ||
Josef Bacik4e121c02013-09-27 16:32:39 -0400900 test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
901 wake_up_process(info->transaction_kthread);
Dave Jones4edc2ca2012-04-12 16:03:56 -0400902 err = -EIO;
Josef Bacik4e121c02013-09-27 16:32:39 -0400903 }
Jan Schmidtedf39272012-06-28 18:04:55 +0200904 assert_qgroups_uptodate(trans);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100905
Dave Jones4edc2ca2012-04-12 16:03:56 -0400906 kmem_cache_free(btrfs_trans_handle_cachep, trans);
Chris Masona79b7d42014-05-22 16:18:52 -0700907 if (must_run_delayed_refs) {
908 btrfs_async_run_delayed_refs(root, cur,
909 must_run_delayed_refs == 1);
910 }
Dave Jones4edc2ca2012-04-12 16:03:56 -0400911 return err;
Chris Mason79154b12007-03-22 15:59:16 -0400912}
913
Chris Mason89ce8a62008-06-25 16:01:31 -0400914int btrfs_end_transaction(struct btrfs_trans_handle *trans,
915 struct btrfs_root *root)
916{
Wang Shilong98ad43be2013-04-14 14:08:49 +0000917 return __btrfs_end_transaction(trans, root, 0);
Chris Mason89ce8a62008-06-25 16:01:31 -0400918}
919
920int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
921 struct btrfs_root *root)
922{
Wang Shilong98ad43be2013-04-14 14:08:49 +0000923 return __btrfs_end_transaction(trans, root, 1);
Miao Xie16cdcec2011-04-22 18:12:22 +0800924}
925
Chris Masond352ac62008-09-29 15:18:18 -0400926/*
927 * when btree blocks are allocated, they have some corresponding bits set for
928 * them in one of two extent_io trees. This is used to make sure all of
Chris Mason690587d2009-10-13 13:29:19 -0400929 * those extents are sent to disk but does not wait on them
Chris Masond352ac62008-09-29 15:18:18 -0400930 */
Chris Mason690587d2009-10-13 13:29:19 -0400931int btrfs_write_marked_extents(struct btrfs_root *root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000932 struct extent_io_tree *dirty_pages, int mark)
Chris Mason79154b12007-03-22 15:59:16 -0400933{
Chris Mason777e6bd2008-08-15 15:34:15 -0400934 int err = 0;
Chris Mason7c4452b2007-04-28 09:29:35 -0400935 int werr = 0;
Josef Bacik17283662011-09-26 13:58:47 -0400936 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
Josef Bacike6138872012-09-27 17:07:30 -0400937 struct extent_state *cached_state = NULL;
Chris Mason777e6bd2008-08-15 15:34:15 -0400938 u64 start = 0;
Chris Mason5f39d392007-10-15 16:14:19 -0400939 u64 end;
Chris Mason7c4452b2007-04-28 09:29:35 -0400940
Josef Bacik17283662011-09-26 13:58:47 -0400941 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
Josef Bacike6138872012-09-27 17:07:30 -0400942 mark, &cached_state)) {
Filipe Manana663dfbb2014-10-13 12:28:37 +0100943 bool wait_writeback = false;
944
945 err = convert_extent_bit(dirty_pages, start, end,
946 EXTENT_NEED_WAIT,
David Sterba210aa272016-04-26 23:54:39 +0200947 mark, &cached_state);
Filipe Manana663dfbb2014-10-13 12:28:37 +0100948 /*
949 * convert_extent_bit can return -ENOMEM, which is most of the
950 * time a temporary error. So when it happens, ignore the error
951 * and wait for writeback of this range to finish - because we
952 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
953 * to btrfs_wait_marked_extents() would not know that writeback
954 * for this range started and therefore wouldn't wait for it to
955 * finish - we don't want to commit a superblock that points to
956 * btree nodes/leafs for which writeback hasn't finished yet
957 * (and without errors).
958 * We cleanup any entries left in the io tree when committing
959 * the transaction (through clear_btree_io_tree()).
960 */
961 if (err == -ENOMEM) {
962 err = 0;
963 wait_writeback = true;
964 }
965 if (!err)
966 err = filemap_fdatawrite_range(mapping, start, end);
Josef Bacik17283662011-09-26 13:58:47 -0400967 if (err)
968 werr = err;
Filipe Manana663dfbb2014-10-13 12:28:37 +0100969 else if (wait_writeback)
970 werr = filemap_fdatawait_range(mapping, start, end);
Filipe Mananae38e2ed2014-10-13 12:28:38 +0100971 free_extent_state(cached_state);
Filipe Manana663dfbb2014-10-13 12:28:37 +0100972 cached_state = NULL;
Josef Bacik17283662011-09-26 13:58:47 -0400973 cond_resched();
974 start = end + 1;
Chris Mason7c4452b2007-04-28 09:29:35 -0400975 }
Chris Mason690587d2009-10-13 13:29:19 -0400976 return werr;
977}
978
979/*
980 * when btree blocks are allocated, they have some corresponding bits set for
981 * them in one of two extent_io trees. This is used to make sure all of
982 * those extents are on disk for transaction or log commit. We wait
983 * on all the pages and clear them from the dirty pages state tree
984 */
985int btrfs_wait_marked_extents(struct btrfs_root *root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000986 struct extent_io_tree *dirty_pages, int mark)
Chris Mason690587d2009-10-13 13:29:19 -0400987{
Chris Mason690587d2009-10-13 13:29:19 -0400988 int err = 0;
989 int werr = 0;
Josef Bacik17283662011-09-26 13:58:47 -0400990 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
Josef Bacike6138872012-09-27 17:07:30 -0400991 struct extent_state *cached_state = NULL;
Chris Mason690587d2009-10-13 13:29:19 -0400992 u64 start = 0;
993 u64 end;
Filipe Manana656f30d2014-09-26 12:25:56 +0100994 struct btrfs_inode *btree_ino = BTRFS_I(root->fs_info->btree_inode);
995 bool errors = false;
Chris Mason690587d2009-10-13 13:29:19 -0400996
Josef Bacik17283662011-09-26 13:58:47 -0400997 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
Josef Bacike6138872012-09-27 17:07:30 -0400998 EXTENT_NEED_WAIT, &cached_state)) {
Filipe Manana663dfbb2014-10-13 12:28:37 +0100999 /*
1000 * Ignore -ENOMEM errors returned by clear_extent_bit().
1001 * When committing the transaction, we'll remove any entries
1002 * left in the io tree. For a log commit, we don't remove them
1003 * after committing the log because the tree can be accessed
1004 * concurrently - we do it only at transaction commit time when
1005 * it's safe to do it (through clear_btree_io_tree()).
1006 */
1007 err = clear_extent_bit(dirty_pages, start, end,
1008 EXTENT_NEED_WAIT,
1009 0, 0, &cached_state, GFP_NOFS);
1010 if (err == -ENOMEM)
1011 err = 0;
1012 if (!err)
1013 err = filemap_fdatawait_range(mapping, start, end);
Josef Bacik17283662011-09-26 13:58:47 -04001014 if (err)
1015 werr = err;
Filipe Mananae38e2ed2014-10-13 12:28:38 +01001016 free_extent_state(cached_state);
1017 cached_state = NULL;
Josef Bacik17283662011-09-26 13:58:47 -04001018 cond_resched();
1019 start = end + 1;
Chris Mason777e6bd2008-08-15 15:34:15 -04001020 }
Chris Mason7c4452b2007-04-28 09:29:35 -04001021 if (err)
1022 werr = err;
Filipe Manana656f30d2014-09-26 12:25:56 +01001023
1024 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
1025 if ((mark & EXTENT_DIRTY) &&
1026 test_and_clear_bit(BTRFS_INODE_BTREE_LOG1_ERR,
1027 &btree_ino->runtime_flags))
1028 errors = true;
1029
1030 if ((mark & EXTENT_NEW) &&
1031 test_and_clear_bit(BTRFS_INODE_BTREE_LOG2_ERR,
1032 &btree_ino->runtime_flags))
1033 errors = true;
1034 } else {
1035 if (test_and_clear_bit(BTRFS_INODE_BTREE_ERR,
1036 &btree_ino->runtime_flags))
1037 errors = true;
1038 }
1039
1040 if (errors && !werr)
1041 werr = -EIO;
1042
Chris Mason7c4452b2007-04-28 09:29:35 -04001043 return werr;
Chris Mason79154b12007-03-22 15:59:16 -04001044}
1045
Chris Mason690587d2009-10-13 13:29:19 -04001046/*
1047 * when btree blocks are allocated, they have some corresponding bits set for
1048 * them in one of two extent_io trees. This is used to make sure all of
1049 * those extents are on disk for transaction or log commit
1050 */
Sergei Trofimovich171170c2013-08-14 23:27:46 +03001051static int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +00001052 struct extent_io_tree *dirty_pages, int mark)
Chris Mason690587d2009-10-13 13:29:19 -04001053{
1054 int ret;
1055 int ret2;
Miao Xiec6adc9c2013-05-28 10:05:39 +00001056 struct blk_plug plug;
Chris Mason690587d2009-10-13 13:29:19 -04001057
Miao Xiec6adc9c2013-05-28 10:05:39 +00001058 blk_start_plug(&plug);
Yan, Zheng8cef4e12009-11-12 09:33:26 +00001059 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
Miao Xiec6adc9c2013-05-28 10:05:39 +00001060 blk_finish_plug(&plug);
Yan, Zheng8cef4e12009-11-12 09:33:26 +00001061 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
Chris Masonbf0da8c2011-11-04 12:29:37 -04001062
1063 if (ret)
1064 return ret;
1065 if (ret2)
1066 return ret2;
1067 return 0;
Chris Mason690587d2009-10-13 13:29:19 -04001068}
1069
Filipe Manana663dfbb2014-10-13 12:28:37 +01001070static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
Chris Masond0c803c2008-09-11 16:17:57 -04001071 struct btrfs_root *root)
1072{
Filipe Manana663dfbb2014-10-13 12:28:37 +01001073 int ret;
1074
1075 ret = btrfs_write_and_wait_marked_extents(root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +00001076 &trans->transaction->dirty_pages,
1077 EXTENT_DIRTY);
Filipe Manana663dfbb2014-10-13 12:28:37 +01001078 clear_btree_io_tree(&trans->transaction->dirty_pages);
1079
1080 return ret;
Chris Masond0c803c2008-09-11 16:17:57 -04001081}
1082
Chris Masond352ac62008-09-29 15:18:18 -04001083/*
1084 * this is used to update the root pointer in the tree of tree roots.
1085 *
1086 * But, in the case of the extent allocation tree, updating the root
1087 * pointer may allocate blocks which may change the root of the extent
1088 * allocation tree.
1089 *
1090 * So, this loops and repeats and makes sure the cowonly root didn't
1091 * change while the root pointer was being updated in the metadata.
1092 */
Chris Mason0b86a832008-03-24 15:01:56 -04001093static int update_cowonly_root(struct btrfs_trans_handle *trans,
1094 struct btrfs_root *root)
1095{
1096 int ret;
1097 u64 old_root_bytenr;
Yan, Zheng86b9f2e2009-11-12 09:36:50 +00001098 u64 old_root_used;
Chris Mason0b86a832008-03-24 15:01:56 -04001099 struct btrfs_root *tree_root = root->fs_info->tree_root;
1100
Yan, Zheng86b9f2e2009-11-12 09:36:50 +00001101 old_root_used = btrfs_root_used(&root->root_item);
Chris Mason56bec292009-03-13 10:10:06 -04001102
Chris Masond3977122009-01-05 21:25:51 -05001103 while (1) {
Chris Mason0b86a832008-03-24 15:01:56 -04001104 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
Yan, Zheng86b9f2e2009-11-12 09:36:50 +00001105 if (old_root_bytenr == root->node->start &&
Josef Bacikea526d12015-03-13 16:40:45 -04001106 old_root_used == btrfs_root_used(&root->root_item))
Chris Mason0b86a832008-03-24 15:01:56 -04001107 break;
Chris Mason87ef2bb2008-10-30 11:23:27 -04001108
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001109 btrfs_set_root_node(&root->root_item, root->node);
Chris Mason0b86a832008-03-24 15:01:56 -04001110 ret = btrfs_update_root(trans, tree_root,
1111 &root->root_key,
1112 &root->root_item);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001113 if (ret)
1114 return ret;
Chris Mason56bec292009-03-13 10:10:06 -04001115
Yan, Zheng86b9f2e2009-11-12 09:36:50 +00001116 old_root_used = btrfs_root_used(&root->root_item);
Chris Mason0b86a832008-03-24 15:01:56 -04001117 }
Yan Zheng276e6802009-07-30 09:40:40 -04001118
Chris Mason0b86a832008-03-24 15:01:56 -04001119 return 0;
1120}
1121
Chris Masond352ac62008-09-29 15:18:18 -04001122/*
1123 * update all the cowonly tree roots on disk
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001124 *
1125 * The error handling in this function may not be obvious. Any of the
1126 * failures will cause the file system to go offline. We still need
1127 * to clean up the delayed refs.
Chris Masond352ac62008-09-29 15:18:18 -04001128 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001129static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
1130 struct btrfs_root *root)
Chris Mason79154b12007-03-22 15:59:16 -04001131{
Chris Mason79154b12007-03-22 15:59:16 -04001132 struct btrfs_fs_info *fs_info = root->fs_info;
Josef Bacikea526d12015-03-13 16:40:45 -04001133 struct list_head *dirty_bgs = &trans->transaction->dirty_bgs;
Chris Mason1bbc6212015-04-06 12:46:08 -07001134 struct list_head *io_bgs = &trans->transaction->io_bgs;
Chris Mason0b86a832008-03-24 15:01:56 -04001135 struct list_head *next;
Yan Zheng84234f32008-10-29 14:49:05 -04001136 struct extent_buffer *eb;
Chris Mason56bec292009-03-13 10:10:06 -04001137 int ret;
Yan Zheng84234f32008-10-29 14:49:05 -04001138
1139 eb = btrfs_lock_root_node(fs_info->tree_root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001140 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
1141 0, &eb);
Yan Zheng84234f32008-10-29 14:49:05 -04001142 btrfs_tree_unlock(eb);
1143 free_extent_buffer(eb);
Chris Mason79154b12007-03-22 15:59:16 -04001144
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001145 if (ret)
1146 return ret;
1147
Chris Mason56bec292009-03-13 10:10:06 -04001148 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001149 if (ret)
1150 return ret;
Chris Mason87ef2bb2008-10-30 11:23:27 -04001151
Stefan Behrens733f4fb2012-05-25 16:06:10 +02001152 ret = btrfs_run_dev_stats(trans, root->fs_info);
Josef Bacikc16ce192013-09-27 16:38:20 -04001153 if (ret)
1154 return ret;
Stefan Behrens8dabb742012-11-06 13:15:27 +01001155 ret = btrfs_run_dev_replace(trans, root->fs_info);
Josef Bacikc16ce192013-09-27 16:38:20 -04001156 if (ret)
1157 return ret;
Jan Schmidt546adb02012-06-14 16:37:44 +02001158 ret = btrfs_run_qgroups(trans, root->fs_info);
Josef Bacikc16ce192013-09-27 16:38:20 -04001159 if (ret)
1160 return ret;
Jan Schmidt546adb02012-06-14 16:37:44 +02001161
Josef Bacikdcdf7f62015-03-02 16:37:31 -05001162 ret = btrfs_setup_space_cache(trans, root);
1163 if (ret)
1164 return ret;
1165
Jan Schmidt546adb02012-06-14 16:37:44 +02001166 /* run_qgroups might have added some more refs */
1167 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
Josef Bacikc16ce192013-09-27 16:38:20 -04001168 if (ret)
1169 return ret;
Josef Bacikea526d12015-03-13 16:40:45 -04001170again:
Chris Masond3977122009-01-05 21:25:51 -05001171 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
Chris Mason0b86a832008-03-24 15:01:56 -04001172 next = fs_info->dirty_cowonly_roots.next;
1173 list_del_init(next);
1174 root = list_entry(next, struct btrfs_root, dirty_list);
Josef Bacike7070be2014-12-16 08:54:43 -08001175 clear_bit(BTRFS_ROOT_DIRTY, &root->state);
Chris Mason87ef2bb2008-10-30 11:23:27 -04001176
Josef Bacik9e351cc2014-03-13 15:42:13 -04001177 if (root != fs_info->extent_root)
1178 list_add_tail(&root->dirty_list,
1179 &trans->transaction->switch_commits);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001180 ret = update_cowonly_root(trans, root);
1181 if (ret)
1182 return ret;
Josef Bacikea526d12015-03-13 16:40:45 -04001183 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1184 if (ret)
1185 return ret;
Chris Mason79154b12007-03-22 15:59:16 -04001186 }
Yan Zheng276e6802009-07-30 09:40:40 -04001187
Chris Mason1bbc6212015-04-06 12:46:08 -07001188 while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) {
Josef Bacikea526d12015-03-13 16:40:45 -04001189 ret = btrfs_write_dirty_block_groups(trans, root);
1190 if (ret)
1191 return ret;
1192 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1193 if (ret)
1194 return ret;
1195 }
1196
1197 if (!list_empty(&fs_info->dirty_cowonly_roots))
1198 goto again;
1199
Josef Bacik9e351cc2014-03-13 15:42:13 -04001200 list_add_tail(&fs_info->extent_root->dirty_list,
1201 &trans->transaction->switch_commits);
Stefan Behrens8dabb742012-11-06 13:15:27 +01001202 btrfs_after_dev_replace_commit(fs_info);
1203
Chris Mason79154b12007-03-22 15:59:16 -04001204 return 0;
1205}
1206
Chris Masond352ac62008-09-29 15:18:18 -04001207/*
1208 * dead roots are old snapshots that need to be deleted. This allocates
1209 * a dirty root struct and adds it into the list of dead roots that need to
1210 * be deleted
1211 */
Josef Bacikcfad3922013-07-25 15:11:47 -04001212void btrfs_add_dead_root(struct btrfs_root *root)
Chris Mason5eda7b52007-06-22 14:16:25 -04001213{
Josef Bacika4abeea2011-04-11 17:25:13 -04001214 spin_lock(&root->fs_info->trans_lock);
Josef Bacikcfad3922013-07-25 15:11:47 -04001215 if (list_empty(&root->root_list))
1216 list_add_tail(&root->root_list, &root->fs_info->dead_roots);
Josef Bacika4abeea2011-04-11 17:25:13 -04001217 spin_unlock(&root->fs_info->trans_lock);
Chris Mason5eda7b52007-06-22 14:16:25 -04001218}
1219
Chris Masond352ac62008-09-29 15:18:18 -04001220/*
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001221 * update all the cowonly tree roots on disk
Chris Masond352ac62008-09-29 15:18:18 -04001222 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001223static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
1224 struct btrfs_root *root)
Chris Mason0f7d52f2007-04-09 10:42:37 -04001225{
Chris Mason0f7d52f2007-04-09 10:42:37 -04001226 struct btrfs_root *gang[8];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001227 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Mason0f7d52f2007-04-09 10:42:37 -04001228 int i;
1229 int ret;
Chris Mason54aa1f42007-06-22 14:16:25 -04001230 int err = 0;
1231
Josef Bacika4abeea2011-04-11 17:25:13 -04001232 spin_lock(&fs_info->fs_roots_radix_lock);
Chris Masond3977122009-01-05 21:25:51 -05001233 while (1) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001234 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1235 (void **)gang, 0,
Chris Mason0f7d52f2007-04-09 10:42:37 -04001236 ARRAY_SIZE(gang),
1237 BTRFS_ROOT_TRANS_TAG);
1238 if (ret == 0)
1239 break;
1240 for (i = 0; i < ret; i++) {
1241 root = gang[i];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001242 radix_tree_tag_clear(&fs_info->fs_roots_radix,
1243 (unsigned long)root->root_key.objectid,
1244 BTRFS_ROOT_TRANS_TAG);
Josef Bacika4abeea2011-04-11 17:25:13 -04001245 spin_unlock(&fs_info->fs_roots_radix_lock);
Yan Zheng31153d82008-07-28 15:32:19 -04001246
Chris Masone02119d2008-09-05 16:13:11 -04001247 btrfs_free_log(trans, root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001248 btrfs_update_reloc_root(trans, root);
Yan, Zhengd68fc572010-05-16 10:49:58 -04001249 btrfs_orphan_commit_root(trans, root);
Chris Masone02119d2008-09-05 16:13:11 -04001250
Li Zefan82d59022011-04-20 10:33:24 +08001251 btrfs_save_ino_cache(root, trans);
1252
Liu Bof1ebcc72011-11-14 20:48:06 -05001253 /* see comments in should_cow_block() */
Miao Xie27cdeb72014-04-02 19:51:05 +08001254 clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
Chris Masonc7548af2014-06-10 13:06:56 -07001255 smp_mb__after_atomic();
Liu Bof1ebcc72011-11-14 20:48:06 -05001256
Yan Zheng978d9102009-06-15 20:01:02 -04001257 if (root->commit_root != root->node) {
Josef Bacik9e351cc2014-03-13 15:42:13 -04001258 list_add_tail(&root->dirty_list,
1259 &trans->transaction->switch_commits);
Yan Zheng978d9102009-06-15 20:01:02 -04001260 btrfs_set_root_node(&root->root_item,
1261 root->node);
1262 }
Chris Mason9f3a7422007-08-07 15:52:19 -04001263
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001264 err = btrfs_update_root(trans, fs_info->tree_root,
Chris Mason0f7d52f2007-04-09 10:42:37 -04001265 &root->root_key,
1266 &root->root_item);
Josef Bacika4abeea2011-04-11 17:25:13 -04001267 spin_lock(&fs_info->fs_roots_radix_lock);
Chris Mason54aa1f42007-06-22 14:16:25 -04001268 if (err)
1269 break;
Qu Wenruo71741092015-09-08 17:22:41 +08001270 btrfs_qgroup_free_meta_all(root);
Chris Mason0f7d52f2007-04-09 10:42:37 -04001271 }
1272 }
Josef Bacika4abeea2011-04-11 17:25:13 -04001273 spin_unlock(&fs_info->fs_roots_radix_lock);
Chris Mason54aa1f42007-06-22 14:16:25 -04001274 return err;
Chris Mason0f7d52f2007-04-09 10:42:37 -04001275}
1276
Chris Masond352ac62008-09-29 15:18:18 -04001277/*
Eric Sandeende78b512013-01-31 18:21:12 +00001278 * defrag a given btree.
1279 * Every leaf in the btree is read and defragged.
Chris Masond352ac62008-09-29 15:18:18 -04001280 */
Eric Sandeende78b512013-01-31 18:21:12 +00001281int btrfs_defrag_root(struct btrfs_root *root)
Chris Masone9d0b132007-08-10 14:06:19 -04001282{
1283 struct btrfs_fs_info *info = root->fs_info;
Chris Masone9d0b132007-08-10 14:06:19 -04001284 struct btrfs_trans_handle *trans;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001285 int ret;
Chris Masone9d0b132007-08-10 14:06:19 -04001286
Miao Xie27cdeb72014-04-02 19:51:05 +08001287 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state))
Chris Masone9d0b132007-08-10 14:06:19 -04001288 return 0;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001289
Chris Mason6b800532007-10-15 16:17:34 -04001290 while (1) {
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001291 trans = btrfs_start_transaction(root, 0);
1292 if (IS_ERR(trans))
1293 return PTR_ERR(trans);
1294
Eric Sandeende78b512013-01-31 18:21:12 +00001295 ret = btrfs_defrag_leaves(trans, root);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001296
Chris Masone9d0b132007-08-10 14:06:19 -04001297 btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00001298 btrfs_btree_balance_dirty(info->tree_root);
Chris Masone9d0b132007-08-10 14:06:19 -04001299 cond_resched();
1300
David Sterba7841cb22011-05-31 18:07:27 +02001301 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
Chris Masone9d0b132007-08-10 14:06:19 -04001302 break;
David Sterba210549e2013-02-09 23:38:06 +00001303
1304 if (btrfs_defrag_cancelled(root->fs_info)) {
Frank Holtonefe120a2013-12-20 11:37:06 -05001305 pr_debug("BTRFS: defrag_root cancelled\n");
David Sterba210549e2013-02-09 23:38:06 +00001306 ret = -EAGAIN;
1307 break;
1308 }
Chris Masone9d0b132007-08-10 14:06:19 -04001309 }
Miao Xie27cdeb72014-04-02 19:51:05 +08001310 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001311 return ret;
Chris Masone9d0b132007-08-10 14:06:19 -04001312}
1313
David Sterba2c1984f2016-05-12 11:05:03 +02001314/* Bisesctability fixup, remove in 4.8 */
1315#ifndef btrfs_std_error
1316#define btrfs_std_error btrfs_handle_fs_error
1317#endif
1318
Chris Masond352ac62008-09-29 15:18:18 -04001319/*
Qu Wenruo6426c7a2016-05-11 12:53:52 -07001320 * Do all special snapshot related qgroup dirty hack.
1321 *
1322 * Will do all needed qgroup inherit and dirty hack like switch commit
1323 * roots inside one transaction and write all btree into disk, to make
1324 * qgroup works.
1325 */
1326static int qgroup_account_snapshot(struct btrfs_trans_handle *trans,
1327 struct btrfs_root *src,
1328 struct btrfs_root *parent,
1329 struct btrfs_qgroup_inherit *inherit,
1330 u64 dst_objectid)
1331{
1332 struct btrfs_fs_info *fs_info = src->fs_info;
1333 int ret;
1334
1335 /*
1336 * Save some performance in the case that qgroups are not
1337 * enabled. If this check races with the ioctl, rescan will
1338 * kick in anyway.
1339 */
1340 mutex_lock(&fs_info->qgroup_ioctl_lock);
1341 if (!fs_info->quota_enabled) {
1342 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1343 return 0;
1344 }
1345 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1346
1347 /*
1348 * We are going to commit transaction, see btrfs_commit_transaction()
1349 * comment for reason locking tree_log_mutex
1350 */
1351 mutex_lock(&fs_info->tree_log_mutex);
1352
1353 ret = commit_fs_roots(trans, src);
1354 if (ret)
1355 goto out;
1356 ret = btrfs_qgroup_prepare_account_extents(trans, fs_info);
1357 if (ret < 0)
1358 goto out;
1359 ret = btrfs_qgroup_account_extents(trans, fs_info);
1360 if (ret < 0)
1361 goto out;
1362
1363 /* Now qgroup are all updated, we can inherit it to new qgroups */
1364 ret = btrfs_qgroup_inherit(trans, fs_info,
1365 src->root_key.objectid, dst_objectid,
1366 inherit);
1367 if (ret < 0)
1368 goto out;
1369
1370 /*
1371 * Now we do a simplified commit transaction, which will:
1372 * 1) commit all subvolume and extent tree
1373 * To ensure all subvolume and extent tree have a valid
1374 * commit_root to accounting later insert_dir_item()
1375 * 2) write all btree blocks onto disk
1376 * This is to make sure later btree modification will be cowed
1377 * Or commit_root can be populated and cause wrong qgroup numbers
1378 * In this simplified commit, we don't really care about other trees
1379 * like chunk and root tree, as they won't affect qgroup.
1380 * And we don't write super to avoid half committed status.
1381 */
1382 ret = commit_cowonly_roots(trans, src);
1383 if (ret)
1384 goto out;
1385 switch_commit_roots(trans->transaction, fs_info);
1386 ret = btrfs_write_and_wait_transaction(trans, src);
1387 if (ret)
1388 btrfs_std_error(fs_info, ret,
1389 "Error while writing out transaction for qgroup");
1390
1391out:
1392 mutex_unlock(&fs_info->tree_log_mutex);
1393
1394 /*
1395 * Force parent root to be updated, as we recorded it before so its
1396 * last_trans == cur_transid.
1397 * Or it won't be committed again onto disk after later
1398 * insert_dir_item()
1399 */
1400 if (!ret)
1401 record_root_in_trans(trans, parent, 1);
1402 return ret;
1403}
1404
1405/*
Chris Masond352ac62008-09-29 15:18:18 -04001406 * new snapshots need to be created at a very specific time in the
Miao Xieaec80302013-03-04 09:44:29 +00001407 * transaction commit. This does the actual creation.
1408 *
1409 * Note:
1410 * If the error which may affect the commitment of the current transaction
1411 * happens, we should return the error number. If the error which just affect
1412 * the creation of the pending snapshots, just return 0.
Chris Masond352ac62008-09-29 15:18:18 -04001413 */
Chris Mason80b67942008-02-01 16:35:04 -05001414static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
Chris Mason3063d292008-01-08 15:46:30 -05001415 struct btrfs_fs_info *fs_info,
1416 struct btrfs_pending_snapshot *pending)
1417{
1418 struct btrfs_key key;
Chris Mason80b67942008-02-01 16:35:04 -05001419 struct btrfs_root_item *new_root_item;
Chris Mason3063d292008-01-08 15:46:30 -05001420 struct btrfs_root *tree_root = fs_info->tree_root;
1421 struct btrfs_root *root = pending->root;
Sage Weil6bdb72d2010-03-15 17:27:13 +00001422 struct btrfs_root *parent_root;
Liu Bo98c99422011-09-11 10:52:24 -04001423 struct btrfs_block_rsv *rsv;
Sage Weil6bdb72d2010-03-15 17:27:13 +00001424 struct inode *parent_inode;
Miao Xie42874b32012-09-06 04:03:32 -06001425 struct btrfs_path *path;
1426 struct btrfs_dir_item *dir_item;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001427 struct dentry *dentry;
Chris Mason3063d292008-01-08 15:46:30 -05001428 struct extent_buffer *tmp;
Chris Mason925baed2008-06-25 16:01:30 -04001429 struct extent_buffer *old;
Deepa Dinamani04b285f2016-02-06 23:57:21 -08001430 struct timespec cur_time;
Miao Xieaec80302013-03-04 09:44:29 +00001431 int ret = 0;
Yan, Zhengd68fc572010-05-16 10:49:58 -04001432 u64 to_reserve = 0;
Sage Weil6bdb72d2010-03-15 17:27:13 +00001433 u64 index = 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001434 u64 objectid;
Li Zefanb83cc962010-12-20 16:04:08 +08001435 u64 root_flags;
Alexander Block8ea05e32012-07-25 17:35:53 +02001436 uuid_le new_uuid;
Chris Mason3063d292008-01-08 15:46:30 -05001437
David Sterba8546b572015-11-10 18:54:03 +01001438 ASSERT(pending->path);
1439 path = pending->path;
Miao Xie42874b32012-09-06 04:03:32 -06001440
David Sterbab0c0ea62015-11-10 18:54:00 +01001441 ASSERT(pending->root_item);
1442 new_root_item = pending->root_item;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001443
Miao Xieaec80302013-03-04 09:44:29 +00001444 pending->error = btrfs_find_free_objectid(tree_root, &objectid);
1445 if (pending->error)
Miao Xie6fa97002012-09-06 04:00:32 -06001446 goto no_free_objectid;
Chris Mason3063d292008-01-08 15:46:30 -05001447
Qu Wenruod6726332015-04-20 10:09:06 +08001448 /*
1449 * Make qgroup to skip current new snapshot's qgroupid, as it is
1450 * accounted by later btrfs_qgroup_inherit().
1451 */
1452 btrfs_set_skip_qgroup(trans, objectid);
1453
Zhaolei147d2562015-08-06 20:58:11 +08001454 btrfs_reloc_pre_snapshot(pending, &to_reserve);
Yan, Zhengd68fc572010-05-16 10:49:58 -04001455
1456 if (to_reserve > 0) {
Miao Xieaec80302013-03-04 09:44:29 +00001457 pending->error = btrfs_block_rsv_add(root,
1458 &pending->block_rsv,
1459 to_reserve,
1460 BTRFS_RESERVE_NO_FLUSH);
1461 if (pending->error)
Qu Wenruod6726332015-04-20 10:09:06 +08001462 goto clear_skip_qgroup;
Yan, Zhengd68fc572010-05-16 10:49:58 -04001463 }
1464
Chris Mason3063d292008-01-08 15:46:30 -05001465 key.objectid = objectid;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001466 key.offset = (u64)-1;
1467 key.type = BTRFS_ROOT_ITEM_KEY;
Chris Mason3063d292008-01-08 15:46:30 -05001468
Miao Xie6fa97002012-09-06 04:00:32 -06001469 rsv = trans->block_rsv;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001470 trans->block_rsv = &pending->block_rsv;
Liu Bo2382c5c2013-02-22 04:33:36 +00001471 trans->bytes_reserved = trans->block_rsv->reserved;
Josef Bacik88d3a5a2016-01-13 13:21:20 -05001472 trace_btrfs_space_reservation(root->fs_info, "transaction",
1473 trans->transid,
1474 trans->bytes_reserved, 1);
Yan, Zhenga22285a2010-05-16 10:48:46 -04001475 dentry = pending->dentry;
Miao Xiee9662f72013-02-28 10:01:15 +00001476 parent_inode = pending->dir;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001477 parent_root = BTRFS_I(parent_inode)->root;
Qu Wenruo6426c7a2016-05-11 12:53:52 -07001478 record_root_in_trans(trans, parent_root, 0);
Yan, Zhenga22285a2010-05-16 10:48:46 -04001479
Deepa Dinamani04b285f2016-02-06 23:57:21 -08001480 cur_time = current_fs_time(parent_inode->i_sb);
1481
Sage Weil6bdb72d2010-03-15 17:27:13 +00001482 /*
1483 * insert the directory item
1484 */
Sage Weil6bdb72d2010-03-15 17:27:13 +00001485 ret = btrfs_set_inode_index(parent_inode, &index);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001486 BUG_ON(ret); /* -ENOMEM */
Miao Xie42874b32012-09-06 04:03:32 -06001487
1488 /* check if there is a file/dir which has the same name. */
1489 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1490 btrfs_ino(parent_inode),
1491 dentry->d_name.name,
1492 dentry->d_name.len, 0);
1493 if (dir_item != NULL && !IS_ERR(dir_item)) {
Chris Masonfe66a052012-02-20 08:40:56 -05001494 pending->error = -EEXIST;
Miao Xieaec80302013-03-04 09:44:29 +00001495 goto dir_item_existed;
Miao Xie42874b32012-09-06 04:03:32 -06001496 } else if (IS_ERR(dir_item)) {
1497 ret = PTR_ERR(dir_item);
Miao Xie8732d442012-09-17 23:52:38 -06001498 btrfs_abort_transaction(trans, root, ret);
1499 goto fail;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001500 }
Miao Xie42874b32012-09-06 04:03:32 -06001501 btrfs_release_path(path);
Sage Weil6bdb72d2010-03-15 17:27:13 +00001502
Chris Masone9993762011-06-17 16:14:09 -04001503 /*
1504 * pull in the delayed directory update
1505 * and the delayed inode item
1506 * otherwise we corrupt the FS during
1507 * snapshot
1508 */
1509 ret = btrfs_run_delayed_items(trans, root);
Miao Xie8732d442012-09-17 23:52:38 -06001510 if (ret) { /* Transaction aborted */
1511 btrfs_abort_transaction(trans, root, ret);
1512 goto fail;
1513 }
Chris Masone9993762011-06-17 16:14:09 -04001514
Qu Wenruo6426c7a2016-05-11 12:53:52 -07001515 record_root_in_trans(trans, root, 0);
Sage Weil6bdb72d2010-03-15 17:27:13 +00001516 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1517 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
Li Zefan08fe4db2011-03-28 02:01:25 +00001518 btrfs_check_and_init_root_item(new_root_item);
Sage Weil6bdb72d2010-03-15 17:27:13 +00001519
Li Zefanb83cc962010-12-20 16:04:08 +08001520 root_flags = btrfs_root_flags(new_root_item);
1521 if (pending->readonly)
1522 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1523 else
1524 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1525 btrfs_set_root_flags(new_root_item, root_flags);
1526
Alexander Block8ea05e32012-07-25 17:35:53 +02001527 btrfs_set_root_generation_v2(new_root_item,
1528 trans->transid);
1529 uuid_le_gen(&new_uuid);
1530 memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
1531 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1532 BTRFS_UUID_SIZE);
Stefan Behrens70023da2013-04-17 09:11:47 +00001533 if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1534 memset(new_root_item->received_uuid, 0,
1535 sizeof(new_root_item->received_uuid));
1536 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1537 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1538 btrfs_set_root_stransid(new_root_item, 0);
1539 btrfs_set_root_rtransid(new_root_item, 0);
1540 }
Qu Wenruo3cae2102013-07-16 11:19:18 +08001541 btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1542 btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
Alexander Block8ea05e32012-07-25 17:35:53 +02001543 btrfs_set_root_otransid(new_root_item, trans->transid);
Alexander Block8ea05e32012-07-25 17:35:53 +02001544
Chris Mason925baed2008-06-25 16:01:30 -04001545 old = btrfs_lock_root_node(root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001546 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001547 if (ret) {
1548 btrfs_tree_unlock(old);
1549 free_extent_buffer(old);
Miao Xie8732d442012-09-17 23:52:38 -06001550 btrfs_abort_transaction(trans, root, ret);
1551 goto fail;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001552 }
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001553
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001554 btrfs_set_lock_blocking(old);
Chris Mason3063d292008-01-08 15:46:30 -05001555
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001556 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001557 /* clean up in any case */
Chris Mason925baed2008-06-25 16:01:30 -04001558 btrfs_tree_unlock(old);
1559 free_extent_buffer(old);
Miao Xie8732d442012-09-17 23:52:38 -06001560 if (ret) {
1561 btrfs_abort_transaction(trans, root, ret);
1562 goto fail;
1563 }
Liu Bof1ebcc72011-11-14 20:48:06 -05001564 /* see comments in should_cow_block() */
Miao Xie27cdeb72014-04-02 19:51:05 +08001565 set_bit(BTRFS_ROOT_FORCE_COW, &root->state);
Liu Bof1ebcc72011-11-14 20:48:06 -05001566 smp_wmb();
1567
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001568 btrfs_set_root_node(new_root_item, tmp);
Yan, Zhenga22285a2010-05-16 10:48:46 -04001569 /* record when the snapshot was created in key.offset */
1570 key.offset = trans->transid;
1571 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
Chris Mason925baed2008-06-25 16:01:30 -04001572 btrfs_tree_unlock(tmp);
Chris Mason3063d292008-01-08 15:46:30 -05001573 free_extent_buffer(tmp);
Miao Xie8732d442012-09-17 23:52:38 -06001574 if (ret) {
1575 btrfs_abort_transaction(trans, root, ret);
1576 goto fail;
1577 }
Chris Mason0660b5a2008-11-17 20:37:39 -05001578
Yan, Zhenga22285a2010-05-16 10:48:46 -04001579 /*
1580 * insert root back/forward references
1581 */
1582 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1583 parent_root->root_key.objectid,
Li Zefan33345d012011-04-20 10:31:50 +08001584 btrfs_ino(parent_inode), index,
Yan, Zhenga22285a2010-05-16 10:48:46 -04001585 dentry->d_name.name, dentry->d_name.len);
Miao Xie8732d442012-09-17 23:52:38 -06001586 if (ret) {
1587 btrfs_abort_transaction(trans, root, ret);
1588 goto fail;
1589 }
Yan, Zhenga22285a2010-05-16 10:48:46 -04001590
1591 key.offset = (u64)-1;
1592 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001593 if (IS_ERR(pending->snap)) {
1594 ret = PTR_ERR(pending->snap);
Miao Xie8732d442012-09-17 23:52:38 -06001595 btrfs_abort_transaction(trans, root, ret);
1596 goto fail;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001597 }
Yan, Zhengd68fc572010-05-16 10:49:58 -04001598
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001599 ret = btrfs_reloc_post_snapshot(trans, pending);
Miao Xie8732d442012-09-17 23:52:38 -06001600 if (ret) {
1601 btrfs_abort_transaction(trans, root, ret);
1602 goto fail;
1603 }
Miao Xie361048f2012-09-06 04:00:57 -06001604
1605 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
Miao Xie8732d442012-09-17 23:52:38 -06001606 if (ret) {
1607 btrfs_abort_transaction(trans, root, ret);
1608 goto fail;
1609 }
Miao Xie42874b32012-09-06 04:03:32 -06001610
Qu Wenruo6426c7a2016-05-11 12:53:52 -07001611 /*
1612 * Do special qgroup accounting for snapshot, as we do some qgroup
1613 * snapshot hack to do fast snapshot.
1614 * To co-operate with that hack, we do hack again.
1615 * Or snapshot will be greatly slowed down by a subtree qgroup rescan
1616 */
1617 ret = qgroup_account_snapshot(trans, root, parent_root,
1618 pending->inherit, objectid);
1619 if (ret < 0)
1620 goto fail;
1621
Miao Xie42874b32012-09-06 04:03:32 -06001622 ret = btrfs_insert_dir_item(trans, parent_root,
1623 dentry->d_name.name, dentry->d_name.len,
1624 parent_inode, &key,
1625 BTRFS_FT_DIR, index);
1626 /* We have check then name at the beginning, so it is impossible. */
Chris Mason9c520572012-12-17 14:26:57 -05001627 BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
Miao Xie8732d442012-09-17 23:52:38 -06001628 if (ret) {
1629 btrfs_abort_transaction(trans, root, ret);
1630 goto fail;
1631 }
Miao Xie42874b32012-09-06 04:03:32 -06001632
1633 btrfs_i_size_write(parent_inode, parent_inode->i_size +
1634 dentry->d_name.len * 2);
Deepa Dinamani04b285f2016-02-06 23:57:21 -08001635 parent_inode->i_mtime = parent_inode->i_ctime =
1636 current_fs_time(parent_inode->i_sb);
Josef Bacikbe6aef62012-10-22 15:43:12 -04001637 ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode);
Stefan Behrensdd5f9612013-08-15 17:11:20 +02001638 if (ret) {
Miao Xie8732d442012-09-17 23:52:38 -06001639 btrfs_abort_transaction(trans, root, ret);
Stefan Behrensdd5f9612013-08-15 17:11:20 +02001640 goto fail;
1641 }
1642 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, new_uuid.b,
1643 BTRFS_UUID_KEY_SUBVOL, objectid);
1644 if (ret) {
1645 btrfs_abort_transaction(trans, root, ret);
1646 goto fail;
1647 }
1648 if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1649 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
1650 new_root_item->received_uuid,
1651 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1652 objectid);
1653 if (ret && ret != -EEXIST) {
1654 btrfs_abort_transaction(trans, root, ret);
1655 goto fail;
1656 }
1657 }
Qu Wenruod6726332015-04-20 10:09:06 +08001658
1659 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1660 if (ret) {
1661 btrfs_abort_transaction(trans, root, ret);
1662 goto fail;
1663 }
1664
Chris Mason3063d292008-01-08 15:46:30 -05001665fail:
Miao Xieaec80302013-03-04 09:44:29 +00001666 pending->error = ret;
1667dir_item_existed:
Liu Bo98c99422011-09-11 10:52:24 -04001668 trans->block_rsv = rsv;
Liu Bo2382c5c2013-02-22 04:33:36 +00001669 trans->bytes_reserved = 0;
Qu Wenruod6726332015-04-20 10:09:06 +08001670clear_skip_qgroup:
1671 btrfs_clear_skip_qgroup(trans);
Miao Xie6fa97002012-09-06 04:00:32 -06001672no_free_objectid:
1673 kfree(new_root_item);
David Sterbab0c0ea62015-11-10 18:54:00 +01001674 pending->root_item = NULL;
Miao Xie42874b32012-09-06 04:03:32 -06001675 btrfs_free_path(path);
David Sterba8546b572015-11-10 18:54:03 +01001676 pending->path = NULL;
1677
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001678 return ret;
Chris Mason3063d292008-01-08 15:46:30 -05001679}
1680
Chris Masond352ac62008-09-29 15:18:18 -04001681/*
1682 * create all the snapshots we've scheduled for creation
1683 */
Chris Mason80b67942008-02-01 16:35:04 -05001684static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1685 struct btrfs_fs_info *fs_info)
Chris Mason3063d292008-01-08 15:46:30 -05001686{
Miao Xieaec80302013-03-04 09:44:29 +00001687 struct btrfs_pending_snapshot *pending, *next;
Chris Mason3063d292008-01-08 15:46:30 -05001688 struct list_head *head = &trans->transaction->pending_snapshots;
Miao Xieaec80302013-03-04 09:44:29 +00001689 int ret = 0;
Chris Mason3de45862008-11-17 21:02:50 -05001690
Miao Xieaec80302013-03-04 09:44:29 +00001691 list_for_each_entry_safe(pending, next, head, list) {
1692 list_del(&pending->list);
1693 ret = create_pending_snapshot(trans, fs_info, pending);
1694 if (ret)
1695 break;
1696 }
1697 return ret;
Chris Mason3de45862008-11-17 21:02:50 -05001698}
1699
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001700static void update_super_roots(struct btrfs_root *root)
1701{
1702 struct btrfs_root_item *root_item;
1703 struct btrfs_super_block *super;
1704
David Sterba6c417612011-04-13 15:41:04 +02001705 super = root->fs_info->super_copy;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001706
1707 root_item = &root->fs_info->chunk_root->root_item;
1708 super->chunk_root = root_item->bytenr;
1709 super->chunk_root_generation = root_item->generation;
1710 super->chunk_root_level = root_item->level;
1711
1712 root_item = &root->fs_info->tree_root->root_item;
1713 super->root = root_item->bytenr;
1714 super->generation = root_item->generation;
1715 super->root_level = root_item->level;
Josef Bacik73bc1872011-10-03 14:07:49 -04001716 if (btrfs_test_opt(root, SPACE_CACHE))
Josef Bacik0af3d002010-06-21 14:48:16 -04001717 super->cache_generation = root_item->generation;
Stefan Behrens70f80172013-08-15 17:11:23 +02001718 if (root->fs_info->update_uuid_tree_gen)
1719 super->uuid_tree_generation = root_item->generation;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001720}
1721
Chris Masonf36f3042009-07-30 10:04:48 -04001722int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1723{
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001724 struct btrfs_transaction *trans;
Chris Masonf36f3042009-07-30 10:04:48 -04001725 int ret = 0;
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001726
Josef Bacika4abeea2011-04-11 17:25:13 -04001727 spin_lock(&info->trans_lock);
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001728 trans = info->running_transaction;
1729 if (trans)
1730 ret = (trans->state >= TRANS_STATE_COMMIT_START);
Josef Bacika4abeea2011-04-11 17:25:13 -04001731 spin_unlock(&info->trans_lock);
Chris Masonf36f3042009-07-30 10:04:48 -04001732 return ret;
1733}
1734
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001735int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1736{
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001737 struct btrfs_transaction *trans;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001738 int ret = 0;
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001739
Josef Bacika4abeea2011-04-11 17:25:13 -04001740 spin_lock(&info->trans_lock);
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001741 trans = info->running_transaction;
1742 if (trans)
1743 ret = is_transaction_blocked(trans);
Josef Bacika4abeea2011-04-11 17:25:13 -04001744 spin_unlock(&info->trans_lock);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001745 return ret;
1746}
1747
Sage Weilbb9c12c2010-10-29 15:37:34 -04001748/*
1749 * wait for the current transaction commit to start and block subsequent
1750 * transaction joins
1751 */
1752static void wait_current_trans_commit_start(struct btrfs_root *root,
1753 struct btrfs_transaction *trans)
1754{
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001755 wait_event(root->fs_info->transaction_blocked_wait,
Josef Bacik501407a2013-06-10 16:47:23 -04001756 trans->state >= TRANS_STATE_COMMIT_START ||
1757 trans->aborted);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001758}
1759
1760/*
1761 * wait for the current transaction to start and then become unblocked.
1762 * caller holds ref.
1763 */
1764static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1765 struct btrfs_transaction *trans)
1766{
Li Zefan72d63ed2011-07-14 03:17:00 +00001767 wait_event(root->fs_info->transaction_wait,
Josef Bacik501407a2013-06-10 16:47:23 -04001768 trans->state >= TRANS_STATE_UNBLOCKED ||
1769 trans->aborted);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001770}
1771
1772/*
1773 * commit transactions asynchronously. once btrfs_commit_transaction_async
1774 * returns, any subsequent transaction will not be allowed to join.
1775 */
1776struct btrfs_async_commit {
1777 struct btrfs_trans_handle *newtrans;
1778 struct btrfs_root *root;
Miao Xie7892b5a2012-11-15 08:14:47 +00001779 struct work_struct work;
Sage Weilbb9c12c2010-10-29 15:37:34 -04001780};
1781
1782static void do_async_commit(struct work_struct *work)
1783{
1784 struct btrfs_async_commit *ac =
Miao Xie7892b5a2012-11-15 08:14:47 +00001785 container_of(work, struct btrfs_async_commit, work);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001786
Sage Weil6fc4e35482012-08-30 16:26:15 -06001787 /*
1788 * We've got freeze protection passed with the transaction.
1789 * Tell lockdep about it.
1790 */
Liu Bob1a06a42013-11-06 16:57:55 +08001791 if (ac->newtrans->type & __TRANS_FREEZABLE)
Oleg Nesterovbee91822015-07-19 23:48:20 +02001792 __sb_writers_acquired(ac->root->fs_info->sb, SB_FREEZE_FS);
Sage Weil6fc4e35482012-08-30 16:26:15 -06001793
Sage Weile209db72012-08-30 16:26:16 -06001794 current->journal_info = ac->newtrans;
1795
Sage Weilbb9c12c2010-10-29 15:37:34 -04001796 btrfs_commit_transaction(ac->newtrans, ac->root);
1797 kfree(ac);
1798}
1799
1800int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1801 struct btrfs_root *root,
1802 int wait_for_unblock)
1803{
1804 struct btrfs_async_commit *ac;
1805 struct btrfs_transaction *cur_trans;
1806
1807 ac = kmalloc(sizeof(*ac), GFP_NOFS);
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00001808 if (!ac)
1809 return -ENOMEM;
Sage Weilbb9c12c2010-10-29 15:37:34 -04001810
Miao Xie7892b5a2012-11-15 08:14:47 +00001811 INIT_WORK(&ac->work, do_async_commit);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001812 ac->root = root;
Josef Bacik7a7eaa42011-04-13 12:54:33 -04001813 ac->newtrans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00001814 if (IS_ERR(ac->newtrans)) {
1815 int err = PTR_ERR(ac->newtrans);
1816 kfree(ac);
1817 return err;
1818 }
Sage Weilbb9c12c2010-10-29 15:37:34 -04001819
1820 /* take transaction reference */
Sage Weilbb9c12c2010-10-29 15:37:34 -04001821 cur_trans = trans->transaction;
Josef Bacik13c5a932011-04-11 15:45:29 -04001822 atomic_inc(&cur_trans->use_count);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001823
1824 btrfs_end_transaction(trans, root);
Sage Weil6fc4e35482012-08-30 16:26:15 -06001825
1826 /*
1827 * Tell lockdep we've released the freeze rwsem, since the
1828 * async commit thread will be the one to unlock it.
1829 */
Liu Bob1a06a42013-11-06 16:57:55 +08001830 if (ac->newtrans->type & __TRANS_FREEZABLE)
Oleg Nesterovbee91822015-07-19 23:48:20 +02001831 __sb_writers_release(root->fs_info->sb, SB_FREEZE_FS);
Sage Weil6fc4e35482012-08-30 16:26:15 -06001832
Miao Xie7892b5a2012-11-15 08:14:47 +00001833 schedule_work(&ac->work);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001834
1835 /* wait for transaction to start and unblock */
Sage Weilbb9c12c2010-10-29 15:37:34 -04001836 if (wait_for_unblock)
1837 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1838 else
1839 wait_current_trans_commit_start(root, cur_trans);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001840
Sage Weil38e88052011-06-10 18:43:13 +00001841 if (current->journal_info == trans)
1842 current->journal_info = NULL;
1843
Josef Bacik724e2312013-09-30 11:36:38 -04001844 btrfs_put_transaction(cur_trans);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001845 return 0;
1846}
1847
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001848
1849static void cleanup_transaction(struct btrfs_trans_handle *trans,
Josef Bacik7b8b92a2012-05-31 15:52:43 -04001850 struct btrfs_root *root, int err)
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001851{
1852 struct btrfs_transaction *cur_trans = trans->transaction;
Liu Bof094ac32013-02-27 13:28:25 +00001853 DEFINE_WAIT(wait);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001854
1855 WARN_ON(trans->use_count > 1);
1856
Josef Bacik7b8b92a2012-05-31 15:52:43 -04001857 btrfs_abort_transaction(trans, root, err);
1858
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001859 spin_lock(&root->fs_info->trans_lock);
Liu Bo66b61352013-03-04 16:25:41 +00001860
Miao Xie25d8c282013-05-15 07:48:26 +00001861 /*
1862 * If the transaction is removed from the list, it means this
1863 * transaction has been committed successfully, so it is impossible
1864 * to call the cleanup function.
1865 */
1866 BUG_ON(list_empty(&cur_trans->list));
Liu Bo66b61352013-03-04 16:25:41 +00001867
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001868 list_del_init(&cur_trans->list);
Josef Bacikd7096fc2012-05-31 15:49:57 -04001869 if (cur_trans == root->fs_info->running_transaction) {
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001870 cur_trans->state = TRANS_STATE_COMMIT_DOING;
Liu Bof094ac32013-02-27 13:28:25 +00001871 spin_unlock(&root->fs_info->trans_lock);
1872 wait_event(cur_trans->writer_wait,
1873 atomic_read(&cur_trans->num_writers) == 1);
1874
1875 spin_lock(&root->fs_info->trans_lock);
Josef Bacikd7096fc2012-05-31 15:49:57 -04001876 }
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001877 spin_unlock(&root->fs_info->trans_lock);
1878
1879 btrfs_cleanup_one_transaction(trans->transaction, root);
1880
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001881 spin_lock(&root->fs_info->trans_lock);
1882 if (cur_trans == root->fs_info->running_transaction)
1883 root->fs_info->running_transaction = NULL;
1884 spin_unlock(&root->fs_info->trans_lock);
1885
Josef Bacike0228282013-09-20 22:26:29 -04001886 if (trans->type & __TRANS_FREEZABLE)
1887 sb_end_intwrite(root->fs_info->sb);
Josef Bacik724e2312013-09-30 11:36:38 -04001888 btrfs_put_transaction(cur_trans);
1889 btrfs_put_transaction(cur_trans);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001890
1891 trace_btrfs_transaction_commit(root);
1892
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001893 if (current->journal_info == trans)
1894 current->journal_info = NULL;
Wang Shilongc0af8f02014-02-19 19:24:18 +08001895 btrfs_scrub_cancel(root->fs_info);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001896
1897 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1898}
1899
Miao Xie82436612013-05-15 07:48:28 +00001900static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
1901{
1902 if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT))
Miao Xie6c255e62014-03-06 13:55:01 +08001903 return btrfs_start_delalloc_roots(fs_info, 1, -1);
Miao Xie82436612013-05-15 07:48:28 +00001904 return 0;
1905}
1906
1907static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
1908{
1909 if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT))
Filipe Manana578def72016-04-26 15:36:38 +01001910 btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1);
Miao Xie82436612013-05-15 07:48:28 +00001911}
1912
Josef Bacik50d9aa92014-11-21 14:52:38 -05001913static inline void
Josef Bacik161c35492015-09-24 16:17:39 -04001914btrfs_wait_pending_ordered(struct btrfs_transaction *cur_trans)
Josef Bacik50d9aa92014-11-21 14:52:38 -05001915{
Josef Bacik161c35492015-09-24 16:17:39 -04001916 wait_event(cur_trans->pending_wait,
1917 atomic_read(&cur_trans->pending_ordered) == 0);
Josef Bacik50d9aa92014-11-21 14:52:38 -05001918}
1919
Chris Mason79154b12007-03-22 15:59:16 -04001920int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1921 struct btrfs_root *root)
1922{
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001923 struct btrfs_transaction *cur_trans = trans->transaction;
Chris Mason8fd17792007-04-19 21:01:03 -04001924 struct btrfs_transaction *prev_trans = NULL;
Filipe Manana656f30d2014-09-26 12:25:56 +01001925 struct btrfs_inode *btree_ino = BTRFS_I(root->fs_info->btree_inode);
Miao Xie25287e02012-10-25 09:31:03 +00001926 int ret;
Chris Mason79154b12007-03-22 15:59:16 -04001927
Miao Xie8d25a082013-01-15 06:27:25 +00001928 /* Stop the commit early if ->aborted is set */
1929 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
Miao Xie25287e02012-10-25 09:31:03 +00001930 ret = cur_trans->aborted;
Josef Bacike4a2bca2013-02-06 16:55:41 -05001931 btrfs_end_transaction(trans, root);
1932 return ret;
Miao Xie25287e02012-10-25 09:31:03 +00001933 }
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001934
Chris Mason56bec292009-03-13 10:10:06 -04001935 /* make a pass through all the delayed refs we have so far
1936 * any runnings procs may add more while we are here
1937 */
1938 ret = btrfs_run_delayed_refs(trans, root, 0);
Josef Bacike4a2bca2013-02-06 16:55:41 -05001939 if (ret) {
1940 btrfs_end_transaction(trans, root);
1941 return ret;
1942 }
Chris Mason56bec292009-03-13 10:10:06 -04001943
Josef Bacik0e721102012-06-26 16:13:18 -04001944 btrfs_trans_release_metadata(trans, root);
1945 trans->block_rsv = NULL;
1946
Chris Masonb7ec40d2009-03-12 20:12:45 -04001947 cur_trans = trans->transaction;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001948
Chris Mason56bec292009-03-13 10:10:06 -04001949 /*
1950 * set the flushing flag so procs in this transaction have to
1951 * start sending their work down.
1952 */
Chris Masonb7ec40d2009-03-12 20:12:45 -04001953 cur_trans->delayed_refs.flushing = 1;
Josef Bacik1be41b72013-06-12 13:56:06 -04001954 smp_wmb();
Chris Mason56bec292009-03-13 10:10:06 -04001955
Josef Bacikea658ba2012-09-11 16:57:25 -04001956 if (!list_empty(&trans->new_bgs))
1957 btrfs_create_pending_block_groups(trans, root);
1958
Chris Masonc3e69d52009-03-13 10:17:05 -04001959 ret = btrfs_run_delayed_refs(trans, root, 0);
Josef Bacike4a2bca2013-02-06 16:55:41 -05001960 if (ret) {
1961 btrfs_end_transaction(trans, root);
1962 return ret;
1963 }
Chris Mason56bec292009-03-13 10:10:06 -04001964
Josef Bacik3204d332015-09-24 10:46:10 -04001965 if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) {
Chris Mason1bbc6212015-04-06 12:46:08 -07001966 int run_it = 0;
1967
1968 /* this mutex is also taken before trying to set
1969 * block groups readonly. We need to make sure
1970 * that nobody has set a block group readonly
1971 * after a extents from that block group have been
1972 * allocated for cache files. btrfs_set_block_group_ro
1973 * will wait for the transaction to commit if it
Josef Bacik3204d332015-09-24 10:46:10 -04001974 * finds BTRFS_TRANS_DIRTY_BG_RUN set.
Chris Mason1bbc6212015-04-06 12:46:08 -07001975 *
Josef Bacik3204d332015-09-24 10:46:10 -04001976 * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure
1977 * only one process starts all the block group IO. It wouldn't
Chris Mason1bbc6212015-04-06 12:46:08 -07001978 * hurt to have more than one go through, but there's no
1979 * real advantage to it either.
1980 */
1981 mutex_lock(&root->fs_info->ro_block_group_mutex);
Josef Bacik3204d332015-09-24 10:46:10 -04001982 if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN,
1983 &cur_trans->flags))
Chris Mason1bbc6212015-04-06 12:46:08 -07001984 run_it = 1;
Chris Mason1bbc6212015-04-06 12:46:08 -07001985 mutex_unlock(&root->fs_info->ro_block_group_mutex);
1986
1987 if (run_it)
1988 ret = btrfs_start_dirty_block_groups(trans, root);
1989 }
1990 if (ret) {
1991 btrfs_end_transaction(trans, root);
1992 return ret;
1993 }
1994
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001995 spin_lock(&root->fs_info->trans_lock);
1996 if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
1997 spin_unlock(&root->fs_info->trans_lock);
Josef Bacik13c5a932011-04-11 15:45:29 -04001998 atomic_inc(&cur_trans->use_count);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001999 ret = btrfs_end_transaction(trans, root);
Chris Masonccd467d2007-06-28 15:57:36 -04002000
Li Zefanb9c83002011-07-14 03:17:14 +00002001 wait_for_commit(root, cur_trans);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04002002
Liu Bob4924a02015-03-06 20:23:44 +08002003 if (unlikely(cur_trans->aborted))
2004 ret = cur_trans->aborted;
2005
Josef Bacik724e2312013-09-30 11:36:38 -04002006 btrfs_put_transaction(cur_trans);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04002007
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002008 return ret;
Chris Mason79154b12007-03-22 15:59:16 -04002009 }
Chris Mason4313b392008-01-03 09:08:48 -05002010
Miao Xie4a9d8bd2013-05-17 03:53:43 +00002011 cur_trans->state = TRANS_STATE_COMMIT_START;
Sage Weilbb9c12c2010-10-29 15:37:34 -04002012 wake_up(&root->fs_info->transaction_blocked_wait);
2013
Chris Masonccd467d2007-06-28 15:57:36 -04002014 if (cur_trans->list.prev != &root->fs_info->trans_list) {
2015 prev_trans = list_entry(cur_trans->list.prev,
2016 struct btrfs_transaction, list);
Miao Xie4a9d8bd2013-05-17 03:53:43 +00002017 if (prev_trans->state != TRANS_STATE_COMPLETED) {
Josef Bacik13c5a932011-04-11 15:45:29 -04002018 atomic_inc(&prev_trans->use_count);
Josef Bacika4abeea2011-04-11 17:25:13 -04002019 spin_unlock(&root->fs_info->trans_lock);
Chris Masonccd467d2007-06-28 15:57:36 -04002020
2021 wait_for_commit(root, prev_trans);
Filipe Manana1f9b8c82015-08-12 11:54:35 +01002022 ret = prev_trans->aborted;
Chris Masonccd467d2007-06-28 15:57:36 -04002023
Josef Bacik724e2312013-09-30 11:36:38 -04002024 btrfs_put_transaction(prev_trans);
Filipe Manana1f9b8c82015-08-12 11:54:35 +01002025 if (ret)
2026 goto cleanup_transaction;
Josef Bacika4abeea2011-04-11 17:25:13 -04002027 } else {
2028 spin_unlock(&root->fs_info->trans_lock);
Chris Masonccd467d2007-06-28 15:57:36 -04002029 }
Josef Bacika4abeea2011-04-11 17:25:13 -04002030 } else {
2031 spin_unlock(&root->fs_info->trans_lock);
Chris Masonccd467d2007-06-28 15:57:36 -04002032 }
Josef Bacik15ee9bc2007-08-10 16:22:09 -04002033
Miao Xie0860adf2013-05-15 07:48:27 +00002034 extwriter_counter_dec(cur_trans, trans->type);
2035
Miao Xie82436612013-05-15 07:48:28 +00002036 ret = btrfs_start_delalloc_flush(root->fs_info);
2037 if (ret)
2038 goto cleanup_transaction;
2039
Chris Mason8d875f92014-08-12 10:47:42 -07002040 ret = btrfs_run_delayed_items(trans, root);
Miao Xie581227d2013-05-15 07:48:30 +00002041 if (ret)
2042 goto cleanup_transaction;
Chris Mason89573b92009-03-12 20:12:45 -04002043
Miao Xie581227d2013-05-15 07:48:30 +00002044 wait_event(cur_trans->writer_wait,
2045 extwriter_counter_read(cur_trans) == 0);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04002046
Miao Xie581227d2013-05-15 07:48:30 +00002047 /* some pending stuffs might be added after the previous flush. */
Chris Mason8d875f92014-08-12 10:47:42 -07002048 ret = btrfs_run_delayed_items(trans, root);
Miao Xieca469632012-11-01 07:33:14 +00002049 if (ret)
2050 goto cleanup_transaction;
2051
Miao Xie82436612013-05-15 07:48:28 +00002052 btrfs_wait_delalloc_flush(root->fs_info);
Wang Shilongcb7ab022013-12-04 21:16:53 +08002053
Josef Bacik161c35492015-09-24 16:17:39 -04002054 btrfs_wait_pending_ordered(cur_trans);
Josef Bacik50d9aa92014-11-21 14:52:38 -05002055
Wang Shilongcb7ab022013-12-04 21:16:53 +08002056 btrfs_scrub_pause(root);
Chris Mason75857172011-06-13 20:00:16 -04002057 /*
Josef Baciked0ca142011-06-14 16:22:15 -04002058 * Ok now we need to make sure to block out any other joins while we
2059 * commit the transaction. We could have started a join before setting
Miao Xie4a9d8bd2013-05-17 03:53:43 +00002060 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
Josef Baciked0ca142011-06-14 16:22:15 -04002061 */
2062 spin_lock(&root->fs_info->trans_lock);
Miao Xie4a9d8bd2013-05-17 03:53:43 +00002063 cur_trans->state = TRANS_STATE_COMMIT_DOING;
Josef Baciked0ca142011-06-14 16:22:15 -04002064 spin_unlock(&root->fs_info->trans_lock);
2065 wait_event(cur_trans->writer_wait,
2066 atomic_read(&cur_trans->num_writers) == 1);
2067
Miao Xie2cba30f2013-01-15 06:29:12 +00002068 /* ->aborted might be set after the previous check, so check it */
2069 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
2070 ret = cur_trans->aborted;
Wang Shilong6cf7f772014-02-19 19:24:16 +08002071 goto scrub_continue;
Miao Xie2cba30f2013-01-15 06:29:12 +00002072 }
Chris Masone038dca2011-06-17 14:16:13 -04002073 /*
Chris Mason75857172011-06-13 20:00:16 -04002074 * the reloc mutex makes sure that we stop
2075 * the balancing code from coming in and moving
2076 * extents around in the middle of the commit
2077 */
2078 mutex_lock(&root->fs_info->reloc_mutex);
2079
Miao Xie42874b32012-09-06 04:03:32 -06002080 /*
2081 * We needn't worry about the delayed items because we will
2082 * deal with them in create_pending_snapshot(), which is the
2083 * core function of the snapshot creation.
2084 */
2085 ret = create_pending_snapshots(trans, root->fs_info);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002086 if (ret) {
2087 mutex_unlock(&root->fs_info->reloc_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08002088 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002089 }
Chris Mason3063d292008-01-08 15:46:30 -05002090
Miao Xie42874b32012-09-06 04:03:32 -06002091 /*
2092 * We insert the dir indexes of the snapshots and update the inode
2093 * of the snapshots' parents after the snapshot creation, so there
2094 * are some delayed items which are not dealt with. Now deal with
2095 * them.
2096 *
2097 * We needn't worry that this operation will corrupt the snapshots,
2098 * because all the tree which are snapshoted will be forced to COW
2099 * the nodes and leaves.
2100 */
2101 ret = btrfs_run_delayed_items(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002102 if (ret) {
2103 mutex_unlock(&root->fs_info->reloc_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08002104 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002105 }
Miao Xie16cdcec2011-04-22 18:12:22 +08002106
Chris Mason56bec292009-03-13 10:10:06 -04002107 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002108 if (ret) {
2109 mutex_unlock(&root->fs_info->reloc_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08002110 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002111 }
Chris Mason56bec292009-03-13 10:10:06 -04002112
Qu Wenruo0ed47922015-04-16 16:55:08 +08002113 /* Reocrd old roots for later qgroup accounting */
2114 ret = btrfs_qgroup_prepare_account_extents(trans, root->fs_info);
2115 if (ret) {
2116 mutex_unlock(&root->fs_info->reloc_mutex);
2117 goto scrub_continue;
2118 }
2119
Chris Masone9993762011-06-17 16:14:09 -04002120 /*
2121 * make sure none of the code above managed to slip in a
2122 * delayed item
2123 */
2124 btrfs_assert_delayed_root_empty(root);
2125
Chris Mason2c90e5d2007-04-02 10:50:19 -04002126 WARN_ON(cur_trans != trans->transaction);
Chris Masondc17ff82008-01-08 15:46:30 -05002127
Chris Masone02119d2008-09-05 16:13:11 -04002128 /* btrfs_commit_tree_roots is responsible for getting the
2129 * various roots consistent with each other. Every pointer
2130 * in the tree of tree roots has to point to the most up to date
2131 * root for every subvolume and other tree. So, we have to keep
2132 * the tree logging code from jumping in and changing any
2133 * of the trees.
2134 *
2135 * At this point in the commit, there can't be any tree-log
2136 * writers, but a little lower down we drop the trans mutex
2137 * and let new people in. By holding the tree_log_mutex
2138 * from now until after the super is written, we avoid races
2139 * with the tree-log code.
2140 */
2141 mutex_lock(&root->fs_info->tree_log_mutex);
Zheng Yan1a40e232008-09-26 10:09:34 -04002142
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002143 ret = commit_fs_roots(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002144 if (ret) {
2145 mutex_unlock(&root->fs_info->tree_log_mutex);
David Sterba871383b2012-04-02 18:31:37 +02002146 mutex_unlock(&root->fs_info->reloc_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08002147 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002148 }
Chris Mason54aa1f42007-06-22 14:16:25 -04002149
Qu Wenruo3818aea2014-01-13 13:36:06 +08002150 /*
David Sterba7e1876a2014-02-05 15:26:17 +01002151 * Since the transaction is done, we can apply the pending changes
2152 * before the next transaction.
Qu Wenruo3818aea2014-01-13 13:36:06 +08002153 */
David Sterba572d9ab2014-02-05 15:26:17 +01002154 btrfs_apply_pending_changes(root->fs_info);
Qu Wenruo3818aea2014-01-13 13:36:06 +08002155
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002156 /* commit_fs_roots gets rid of all the tree log roots, it is now
Chris Masone02119d2008-09-05 16:13:11 -04002157 * safe to free the root of tree log roots
2158 */
2159 btrfs_free_log_root_tree(trans, root->fs_info);
2160
Qu Wenruo0ed47922015-04-16 16:55:08 +08002161 /*
2162 * Since fs roots are all committed, we can get a quite accurate
2163 * new_roots. So let's do quota accounting.
2164 */
2165 ret = btrfs_qgroup_account_extents(trans, root->fs_info);
2166 if (ret < 0) {
2167 mutex_unlock(&root->fs_info->tree_log_mutex);
2168 mutex_unlock(&root->fs_info->reloc_mutex);
2169 goto scrub_continue;
2170 }
2171
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002172 ret = commit_cowonly_roots(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002173 if (ret) {
2174 mutex_unlock(&root->fs_info->tree_log_mutex);
David Sterba871383b2012-04-02 18:31:37 +02002175 mutex_unlock(&root->fs_info->reloc_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08002176 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002177 }
Chris Mason54aa1f42007-06-22 14:16:25 -04002178
Miao Xie2cba30f2013-01-15 06:29:12 +00002179 /*
2180 * The tasks which save the space cache and inode cache may also
2181 * update ->aborted, check it.
2182 */
2183 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
2184 ret = cur_trans->aborted;
2185 mutex_unlock(&root->fs_info->tree_log_mutex);
2186 mutex_unlock(&root->fs_info->reloc_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08002187 goto scrub_continue;
Miao Xie2cba30f2013-01-15 06:29:12 +00002188 }
2189
Yan Zheng11833d62009-09-11 16:11:19 -04002190 btrfs_prepare_extent_commit(trans, root);
2191
Chris Mason78fae272007-03-25 11:35:08 -04002192 cur_trans = root->fs_info->running_transaction;
Chris Mason5f39d392007-10-15 16:14:19 -04002193
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002194 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
2195 root->fs_info->tree_root->node);
Josef Bacik9e351cc2014-03-13 15:42:13 -04002196 list_add_tail(&root->fs_info->tree_root->dirty_list,
2197 &cur_trans->switch_commits);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002198
2199 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
2200 root->fs_info->chunk_root->node);
Josef Bacik9e351cc2014-03-13 15:42:13 -04002201 list_add_tail(&root->fs_info->chunk_root->dirty_list,
2202 &cur_trans->switch_commits);
2203
2204 switch_commit_roots(cur_trans, root->fs_info);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002205
Jan Schmidtedf39272012-06-28 18:04:55 +02002206 assert_qgroups_uptodate(trans);
Josef Bacikce93ec52014-11-17 15:45:48 -05002207 ASSERT(list_empty(&cur_trans->dirty_bgs));
Chris Mason1bbc6212015-04-06 12:46:08 -07002208 ASSERT(list_empty(&cur_trans->io_bgs));
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002209 update_super_roots(root);
Chris Masone02119d2008-09-05 16:13:11 -04002210
Josef Bacik60e7cd32013-09-30 14:10:43 -04002211 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
2212 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
David Sterba6c417612011-04-13 15:41:04 +02002213 memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
2214 sizeof(*root->fs_info->super_copy));
Chris Masonccd467d2007-06-28 15:57:36 -04002215
Miao Xie935e5cc2014-09-03 21:35:33 +08002216 btrfs_update_commit_device_size(root->fs_info);
Miao Xiece7213c2014-09-03 21:35:34 +08002217 btrfs_update_commit_device_bytes_used(root, cur_trans);
Miao Xie935e5cc2014-09-03 21:35:33 +08002218
Filipe Manana656f30d2014-09-26 12:25:56 +01002219 clear_bit(BTRFS_INODE_BTREE_LOG1_ERR, &btree_ino->runtime_flags);
2220 clear_bit(BTRFS_INODE_BTREE_LOG2_ERR, &btree_ino->runtime_flags);
2221
Filipe Manana4fbcdf62015-05-20 14:01:54 +01002222 btrfs_trans_release_chunk_metadata(trans);
2223
Josef Bacika4abeea2011-04-11 17:25:13 -04002224 spin_lock(&root->fs_info->trans_lock);
Miao Xie4a9d8bd2013-05-17 03:53:43 +00002225 cur_trans->state = TRANS_STATE_UNBLOCKED;
Josef Bacika4abeea2011-04-11 17:25:13 -04002226 root->fs_info->running_transaction = NULL;
Josef Bacika4abeea2011-04-11 17:25:13 -04002227 spin_unlock(&root->fs_info->trans_lock);
Chris Mason75857172011-06-13 20:00:16 -04002228 mutex_unlock(&root->fs_info->reloc_mutex);
Chris Masonb7ec40d2009-03-12 20:12:45 -04002229
Chris Masonf9295742008-07-17 12:54:14 -04002230 wake_up(&root->fs_info->transaction_wait);
Chris Masone6dcd2d2008-07-17 12:53:50 -04002231
Chris Mason79154b12007-03-22 15:59:16 -04002232 ret = btrfs_write_and_wait_transaction(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002233 if (ret) {
Anand Jain34d97002016-03-16 16:43:06 +08002234 btrfs_handle_fs_error(root->fs_info, ret,
David Sterba08748812013-03-12 14:46:08 +00002235 "Error while writing out transaction");
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002236 mutex_unlock(&root->fs_info->tree_log_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08002237 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002238 }
2239
2240 ret = write_ctree_super(trans, root, 0);
2241 if (ret) {
2242 mutex_unlock(&root->fs_info->tree_log_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08002243 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002244 }
Chris Mason4313b392008-01-03 09:08:48 -05002245
Chris Masone02119d2008-09-05 16:13:11 -04002246 /*
2247 * the super is written, we can safely allow the tree-loggers
2248 * to go about their business
2249 */
2250 mutex_unlock(&root->fs_info->tree_log_mutex);
2251
Yan Zheng11833d62009-09-11 16:11:19 -04002252 btrfs_finish_extent_commit(trans, root);
Chris Mason4313b392008-01-03 09:08:48 -05002253
Josef Bacik3204d332015-09-24 10:46:10 -04002254 if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags))
Zhao Lei13212b52015-02-12 14:18:17 +08002255 btrfs_clear_space_info_full(root->fs_info);
2256
Josef Bacik15ee9bc2007-08-10 16:22:09 -04002257 root->fs_info->last_trans_committed = cur_trans->transid;
Miao Xie4a9d8bd2013-05-17 03:53:43 +00002258 /*
2259 * We needn't acquire the lock here because there is no other task
2260 * which can change it.
2261 */
2262 cur_trans->state = TRANS_STATE_COMPLETED;
Chris Mason2c90e5d2007-04-02 10:50:19 -04002263 wake_up(&cur_trans->commit_wait);
Chris Mason3de45862008-11-17 21:02:50 -05002264
Josef Bacika4abeea2011-04-11 17:25:13 -04002265 spin_lock(&root->fs_info->trans_lock);
Josef Bacik13c5a932011-04-11 15:45:29 -04002266 list_del_init(&cur_trans->list);
Josef Bacika4abeea2011-04-11 17:25:13 -04002267 spin_unlock(&root->fs_info->trans_lock);
2268
Josef Bacik724e2312013-09-30 11:36:38 -04002269 btrfs_put_transaction(cur_trans);
2270 btrfs_put_transaction(cur_trans);
Josef Bacik58176a92007-08-29 15:47:34 -04002271
Miao Xie0860adf2013-05-15 07:48:27 +00002272 if (trans->type & __TRANS_FREEZABLE)
Miao Xie354aa0f2012-09-20 01:54:00 -06002273 sb_end_intwrite(root->fs_info->sb);
Jan Karab2b5ef52012-06-12 16:20:45 +02002274
liubo1abe9b82011-03-24 11:18:59 +00002275 trace_btrfs_transaction_commit(root);
2276
Arne Jansena2de7332011-03-08 14:14:00 +01002277 btrfs_scrub_continue(root);
2278
Josef Bacik9ed74f22009-09-11 16:12:44 -04002279 if (current->journal_info == trans)
2280 current->journal_info = NULL;
2281
Chris Mason2c90e5d2007-04-02 10:50:19 -04002282 kmem_cache_free(btrfs_trans_handle_cachep, trans);
Yan, Zheng24bbcf02009-11-12 09:36:34 +00002283
Zhao Lei8a733012015-07-15 11:48:14 +08002284 if (current != root->fs_info->transaction_kthread &&
2285 current != root->fs_info->cleaner_kthread)
Yan, Zheng24bbcf02009-11-12 09:36:34 +00002286 btrfs_run_delayed_iputs(root);
2287
Chris Mason79154b12007-03-22 15:59:16 -04002288 return ret;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002289
Wang Shilong6cf7f772014-02-19 19:24:16 +08002290scrub_continue:
2291 btrfs_scrub_continue(root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002292cleanup_transaction:
Josef Bacik0e721102012-06-26 16:13:18 -04002293 btrfs_trans_release_metadata(trans, root);
Filipe Manana4fbcdf62015-05-20 14:01:54 +01002294 btrfs_trans_release_chunk_metadata(trans);
Josef Bacik0e721102012-06-26 16:13:18 -04002295 trans->block_rsv = NULL;
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00002296 btrfs_warn(root->fs_info, "Skipping commit of aborted transaction.");
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002297 if (current->journal_info == trans)
2298 current->journal_info = NULL;
Josef Bacik7b8b92a2012-05-31 15:52:43 -04002299 cleanup_transaction(trans, root, ret);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002300
2301 return ret;
Chris Mason79154b12007-03-22 15:59:16 -04002302}
2303
Chris Masond352ac62008-09-29 15:18:18 -04002304/*
David Sterba9d1a2a32013-03-12 15:13:28 +00002305 * return < 0 if error
2306 * 0 if there are no more dead_roots at the time of call
2307 * 1 there are more to be processed, call me again
2308 *
2309 * The return value indicates there are certainly more snapshots to delete, but
2310 * if there comes a new one during processing, it may return 0. We don't mind,
2311 * because btrfs_commit_super will poke cleaner thread and it will process it a
2312 * few seconds later.
Chris Masond352ac62008-09-29 15:18:18 -04002313 */
David Sterba9d1a2a32013-03-12 15:13:28 +00002314int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root)
Chris Masone9d0b132007-08-10 14:06:19 -04002315{
David Sterba9d1a2a32013-03-12 15:13:28 +00002316 int ret;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002317 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone9d0b132007-08-10 14:06:19 -04002318
Josef Bacika4abeea2011-04-11 17:25:13 -04002319 spin_lock(&fs_info->trans_lock);
David Sterba9d1a2a32013-03-12 15:13:28 +00002320 if (list_empty(&fs_info->dead_roots)) {
2321 spin_unlock(&fs_info->trans_lock);
2322 return 0;
2323 }
2324 root = list_first_entry(&fs_info->dead_roots,
2325 struct btrfs_root, root_list);
Josef Bacikcfad3922013-07-25 15:11:47 -04002326 list_del_init(&root->root_list);
Josef Bacika4abeea2011-04-11 17:25:13 -04002327 spin_unlock(&fs_info->trans_lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002328
Frank Holtonefe120a2013-12-20 11:37:06 -05002329 pr_debug("BTRFS: cleaner removing %llu\n", root->objectid);
Jeff Mahoney2c536792011-10-03 23:22:41 -04002330
David Sterba9d1a2a32013-03-12 15:13:28 +00002331 btrfs_kill_all_delayed_nodes(root);
Yan, Zheng76dda932009-09-21 16:00:26 -04002332
David Sterba9d1a2a32013-03-12 15:13:28 +00002333 if (btrfs_header_backref_rev(root->node) <
2334 BTRFS_MIXED_BACKREF_REV)
2335 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
2336 else
2337 ret = btrfs_drop_snapshot(root, NULL, 1, 0);
David Sterba32471dc2014-02-05 02:03:47 +01002338
Josef Bacik6596a9282013-07-31 10:28:05 -04002339 return (ret < 0) ? 0 : 1;
Chris Masone9d0b132007-08-10 14:06:19 -04002340}
David Sterba572d9ab2014-02-05 15:26:17 +01002341
2342void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info)
2343{
2344 unsigned long prev;
2345 unsigned long bit;
2346
Qu Wenruo6c9fe142015-01-20 17:05:33 +08002347 prev = xchg(&fs_info->pending_changes, 0);
David Sterba572d9ab2014-02-05 15:26:17 +01002348 if (!prev)
2349 return;
2350
David Sterba7e1876a2014-02-05 15:26:17 +01002351 bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE;
2352 if (prev & bit)
2353 btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2354 prev &= ~bit;
2355
2356 bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE;
2357 if (prev & bit)
2358 btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2359 prev &= ~bit;
2360
David Sterbad51033d2014-11-12 14:24:35 +01002361 bit = 1 << BTRFS_PENDING_COMMIT;
2362 if (prev & bit)
2363 btrfs_debug(fs_info, "pending commit done");
2364 prev &= ~bit;
2365
David Sterba572d9ab2014-02-05 15:26:17 +01002366 if (prev)
2367 btrfs_warn(fs_info,
2368 "unknown pending changes left 0x%lx, ignoring", prev);
2369}