blob: b53104042e954f493aee84965072567a040a6d47 [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 Bacikafcdd122016-09-02 15:40:02 -0400444 if (test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags))
Josef Bacika4abeea2011-04-11 17:25:13 -0400445 return 0;
446
447 if (type == TRANS_USERSPACE)
Yan, Zhenga22285a2010-05-16 10:48:46 -0400448 return 1;
Josef Bacika4abeea2011-04-11 17:25:13 -0400449
450 if (type == TRANS_START &&
451 !atomic_read(&root->fs_info->open_ioctl_trans))
452 return 1;
453
Yan, Zhenga22285a2010-05-16 10:48:46 -0400454 return 0;
455}
456
Miao Xie20dd2cb2013-09-25 21:47:45 +0800457static inline bool need_reserve_reloc_root(struct btrfs_root *root)
458{
459 if (!root->fs_info->reloc_ctl ||
Miao Xie27cdeb72014-04-02 19:51:05 +0800460 !test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
Miao Xie20dd2cb2013-09-25 21:47:45 +0800461 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
462 root->reloc_root)
463 return false;
464
465 return true;
466}
467
Miao Xie08e007d2012-10-16 11:33:38 +0000468static struct btrfs_trans_handle *
Alexandru Moise5aed1dd2015-09-22 20:59:15 +0000469start_transaction(struct btrfs_root *root, unsigned int num_items,
470 unsigned int type, enum btrfs_reserve_flush_enum flush)
Yan, Zhenga22285a2010-05-16 10:48:46 -0400471{
472 struct btrfs_trans_handle *h;
473 struct btrfs_transaction *cur_trans;
Josef Bacikb5009942011-06-07 15:07:51 -0400474 u64 num_bytes = 0;
Arne Jansenc5567232011-09-14 15:44:05 +0200475 u64 qgroup_reserved = 0;
Miao Xie20dd2cb2013-09-25 21:47:45 +0800476 bool reloc_reserved = false;
477 int ret;
liuboacce9522011-01-06 19:30:25 +0800478
Filipe Manana46c4e712014-06-24 17:48:28 +0100479 /* Send isn't supposed to start transactions. */
David Sterba2755a0d2014-07-31 00:43:18 +0200480 ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB);
Filipe Manana46c4e712014-06-24 17:48:28 +0100481
Miao Xie87533c42013-01-29 10:14:48 +0000482 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
liuboacce9522011-01-06 19:30:25 +0800483 return ERR_PTR(-EROFS);
Josef Bacik2a1eb462011-04-13 15:15:59 -0400484
Filipe Manana46c4e712014-06-24 17:48:28 +0100485 if (current->journal_info) {
Miao Xie0860adf2013-05-15 07:48:27 +0000486 WARN_ON(type & TRANS_EXTWRITERS);
Josef Bacik2a1eb462011-04-13 15:15:59 -0400487 h = current->journal_info;
488 h->use_count++;
Miao Xieb7d5b0a2012-11-01 07:32:18 +0000489 WARN_ON(h->use_count > 2);
Josef Bacik2a1eb462011-04-13 15:15:59 -0400490 h->orig_rsv = h->block_rsv;
491 h->block_rsv = NULL;
492 goto got_it;
493 }
Josef Bacikb5009942011-06-07 15:07:51 -0400494
495 /*
496 * Do the reservation before we join the transaction so we can do all
497 * the appropriate flushing if need be.
498 */
499 if (num_items > 0 && root != root->fs_info->chunk_root) {
Qu Wenruo71741092015-09-08 17:22:41 +0800500 qgroup_reserved = num_items * root->nodesize;
501 ret = btrfs_qgroup_reserve_meta(root, qgroup_reserved);
502 if (ret)
503 return ERR_PTR(ret);
Arne Jansenc5567232011-09-14 15:44:05 +0200504
Josef Bacikb5009942011-06-07 15:07:51 -0400505 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
Miao Xie20dd2cb2013-09-25 21:47:45 +0800506 /*
507 * Do the reservation for the relocation root creation
508 */
David Sterbaee39b432014-09-30 01:33:33 +0200509 if (need_reserve_reloc_root(root)) {
Miao Xie20dd2cb2013-09-25 21:47:45 +0800510 num_bytes += root->nodesize;
511 reloc_reserved = true;
512 }
513
Miao Xie08e007d2012-10-16 11:33:38 +0000514 ret = btrfs_block_rsv_add(root,
515 &root->fs_info->trans_block_rsv,
516 num_bytes, flush);
Josef Bacikb5009942011-06-07 15:07:51 -0400517 if (ret)
Miao Xie843fcf32013-01-28 12:36:22 +0000518 goto reserve_fail;
Josef Bacikb5009942011-06-07 15:07:51 -0400519 }
Yan, Zhenga22285a2010-05-16 10:48:46 -0400520again:
Alexandru Moisef2f767e2015-08-27 23:53:45 +0000521 h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS);
Miao Xie843fcf32013-01-28 12:36:22 +0000522 if (!h) {
523 ret = -ENOMEM;
524 goto alloc_fail;
525 }
Yan, Zhenga22285a2010-05-16 10:48:46 -0400526
Josef Bacik98114652012-09-14 11:22:38 -0400527 /*
528 * If we are JOIN_NOLOCK we're already committing a transaction and
529 * waiting on this guy, so we don't need to do the sb_start_intwrite
530 * because we're already holding a ref. We need this because we could
531 * have raced in and did an fsync() on a file which can kick a commit
532 * and then we deadlock with somebody doing a freeze.
Miao Xie354aa0f2012-09-20 01:54:00 -0600533 *
534 * If we are ATTACH, it means we just want to catch the current
535 * transaction and commit it, so we needn't do sb_start_intwrite().
Josef Bacik98114652012-09-14 11:22:38 -0400536 */
Miao Xie0860adf2013-05-15 07:48:27 +0000537 if (type & __TRANS_FREEZABLE)
Josef Bacik60376ce2012-09-14 10:34:40 -0400538 sb_start_intwrite(root->fs_info->sb);
Jan Karab2b5ef52012-06-12 16:20:45 +0200539
Yan, Zhenga22285a2010-05-16 10:48:46 -0400540 if (may_wait_transaction(root, type))
Chris Mason37d1aee2008-07-31 10:48:37 -0400541 wait_current_trans(root);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400542
Josef Bacika4abeea2011-04-11 17:25:13 -0400543 do {
Miao Xie354aa0f2012-09-20 01:54:00 -0600544 ret = join_transaction(root, type);
Miao Xie178260b2013-02-20 09:16:24 +0000545 if (ret == -EBUSY) {
Josef Bacika4abeea2011-04-11 17:25:13 -0400546 wait_current_trans(root);
Miao Xie178260b2013-02-20 09:16:24 +0000547 if (unlikely(type == TRANS_ATTACH))
548 ret = -ENOENT;
549 }
Josef Bacika4abeea2011-04-11 17:25:13 -0400550 } while (ret == -EBUSY);
551
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;
Jeff Mahoney64b63582016-06-20 17:23:41 -0400564 h->fs_info = root->fs_info;
Qu Wenruo71741092015-09-08 17:22:41 +0800565
Miao Xiea698d0752012-09-20 01:51:59 -0600566 h->type = type;
Filipe Mananad9a05402015-10-03 13:13:13 +0100567 h->can_flush_pending_bgs = true;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200568 INIT_LIST_HEAD(&h->qgroup_ref_list);
Josef Bacikea658ba2012-09-11 16:57:25 -0400569 INIT_LIST_HEAD(&h->new_bgs);
Chris Masonb7ec40d2009-03-12 20:12:45 -0400570
Yan, Zhenga22285a2010-05-16 10:48:46 -0400571 smp_mb();
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000572 if (cur_trans->state >= TRANS_STATE_BLOCKED &&
573 may_wait_transaction(root, type)) {
Filipe Mananaabdd2e82014-06-24 17:46:58 +0100574 current->journal_info = h;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400575 btrfs_commit_transaction(h, root);
576 goto again;
577 }
Josef Bacik9ed74f22009-09-11 16:12:44 -0400578
Josef Bacikb5009942011-06-07 15:07:51 -0400579 if (num_bytes) {
Josef Bacik8c2a3ca2012-01-10 10:31:31 -0500580 trace_btrfs_space_reservation(root->fs_info, "transaction",
Liu Bo2bcc0322012-03-29 09:57:44 -0400581 h->transid, num_bytes, 1);
Josef Bacikb5009942011-06-07 15:07:51 -0400582 h->block_rsv = &root->fs_info->trans_block_rsv;
583 h->bytes_reserved = num_bytes;
Miao Xie20dd2cb2013-09-25 21:47:45 +0800584 h->reloc_reserved = reloc_reserved;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400585 }
586
Josef Bacik2a1eb462011-04-13 15:15:59 -0400587got_it:
Josef Bacika4abeea2011-04-11 17:25:13 -0400588 btrfs_record_root_in_trans(h, root);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400589
590 if (!current->journal_info && type != TRANS_USERSPACE)
591 current->journal_info = h;
Chris Mason79154b12007-03-22 15:59:16 -0400592 return h;
Miao Xie843fcf32013-01-28 12:36:22 +0000593
594join_fail:
Miao Xie0860adf2013-05-15 07:48:27 +0000595 if (type & __TRANS_FREEZABLE)
Miao Xie843fcf32013-01-28 12:36:22 +0000596 sb_end_intwrite(root->fs_info->sb);
597 kmem_cache_free(btrfs_trans_handle_cachep, h);
598alloc_fail:
599 if (num_bytes)
600 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
601 num_bytes);
602reserve_fail:
Qu Wenruo71741092015-09-08 17:22:41 +0800603 btrfs_qgroup_free_meta(root, qgroup_reserved);
Miao Xie843fcf32013-01-28 12:36:22 +0000604 return ERR_PTR(ret);
Chris Mason79154b12007-03-22 15:59:16 -0400605}
606
Chris Masonf9295742008-07-17 12:54:14 -0400607struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
Alexandru Moise5aed1dd2015-09-22 20:59:15 +0000608 unsigned int num_items)
Chris Masonf9295742008-07-17 12:54:14 -0400609{
Miao Xie08e007d2012-10-16 11:33:38 +0000610 return start_transaction(root, num_items, TRANS_START,
611 BTRFS_RESERVE_FLUSH_ALL);
Chris Masonf9295742008-07-17 12:54:14 -0400612}
Filipe Manana8eab77f2015-11-13 23:57:16 +0000613struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv(
614 struct btrfs_root *root,
615 unsigned int num_items,
616 int min_factor)
617{
618 struct btrfs_trans_handle *trans;
619 u64 num_bytes;
620 int ret;
621
622 trans = btrfs_start_transaction(root, num_items);
623 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
624 return trans;
625
626 trans = btrfs_start_transaction(root, 0);
627 if (IS_ERR(trans))
628 return trans;
629
630 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
631 ret = btrfs_cond_migrate_bytes(root->fs_info,
632 &root->fs_info->trans_block_rsv,
633 num_bytes,
634 min_factor);
635 if (ret) {
636 btrfs_end_transaction(trans, root);
637 return ERR_PTR(ret);
638 }
639
640 trans->block_rsv = &root->fs_info->trans_block_rsv;
641 trans->bytes_reserved = num_bytes;
Josef Bacik88d3a5a2016-01-13 13:21:20 -0500642 trace_btrfs_space_reservation(root->fs_info, "transaction",
643 trans->transid, num_bytes, 1);
Filipe Manana8eab77f2015-11-13 23:57:16 +0000644
645 return trans;
646}
Miao Xie8407aa42012-09-07 01:43:32 -0600647
Miao Xie08e007d2012-10-16 11:33:38 +0000648struct btrfs_trans_handle *btrfs_start_transaction_lflush(
Alexandru Moise5aed1dd2015-09-22 20:59:15 +0000649 struct btrfs_root *root,
650 unsigned int num_items)
Miao Xie8407aa42012-09-07 01:43:32 -0600651{
Miao Xie08e007d2012-10-16 11:33:38 +0000652 return start_transaction(root, num_items, TRANS_START,
653 BTRFS_RESERVE_FLUSH_LIMIT);
Miao Xie8407aa42012-09-07 01:43:32 -0600654}
655
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400656struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
Chris Masonf9295742008-07-17 12:54:14 -0400657{
Alexandru Moise575a75d2015-10-25 19:35:44 +0000658 return start_transaction(root, 0, TRANS_JOIN,
659 BTRFS_RESERVE_NO_FLUSH);
Chris Masonf9295742008-07-17 12:54:14 -0400660}
661
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400662struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
Josef Bacik0af3d002010-06-21 14:48:16 -0400663{
Alexandru Moise575a75d2015-10-25 19:35:44 +0000664 return start_transaction(root, 0, TRANS_JOIN_NOLOCK,
665 BTRFS_RESERVE_NO_FLUSH);
Josef Bacik0af3d002010-06-21 14:48:16 -0400666}
667
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400668struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
Sage Weil9ca9ee02008-08-04 10:41:27 -0400669{
Alexandru Moise575a75d2015-10-25 19:35:44 +0000670 return start_transaction(root, 0, TRANS_USERSPACE,
671 BTRFS_RESERVE_NO_FLUSH);
Sage Weil9ca9ee02008-08-04 10:41:27 -0400672}
673
Miao Xied4edf392013-02-20 09:17:06 +0000674/*
675 * btrfs_attach_transaction() - catch the running transaction
676 *
677 * It is used when we want to commit the current the transaction, but
678 * don't want to start a new one.
679 *
680 * Note: If this function return -ENOENT, it just means there is no
681 * running transaction. But it is possible that the inactive transaction
682 * is still in the memory, not fully on disk. If you hope there is no
683 * inactive transaction in the fs when -ENOENT is returned, you should
684 * invoke
685 * btrfs_attach_transaction_barrier()
686 */
Miao Xie354aa0f2012-09-20 01:54:00 -0600687struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
Josef Bacik60376ce2012-09-14 10:34:40 -0400688{
Alexandru Moise575a75d2015-10-25 19:35:44 +0000689 return start_transaction(root, 0, TRANS_ATTACH,
690 BTRFS_RESERVE_NO_FLUSH);
Josef Bacik60376ce2012-09-14 10:34:40 -0400691}
692
Miao Xied4edf392013-02-20 09:17:06 +0000693/*
Wang Sheng-Hui90b6d282013-06-14 16:21:24 +0800694 * btrfs_attach_transaction_barrier() - catch the running transaction
Miao Xied4edf392013-02-20 09:17:06 +0000695 *
696 * It is similar to the above function, the differentia is this one
697 * will wait for all the inactive transactions until they fully
698 * complete.
699 */
700struct btrfs_trans_handle *
701btrfs_attach_transaction_barrier(struct btrfs_root *root)
702{
703 struct btrfs_trans_handle *trans;
704
Alexandru Moise575a75d2015-10-25 19:35:44 +0000705 trans = start_transaction(root, 0, TRANS_ATTACH,
706 BTRFS_RESERVE_NO_FLUSH);
Miao Xied4edf392013-02-20 09:17:06 +0000707 if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT)
708 btrfs_wait_for_commit(root, 0);
709
710 return trans;
711}
712
Chris Masond352ac62008-09-29 15:18:18 -0400713/* wait for a transaction commit to be fully complete */
Li Zefanb9c83002011-07-14 03:17:14 +0000714static noinline void wait_for_commit(struct btrfs_root *root,
Chris Mason89ce8a62008-06-25 16:01:31 -0400715 struct btrfs_transaction *commit)
716{
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000717 wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED);
Chris Mason89ce8a62008-06-25 16:01:31 -0400718}
719
Sage Weil46204592010-10-29 15:41:32 -0400720int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
721{
722 struct btrfs_transaction *cur_trans = NULL, *t;
Miao Xie8cd28072012-11-26 08:42:07 +0000723 int ret = 0;
Sage Weil46204592010-10-29 15:41:32 -0400724
Sage Weil46204592010-10-29 15:41:32 -0400725 if (transid) {
726 if (transid <= root->fs_info->last_trans_committed)
Josef Bacika4abeea2011-04-11 17:25:13 -0400727 goto out;
Sage Weil46204592010-10-29 15:41:32 -0400728
729 /* find specified transaction */
Josef Bacika4abeea2011-04-11 17:25:13 -0400730 spin_lock(&root->fs_info->trans_lock);
Sage Weil46204592010-10-29 15:41:32 -0400731 list_for_each_entry(t, &root->fs_info->trans_list, list) {
732 if (t->transid == transid) {
733 cur_trans = t;
Josef Bacika4abeea2011-04-11 17:25:13 -0400734 atomic_inc(&cur_trans->use_count);
Miao Xie8cd28072012-11-26 08:42:07 +0000735 ret = 0;
Sage Weil46204592010-10-29 15:41:32 -0400736 break;
737 }
Miao Xie8cd28072012-11-26 08:42:07 +0000738 if (t->transid > transid) {
739 ret = 0;
Sage Weil46204592010-10-29 15:41:32 -0400740 break;
Miao Xie8cd28072012-11-26 08:42:07 +0000741 }
Sage Weil46204592010-10-29 15:41:32 -0400742 }
Josef Bacika4abeea2011-04-11 17:25:13 -0400743 spin_unlock(&root->fs_info->trans_lock);
Sage Weil42383022014-09-26 08:30:06 -0700744
745 /*
746 * The specified transaction doesn't exist, or we
747 * raced with btrfs_commit_transaction
748 */
749 if (!cur_trans) {
750 if (transid > root->fs_info->last_trans_committed)
751 ret = -EINVAL;
Miao Xie8cd28072012-11-26 08:42:07 +0000752 goto out;
Sage Weil42383022014-09-26 08:30:06 -0700753 }
Sage Weil46204592010-10-29 15:41:32 -0400754 } else {
755 /* find newest transaction that is committing | committed */
Josef Bacika4abeea2011-04-11 17:25:13 -0400756 spin_lock(&root->fs_info->trans_lock);
Sage Weil46204592010-10-29 15:41:32 -0400757 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
758 list) {
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000759 if (t->state >= TRANS_STATE_COMMIT_START) {
760 if (t->state == TRANS_STATE_COMPLETED)
Josef Bacik3473f3c2011-06-09 10:15:17 -0400761 break;
Sage Weil46204592010-10-29 15:41:32 -0400762 cur_trans = t;
Josef Bacika4abeea2011-04-11 17:25:13 -0400763 atomic_inc(&cur_trans->use_count);
Sage Weil46204592010-10-29 15:41:32 -0400764 break;
765 }
766 }
Josef Bacika4abeea2011-04-11 17:25:13 -0400767 spin_unlock(&root->fs_info->trans_lock);
Sage Weil46204592010-10-29 15:41:32 -0400768 if (!cur_trans)
Josef Bacika4abeea2011-04-11 17:25:13 -0400769 goto out; /* nothing committing|committed */
Sage Weil46204592010-10-29 15:41:32 -0400770 }
771
Sage Weil46204592010-10-29 15:41:32 -0400772 wait_for_commit(root, cur_trans);
Josef Bacik724e2312013-09-30 11:36:38 -0400773 btrfs_put_transaction(cur_trans);
Josef Bacika4abeea2011-04-11 17:25:13 -0400774out:
Sage Weil46204592010-10-29 15:41:32 -0400775 return ret;
776}
777
Chris Mason37d1aee2008-07-31 10:48:37 -0400778void btrfs_throttle(struct btrfs_root *root)
779{
Josef Bacika4abeea2011-04-11 17:25:13 -0400780 if (!atomic_read(&root->fs_info->open_ioctl_trans))
Sage Weil9ca9ee02008-08-04 10:41:27 -0400781 wait_current_trans(root);
Chris Mason37d1aee2008-07-31 10:48:37 -0400782}
783
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400784static int should_end_transaction(struct btrfs_trans_handle *trans,
785 struct btrfs_root *root)
786{
Josef Bacik1be41b72013-06-12 13:56:06 -0400787 if (root->fs_info->global_block_rsv.space_info->full &&
Josef Bacik0a2b2a82014-01-23 10:54:11 -0500788 btrfs_check_space_for_delayed_refs(trans, root))
Josef Bacik1be41b72013-06-12 13:56:06 -0400789 return 1;
Josef Bacik36ba0222011-10-18 12:15:48 -0400790
Josef Bacik1be41b72013-06-12 13:56:06 -0400791 return !!btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400792}
793
794int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
795 struct btrfs_root *root)
796{
797 struct btrfs_transaction *cur_trans = trans->transaction;
798 int updates;
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100799 int err;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400800
Josef Bacika4abeea2011-04-11 17:25:13 -0400801 smp_mb();
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000802 if (cur_trans->state >= TRANS_STATE_BLOCKED ||
803 cur_trans->delayed_refs.flushing)
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400804 return 1;
805
806 updates = trans->delayed_ref_updates;
807 trans->delayed_ref_updates = 0;
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100808 if (updates) {
Chris Mason28ed1342014-12-17 09:41:04 -0800809 err = btrfs_run_delayed_refs(trans, root, updates * 2);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100810 if (err) /* Error code will also eval true */
811 return err;
812 }
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400813
814 return should_end_transaction(trans, root);
815}
816
Chris Mason89ce8a62008-06-25 16:01:31 -0400817static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
Miao Xiea698d0752012-09-20 01:51:59 -0600818 struct btrfs_root *root, int throttle)
Chris Mason79154b12007-03-22 15:59:16 -0400819{
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400820 struct btrfs_transaction *cur_trans = trans->transaction;
Chris Masonab78c842008-07-29 16:15:18 -0400821 struct btrfs_fs_info *info = root->fs_info;
Josef Bacik31b96552016-04-11 17:37:40 -0400822 u64 transid = trans->transid;
Josef Bacik1be41b72013-06-12 13:56:06 -0400823 unsigned long cur = trans->delayed_ref_updates;
Miao Xiea698d0752012-09-20 01:51:59 -0600824 int lock = (trans->type != TRANS_JOIN_NOLOCK);
Dave Jones4edc2ca2012-04-12 16:03:56 -0400825 int err = 0;
Chris Masona79b7d42014-05-22 16:18:52 -0700826 int must_run_delayed_refs = 0;
Chris Masond6e4a422007-04-06 15:37:36 -0400827
Josef Bacik3bbb24b2014-03-06 19:01:07 -0500828 if (trans->use_count > 1) {
829 trans->use_count--;
Josef Bacik2a1eb462011-04-13 15:15:59 -0400830 trans->block_rsv = trans->orig_rsv;
831 return 0;
832 }
833
Josef Bacikb24e03d2011-10-14 14:40:17 -0400834 btrfs_trans_release_metadata(trans, root);
Josef Bacik4c13d752011-08-30 11:31:29 -0400835 trans->block_rsv = NULL;
Arne Jansenc5567232011-09-14 15:44:05 +0200836
Josef Bacikea658ba2012-09-11 16:57:25 -0400837 if (!list_empty(&trans->new_bgs))
838 btrfs_create_pending_block_groups(trans, root);
839
Josef Bacik1be41b72013-06-12 13:56:06 -0400840 trans->delayed_ref_updates = 0;
Chris Masona79b7d42014-05-22 16:18:52 -0700841 if (!trans->sync) {
842 must_run_delayed_refs =
843 btrfs_should_throttle_delayed_refs(trans, root);
Josef Bacik0a2b2a82014-01-23 10:54:11 -0500844 cur = max_t(unsigned long, cur, 32);
Chris Masona79b7d42014-05-22 16:18:52 -0700845
846 /*
847 * don't make the caller wait if they are from a NOLOCK
848 * or ATTACH transaction, it will deadlock with commit
849 */
850 if (must_run_delayed_refs == 1 &&
851 (trans->type & (__TRANS_JOIN_NOLOCK | __TRANS_ATTACH)))
852 must_run_delayed_refs = 2;
Chris Mason56bec292009-03-13 10:10:06 -0400853 }
Chris Masonbb721702013-01-29 18:44:12 -0500854
Josef Bacik0e721102012-06-26 16:13:18 -0400855 btrfs_trans_release_metadata(trans, root);
856 trans->block_rsv = NULL;
Chris Mason56bec292009-03-13 10:10:06 -0400857
Josef Bacikea658ba2012-09-11 16:57:25 -0400858 if (!list_empty(&trans->new_bgs))
859 btrfs_create_pending_block_groups(trans, root);
860
Filipe Manana4fbcdf62015-05-20 14:01:54 +0100861 btrfs_trans_release_chunk_metadata(trans);
862
Josef Bacika4abeea2011-04-11 17:25:13 -0400863 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000864 should_end_transaction(trans, root) &&
865 ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) {
866 spin_lock(&info->trans_lock);
867 if (cur_trans->state == TRANS_STATE_RUNNING)
868 cur_trans->state = TRANS_STATE_BLOCKED;
869 spin_unlock(&info->trans_lock);
Josef Bacika4abeea2011-04-11 17:25:13 -0400870 }
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400871
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000872 if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) {
Josef Bacik3bbb24b2014-03-06 19:01:07 -0500873 if (throttle)
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400874 return btrfs_commit_transaction(trans, root);
Josef Bacik3bbb24b2014-03-06 19:01:07 -0500875 else
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400876 wake_up_process(info->transaction_kthread);
877 }
878
Miao Xie0860adf2013-05-15 07:48:27 +0000879 if (trans->type & __TRANS_FREEZABLE)
Josef Bacik98114652012-09-14 11:22:38 -0400880 sb_end_intwrite(root->fs_info->sb);
Josef Bacik6df78812012-09-05 08:08:30 -0600881
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400882 WARN_ON(cur_trans != info->running_transaction);
Josef Bacik13c5a932011-04-11 15:45:29 -0400883 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
884 atomic_dec(&cur_trans->num_writers);
Miao Xie0860adf2013-05-15 07:48:27 +0000885 extwriter_counter_dec(cur_trans, trans->type);
Chris Mason89ce8a62008-06-25 16:01:31 -0400886
David Sterbaa83342a2015-02-16 19:36:47 +0100887 /*
888 * Make sure counter is updated before we wake up waiters.
889 */
Sage Weil99d16cb2010-10-29 15:37:34 -0400890 smp_mb();
Chris Mason79154b12007-03-22 15:59:16 -0400891 if (waitqueue_active(&cur_trans->writer_wait))
892 wake_up(&cur_trans->writer_wait);
Josef Bacik724e2312013-09-30 11:36:38 -0400893 btrfs_put_transaction(cur_trans);
Josef Bacik9ed74f22009-09-11 16:12:44 -0400894
895 if (current->journal_info == trans)
896 current->journal_info = NULL;
Chris Masonab78c842008-07-29 16:15:18 -0400897
Yan, Zheng24bbcf02009-11-12 09:36:34 +0000898 if (throttle)
899 btrfs_run_delayed_iputs(root);
900
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100901 if (trans->aborted ||
Josef Bacik4e121c02013-09-27 16:32:39 -0400902 test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
903 wake_up_process(info->transaction_kthread);
Dave Jones4edc2ca2012-04-12 16:03:56 -0400904 err = -EIO;
Josef Bacik4e121c02013-09-27 16:32:39 -0400905 }
Jan Schmidtedf39272012-06-28 18:04:55 +0200906 assert_qgroups_uptodate(trans);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100907
Dave Jones4edc2ca2012-04-12 16:03:56 -0400908 kmem_cache_free(btrfs_trans_handle_cachep, trans);
Chris Masona79b7d42014-05-22 16:18:52 -0700909 if (must_run_delayed_refs) {
Josef Bacik31b96552016-04-11 17:37:40 -0400910 btrfs_async_run_delayed_refs(root, cur, transid,
Chris Masona79b7d42014-05-22 16:18:52 -0700911 must_run_delayed_refs == 1);
912 }
Dave Jones4edc2ca2012-04-12 16:03:56 -0400913 return err;
Chris Mason79154b12007-03-22 15:59:16 -0400914}
915
Chris Mason89ce8a62008-06-25 16:01:31 -0400916int btrfs_end_transaction(struct btrfs_trans_handle *trans,
917 struct btrfs_root *root)
918{
Wang Shilong98ad43be2013-04-14 14:08:49 +0000919 return __btrfs_end_transaction(trans, root, 0);
Chris Mason89ce8a62008-06-25 16:01:31 -0400920}
921
922int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
923 struct btrfs_root *root)
924{
Wang Shilong98ad43be2013-04-14 14:08:49 +0000925 return __btrfs_end_transaction(trans, root, 1);
Miao Xie16cdcec2011-04-22 18:12:22 +0800926}
927
Chris Masond352ac62008-09-29 15:18:18 -0400928/*
929 * when btree blocks are allocated, they have some corresponding bits set for
930 * them in one of two extent_io trees. This is used to make sure all of
Chris Mason690587d2009-10-13 13:29:19 -0400931 * those extents are sent to disk but does not wait on them
Chris Masond352ac62008-09-29 15:18:18 -0400932 */
Chris Mason690587d2009-10-13 13:29:19 -0400933int btrfs_write_marked_extents(struct btrfs_root *root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000934 struct extent_io_tree *dirty_pages, int mark)
Chris Mason79154b12007-03-22 15:59:16 -0400935{
Chris Mason777e6bd2008-08-15 15:34:15 -0400936 int err = 0;
Chris Mason7c4452b2007-04-28 09:29:35 -0400937 int werr = 0;
Josef Bacik17283662011-09-26 13:58:47 -0400938 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
Josef Bacike6138872012-09-27 17:07:30 -0400939 struct extent_state *cached_state = NULL;
Chris Mason777e6bd2008-08-15 15:34:15 -0400940 u64 start = 0;
Chris Mason5f39d392007-10-15 16:14:19 -0400941 u64 end;
Chris Mason7c4452b2007-04-28 09:29:35 -0400942
Josef Bacik17283662011-09-26 13:58:47 -0400943 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
Josef Bacike6138872012-09-27 17:07:30 -0400944 mark, &cached_state)) {
Filipe Manana663dfbb2014-10-13 12:28:37 +0100945 bool wait_writeback = false;
946
947 err = convert_extent_bit(dirty_pages, start, end,
948 EXTENT_NEED_WAIT,
David Sterba210aa272016-04-26 23:54:39 +0200949 mark, &cached_state);
Filipe Manana663dfbb2014-10-13 12:28:37 +0100950 /*
951 * convert_extent_bit can return -ENOMEM, which is most of the
952 * time a temporary error. So when it happens, ignore the error
953 * and wait for writeback of this range to finish - because we
954 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
955 * to btrfs_wait_marked_extents() would not know that writeback
956 * for this range started and therefore wouldn't wait for it to
957 * finish - we don't want to commit a superblock that points to
958 * btree nodes/leafs for which writeback hasn't finished yet
959 * (and without errors).
960 * We cleanup any entries left in the io tree when committing
961 * the transaction (through clear_btree_io_tree()).
962 */
963 if (err == -ENOMEM) {
964 err = 0;
965 wait_writeback = true;
966 }
967 if (!err)
968 err = filemap_fdatawrite_range(mapping, start, end);
Josef Bacik17283662011-09-26 13:58:47 -0400969 if (err)
970 werr = err;
Filipe Manana663dfbb2014-10-13 12:28:37 +0100971 else if (wait_writeback)
972 werr = filemap_fdatawait_range(mapping, start, end);
Filipe Mananae38e2ed2014-10-13 12:28:38 +0100973 free_extent_state(cached_state);
Filipe Manana663dfbb2014-10-13 12:28:37 +0100974 cached_state = NULL;
Josef Bacik17283662011-09-26 13:58:47 -0400975 cond_resched();
976 start = end + 1;
Chris Mason7c4452b2007-04-28 09:29:35 -0400977 }
Chris Mason690587d2009-10-13 13:29:19 -0400978 return werr;
979}
980
981/*
982 * when btree blocks are allocated, they have some corresponding bits set for
983 * them in one of two extent_io trees. This is used to make sure all of
984 * those extents are on disk for transaction or log commit. We wait
985 * on all the pages and clear them from the dirty pages state tree
986 */
987int btrfs_wait_marked_extents(struct btrfs_root *root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000988 struct extent_io_tree *dirty_pages, int mark)
Chris Mason690587d2009-10-13 13:29:19 -0400989{
Chris Mason690587d2009-10-13 13:29:19 -0400990 int err = 0;
991 int werr = 0;
Josef Bacik17283662011-09-26 13:58:47 -0400992 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
Josef Bacike6138872012-09-27 17:07:30 -0400993 struct extent_state *cached_state = NULL;
Chris Mason690587d2009-10-13 13:29:19 -0400994 u64 start = 0;
995 u64 end;
Filipe Manana656f30d2014-09-26 12:25:56 +0100996 bool errors = false;
Chris Mason690587d2009-10-13 13:29:19 -0400997
Josef Bacik17283662011-09-26 13:58:47 -0400998 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
Josef Bacike6138872012-09-27 17:07:30 -0400999 EXTENT_NEED_WAIT, &cached_state)) {
Filipe Manana663dfbb2014-10-13 12:28:37 +01001000 /*
1001 * Ignore -ENOMEM errors returned by clear_extent_bit().
1002 * When committing the transaction, we'll remove any entries
1003 * left in the io tree. For a log commit, we don't remove them
1004 * after committing the log because the tree can be accessed
1005 * concurrently - we do it only at transaction commit time when
1006 * it's safe to do it (through clear_btree_io_tree()).
1007 */
1008 err = clear_extent_bit(dirty_pages, start, end,
1009 EXTENT_NEED_WAIT,
1010 0, 0, &cached_state, GFP_NOFS);
1011 if (err == -ENOMEM)
1012 err = 0;
1013 if (!err)
1014 err = filemap_fdatawait_range(mapping, start, end);
Josef Bacik17283662011-09-26 13:58:47 -04001015 if (err)
1016 werr = err;
Filipe Mananae38e2ed2014-10-13 12:28:38 +01001017 free_extent_state(cached_state);
1018 cached_state = NULL;
Josef Bacik17283662011-09-26 13:58:47 -04001019 cond_resched();
1020 start = end + 1;
Chris Mason777e6bd2008-08-15 15:34:15 -04001021 }
Chris Mason7c4452b2007-04-28 09:29:35 -04001022 if (err)
1023 werr = err;
Filipe Manana656f30d2014-09-26 12:25:56 +01001024
1025 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
1026 if ((mark & EXTENT_DIRTY) &&
Josef Bacikafcdd122016-09-02 15:40:02 -04001027 test_and_clear_bit(BTRFS_FS_LOG1_ERR,
1028 &root->fs_info->flags))
Filipe Manana656f30d2014-09-26 12:25:56 +01001029 errors = true;
1030
1031 if ((mark & EXTENT_NEW) &&
Josef Bacikafcdd122016-09-02 15:40:02 -04001032 test_and_clear_bit(BTRFS_FS_LOG2_ERR,
1033 &root->fs_info->flags))
Filipe Manana656f30d2014-09-26 12:25:56 +01001034 errors = true;
1035 } else {
Josef Bacikafcdd122016-09-02 15:40:02 -04001036 if (test_and_clear_bit(BTRFS_FS_BTREE_ERR,
1037 &root->fs_info->flags))
Filipe Manana656f30d2014-09-26 12:25:56 +01001038 errors = true;
1039 }
1040
1041 if (errors && !werr)
1042 werr = -EIO;
1043
Chris Mason7c4452b2007-04-28 09:29:35 -04001044 return werr;
Chris Mason79154b12007-03-22 15:59:16 -04001045}
1046
Chris Mason690587d2009-10-13 13:29:19 -04001047/*
1048 * when btree blocks are allocated, they have some corresponding bits set for
1049 * them in one of two extent_io trees. This is used to make sure all of
1050 * those extents are on disk for transaction or log commit
1051 */
Sergei Trofimovich171170c2013-08-14 23:27:46 +03001052static int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +00001053 struct extent_io_tree *dirty_pages, int mark)
Chris Mason690587d2009-10-13 13:29:19 -04001054{
1055 int ret;
1056 int ret2;
Miao Xiec6adc9c2013-05-28 10:05:39 +00001057 struct blk_plug plug;
Chris Mason690587d2009-10-13 13:29:19 -04001058
Miao Xiec6adc9c2013-05-28 10:05:39 +00001059 blk_start_plug(&plug);
Yan, Zheng8cef4e12009-11-12 09:33:26 +00001060 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
Miao Xiec6adc9c2013-05-28 10:05:39 +00001061 blk_finish_plug(&plug);
Yan, Zheng8cef4e12009-11-12 09:33:26 +00001062 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
Chris Masonbf0da8c2011-11-04 12:29:37 -04001063
1064 if (ret)
1065 return ret;
1066 if (ret2)
1067 return ret2;
1068 return 0;
Chris Mason690587d2009-10-13 13:29:19 -04001069}
1070
Filipe Manana663dfbb2014-10-13 12:28:37 +01001071static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
Chris Masond0c803c2008-09-11 16:17:57 -04001072 struct btrfs_root *root)
1073{
Filipe Manana663dfbb2014-10-13 12:28:37 +01001074 int ret;
1075
1076 ret = btrfs_write_and_wait_marked_extents(root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +00001077 &trans->transaction->dirty_pages,
1078 EXTENT_DIRTY);
Filipe Manana663dfbb2014-10-13 12:28:37 +01001079 clear_btree_io_tree(&trans->transaction->dirty_pages);
1080
1081 return ret;
Chris Masond0c803c2008-09-11 16:17:57 -04001082}
1083
Chris Masond352ac62008-09-29 15:18:18 -04001084/*
1085 * this is used to update the root pointer in the tree of tree roots.
1086 *
1087 * But, in the case of the extent allocation tree, updating the root
1088 * pointer may allocate blocks which may change the root of the extent
1089 * allocation tree.
1090 *
1091 * So, this loops and repeats and makes sure the cowonly root didn't
1092 * change while the root pointer was being updated in the metadata.
1093 */
Chris Mason0b86a832008-03-24 15:01:56 -04001094static int update_cowonly_root(struct btrfs_trans_handle *trans,
1095 struct btrfs_root *root)
1096{
1097 int ret;
1098 u64 old_root_bytenr;
Yan, Zheng86b9f2e2009-11-12 09:36:50 +00001099 u64 old_root_used;
Chris Mason0b86a832008-03-24 15:01:56 -04001100 struct btrfs_root *tree_root = root->fs_info->tree_root;
1101
Yan, Zheng86b9f2e2009-11-12 09:36:50 +00001102 old_root_used = btrfs_root_used(&root->root_item);
Chris Mason56bec292009-03-13 10:10:06 -04001103
Chris Masond3977122009-01-05 21:25:51 -05001104 while (1) {
Chris Mason0b86a832008-03-24 15:01:56 -04001105 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
Yan, Zheng86b9f2e2009-11-12 09:36:50 +00001106 if (old_root_bytenr == root->node->start &&
Josef Bacikea526d12015-03-13 16:40:45 -04001107 old_root_used == btrfs_root_used(&root->root_item))
Chris Mason0b86a832008-03-24 15:01:56 -04001108 break;
Chris Mason87ef2bb2008-10-30 11:23:27 -04001109
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001110 btrfs_set_root_node(&root->root_item, root->node);
Chris Mason0b86a832008-03-24 15:01:56 -04001111 ret = btrfs_update_root(trans, tree_root,
1112 &root->root_key,
1113 &root->root_item);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001114 if (ret)
1115 return ret;
Chris Mason56bec292009-03-13 10:10:06 -04001116
Yan, Zheng86b9f2e2009-11-12 09:36:50 +00001117 old_root_used = btrfs_root_used(&root->root_item);
Chris Mason0b86a832008-03-24 15:01:56 -04001118 }
Yan Zheng276e6802009-07-30 09:40:40 -04001119
Chris Mason0b86a832008-03-24 15:01:56 -04001120 return 0;
1121}
1122
Chris Masond352ac62008-09-29 15:18:18 -04001123/*
1124 * update all the cowonly tree roots on disk
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001125 *
1126 * The error handling in this function may not be obvious. Any of the
1127 * failures will cause the file system to go offline. We still need
1128 * to clean up the delayed refs.
Chris Masond352ac62008-09-29 15:18:18 -04001129 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001130static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
1131 struct btrfs_root *root)
Chris Mason79154b12007-03-22 15:59:16 -04001132{
Chris Mason79154b12007-03-22 15:59:16 -04001133 struct btrfs_fs_info *fs_info = root->fs_info;
Josef Bacikea526d12015-03-13 16:40:45 -04001134 struct list_head *dirty_bgs = &trans->transaction->dirty_bgs;
Chris Mason1bbc6212015-04-06 12:46:08 -07001135 struct list_head *io_bgs = &trans->transaction->io_bgs;
Chris Mason0b86a832008-03-24 15:01:56 -04001136 struct list_head *next;
Yan Zheng84234f32008-10-29 14:49:05 -04001137 struct extent_buffer *eb;
Chris Mason56bec292009-03-13 10:10:06 -04001138 int ret;
Yan Zheng84234f32008-10-29 14:49:05 -04001139
1140 eb = btrfs_lock_root_node(fs_info->tree_root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001141 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
1142 0, &eb);
Yan Zheng84234f32008-10-29 14:49:05 -04001143 btrfs_tree_unlock(eb);
1144 free_extent_buffer(eb);
Chris Mason79154b12007-03-22 15:59:16 -04001145
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001146 if (ret)
1147 return ret;
1148
Chris Mason56bec292009-03-13 10:10:06 -04001149 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001150 if (ret)
1151 return ret;
Chris Mason87ef2bb2008-10-30 11:23:27 -04001152
Stefan Behrens733f4fb2012-05-25 16:06:10 +02001153 ret = btrfs_run_dev_stats(trans, root->fs_info);
Josef Bacikc16ce192013-09-27 16:38:20 -04001154 if (ret)
1155 return ret;
Stefan Behrens8dabb742012-11-06 13:15:27 +01001156 ret = btrfs_run_dev_replace(trans, root->fs_info);
Josef Bacikc16ce192013-09-27 16:38:20 -04001157 if (ret)
1158 return ret;
Jan Schmidt546adb02012-06-14 16:37:44 +02001159 ret = btrfs_run_qgroups(trans, root->fs_info);
Josef Bacikc16ce192013-09-27 16:38:20 -04001160 if (ret)
1161 return ret;
Jan Schmidt546adb02012-06-14 16:37:44 +02001162
Josef Bacikdcdf7f62015-03-02 16:37:31 -05001163 ret = btrfs_setup_space_cache(trans, root);
1164 if (ret)
1165 return ret;
1166
Jan Schmidt546adb02012-06-14 16:37:44 +02001167 /* run_qgroups might have added some more refs */
1168 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
Josef Bacikc16ce192013-09-27 16:38:20 -04001169 if (ret)
1170 return ret;
Josef Bacikea526d12015-03-13 16:40:45 -04001171again:
Chris Masond3977122009-01-05 21:25:51 -05001172 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
Chris Mason0b86a832008-03-24 15:01:56 -04001173 next = fs_info->dirty_cowonly_roots.next;
1174 list_del_init(next);
1175 root = list_entry(next, struct btrfs_root, dirty_list);
Josef Bacike7070be2014-12-16 08:54:43 -08001176 clear_bit(BTRFS_ROOT_DIRTY, &root->state);
Chris Mason87ef2bb2008-10-30 11:23:27 -04001177
Josef Bacik9e351cc2014-03-13 15:42:13 -04001178 if (root != fs_info->extent_root)
1179 list_add_tail(&root->dirty_list,
1180 &trans->transaction->switch_commits);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001181 ret = update_cowonly_root(trans, root);
1182 if (ret)
1183 return ret;
Josef Bacikea526d12015-03-13 16:40:45 -04001184 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1185 if (ret)
1186 return ret;
Chris Mason79154b12007-03-22 15:59:16 -04001187 }
Yan Zheng276e6802009-07-30 09:40:40 -04001188
Chris Mason1bbc6212015-04-06 12:46:08 -07001189 while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) {
Josef Bacikea526d12015-03-13 16:40:45 -04001190 ret = btrfs_write_dirty_block_groups(trans, root);
1191 if (ret)
1192 return ret;
1193 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1194 if (ret)
1195 return ret;
1196 }
1197
1198 if (!list_empty(&fs_info->dirty_cowonly_roots))
1199 goto again;
1200
Josef Bacik9e351cc2014-03-13 15:42:13 -04001201 list_add_tail(&fs_info->extent_root->dirty_list,
1202 &trans->transaction->switch_commits);
Stefan Behrens8dabb742012-11-06 13:15:27 +01001203 btrfs_after_dev_replace_commit(fs_info);
1204
Chris Mason79154b12007-03-22 15:59:16 -04001205 return 0;
1206}
1207
Chris Masond352ac62008-09-29 15:18:18 -04001208/*
1209 * dead roots are old snapshots that need to be deleted. This allocates
1210 * a dirty root struct and adds it into the list of dead roots that need to
1211 * be deleted
1212 */
Josef Bacikcfad3922013-07-25 15:11:47 -04001213void btrfs_add_dead_root(struct btrfs_root *root)
Chris Mason5eda7b52007-06-22 14:16:25 -04001214{
Josef Bacika4abeea2011-04-11 17:25:13 -04001215 spin_lock(&root->fs_info->trans_lock);
Josef Bacikcfad3922013-07-25 15:11:47 -04001216 if (list_empty(&root->root_list))
1217 list_add_tail(&root->root_list, &root->fs_info->dead_roots);
Josef Bacika4abeea2011-04-11 17:25:13 -04001218 spin_unlock(&root->fs_info->trans_lock);
Chris Mason5eda7b52007-06-22 14:16:25 -04001219}
1220
Chris Masond352ac62008-09-29 15:18:18 -04001221/*
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001222 * update all the cowonly tree roots on disk
Chris Masond352ac62008-09-29 15:18:18 -04001223 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001224static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
1225 struct btrfs_root *root)
Chris Mason0f7d52f2007-04-09 10:42:37 -04001226{
Chris Mason0f7d52f2007-04-09 10:42:37 -04001227 struct btrfs_root *gang[8];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001228 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Mason0f7d52f2007-04-09 10:42:37 -04001229 int i;
1230 int ret;
Chris Mason54aa1f42007-06-22 14:16:25 -04001231 int err = 0;
1232
Josef Bacika4abeea2011-04-11 17:25:13 -04001233 spin_lock(&fs_info->fs_roots_radix_lock);
Chris Masond3977122009-01-05 21:25:51 -05001234 while (1) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001235 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1236 (void **)gang, 0,
Chris Mason0f7d52f2007-04-09 10:42:37 -04001237 ARRAY_SIZE(gang),
1238 BTRFS_ROOT_TRANS_TAG);
1239 if (ret == 0)
1240 break;
1241 for (i = 0; i < ret; i++) {
1242 root = gang[i];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001243 radix_tree_tag_clear(&fs_info->fs_roots_radix,
1244 (unsigned long)root->root_key.objectid,
1245 BTRFS_ROOT_TRANS_TAG);
Josef Bacika4abeea2011-04-11 17:25:13 -04001246 spin_unlock(&fs_info->fs_roots_radix_lock);
Yan Zheng31153d82008-07-28 15:32:19 -04001247
Chris Masone02119d2008-09-05 16:13:11 -04001248 btrfs_free_log(trans, root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001249 btrfs_update_reloc_root(trans, root);
Yan, Zhengd68fc572010-05-16 10:49:58 -04001250 btrfs_orphan_commit_root(trans, root);
Chris Masone02119d2008-09-05 16:13:11 -04001251
Li Zefan82d59022011-04-20 10:33:24 +08001252 btrfs_save_ino_cache(root, trans);
1253
Liu Bof1ebcc72011-11-14 20:48:06 -05001254 /* see comments in should_cow_block() */
Miao Xie27cdeb72014-04-02 19:51:05 +08001255 clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
Chris Masonc7548af2014-06-10 13:06:56 -07001256 smp_mb__after_atomic();
Liu Bof1ebcc72011-11-14 20:48:06 -05001257
Yan Zheng978d9102009-06-15 20:01:02 -04001258 if (root->commit_root != root->node) {
Josef Bacik9e351cc2014-03-13 15:42:13 -04001259 list_add_tail(&root->dirty_list,
1260 &trans->transaction->switch_commits);
Yan Zheng978d9102009-06-15 20:01:02 -04001261 btrfs_set_root_node(&root->root_item,
1262 root->node);
1263 }
Chris Mason9f3a7422007-08-07 15:52:19 -04001264
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001265 err = btrfs_update_root(trans, fs_info->tree_root,
Chris Mason0f7d52f2007-04-09 10:42:37 -04001266 &root->root_key,
1267 &root->root_item);
Josef Bacika4abeea2011-04-11 17:25:13 -04001268 spin_lock(&fs_info->fs_roots_radix_lock);
Chris Mason54aa1f42007-06-22 14:16:25 -04001269 if (err)
1270 break;
Qu Wenruo71741092015-09-08 17:22:41 +08001271 btrfs_qgroup_free_meta_all(root);
Chris Mason0f7d52f2007-04-09 10:42:37 -04001272 }
1273 }
Josef Bacika4abeea2011-04-11 17:25:13 -04001274 spin_unlock(&fs_info->fs_roots_radix_lock);
Chris Mason54aa1f42007-06-22 14:16:25 -04001275 return err;
Chris Mason0f7d52f2007-04-09 10:42:37 -04001276}
1277
Chris Masond352ac62008-09-29 15:18:18 -04001278/*
Eric Sandeende78b512013-01-31 18:21:12 +00001279 * defrag a given btree.
1280 * Every leaf in the btree is read and defragged.
Chris Masond352ac62008-09-29 15:18:18 -04001281 */
Eric Sandeende78b512013-01-31 18:21:12 +00001282int btrfs_defrag_root(struct btrfs_root *root)
Chris Masone9d0b132007-08-10 14:06:19 -04001283{
1284 struct btrfs_fs_info *info = root->fs_info;
Chris Masone9d0b132007-08-10 14:06:19 -04001285 struct btrfs_trans_handle *trans;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001286 int ret;
Chris Masone9d0b132007-08-10 14:06:19 -04001287
Miao Xie27cdeb72014-04-02 19:51:05 +08001288 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state))
Chris Masone9d0b132007-08-10 14:06:19 -04001289 return 0;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001290
Chris Mason6b800532007-10-15 16:17:34 -04001291 while (1) {
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001292 trans = btrfs_start_transaction(root, 0);
1293 if (IS_ERR(trans))
1294 return PTR_ERR(trans);
1295
Eric Sandeende78b512013-01-31 18:21:12 +00001296 ret = btrfs_defrag_leaves(trans, root);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001297
Chris Masone9d0b132007-08-10 14:06:19 -04001298 btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00001299 btrfs_btree_balance_dirty(info->tree_root);
Chris Masone9d0b132007-08-10 14:06:19 -04001300 cond_resched();
1301
David Sterba7841cb22011-05-31 18:07:27 +02001302 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
Chris Masone9d0b132007-08-10 14:06:19 -04001303 break;
David Sterba210549e2013-02-09 23:38:06 +00001304
1305 if (btrfs_defrag_cancelled(root->fs_info)) {
Frank Holtonefe120a2013-12-20 11:37:06 -05001306 pr_debug("BTRFS: defrag_root cancelled\n");
David Sterba210549e2013-02-09 23:38:06 +00001307 ret = -EAGAIN;
1308 break;
1309 }
Chris Masone9d0b132007-08-10 14:06:19 -04001310 }
Miao Xie27cdeb72014-04-02 19:51:05 +08001311 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001312 return ret;
Chris Masone9d0b132007-08-10 14:06:19 -04001313}
1314
Chris Masond352ac62008-09-29 15:18:18 -04001315/*
Qu Wenruo6426c7a2016-05-11 12:53:52 -07001316 * Do all special snapshot related qgroup dirty hack.
1317 *
1318 * Will do all needed qgroup inherit and dirty hack like switch commit
1319 * roots inside one transaction and write all btree into disk, to make
1320 * qgroup works.
1321 */
1322static int qgroup_account_snapshot(struct btrfs_trans_handle *trans,
1323 struct btrfs_root *src,
1324 struct btrfs_root *parent,
1325 struct btrfs_qgroup_inherit *inherit,
1326 u64 dst_objectid)
1327{
1328 struct btrfs_fs_info *fs_info = src->fs_info;
1329 int ret;
1330
1331 /*
1332 * Save some performance in the case that qgroups are not
1333 * enabled. If this check races with the ioctl, rescan will
1334 * kick in anyway.
1335 */
1336 mutex_lock(&fs_info->qgroup_ioctl_lock);
Josef Bacikafcdd122016-09-02 15:40:02 -04001337 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
Qu Wenruo6426c7a2016-05-11 12:53:52 -07001338 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1339 return 0;
1340 }
1341 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1342
1343 /*
1344 * We are going to commit transaction, see btrfs_commit_transaction()
1345 * comment for reason locking tree_log_mutex
1346 */
1347 mutex_lock(&fs_info->tree_log_mutex);
1348
1349 ret = commit_fs_roots(trans, src);
1350 if (ret)
1351 goto out;
1352 ret = btrfs_qgroup_prepare_account_extents(trans, fs_info);
1353 if (ret < 0)
1354 goto out;
1355 ret = btrfs_qgroup_account_extents(trans, fs_info);
1356 if (ret < 0)
1357 goto out;
1358
1359 /* Now qgroup are all updated, we can inherit it to new qgroups */
1360 ret = btrfs_qgroup_inherit(trans, fs_info,
1361 src->root_key.objectid, dst_objectid,
1362 inherit);
1363 if (ret < 0)
1364 goto out;
1365
1366 /*
1367 * Now we do a simplified commit transaction, which will:
1368 * 1) commit all subvolume and extent tree
1369 * To ensure all subvolume and extent tree have a valid
1370 * commit_root to accounting later insert_dir_item()
1371 * 2) write all btree blocks onto disk
1372 * This is to make sure later btree modification will be cowed
1373 * Or commit_root can be populated and cause wrong qgroup numbers
1374 * In this simplified commit, we don't really care about other trees
1375 * like chunk and root tree, as they won't affect qgroup.
1376 * And we don't write super to avoid half committed status.
1377 */
1378 ret = commit_cowonly_roots(trans, src);
1379 if (ret)
1380 goto out;
1381 switch_commit_roots(trans->transaction, fs_info);
1382 ret = btrfs_write_and_wait_transaction(trans, src);
1383 if (ret)
David Sterbaf7af3932016-06-17 18:15:25 +02001384 btrfs_handle_fs_error(fs_info, ret,
Qu Wenruo6426c7a2016-05-11 12:53:52 -07001385 "Error while writing out transaction for qgroup");
1386
1387out:
1388 mutex_unlock(&fs_info->tree_log_mutex);
1389
1390 /*
1391 * Force parent root to be updated, as we recorded it before so its
1392 * last_trans == cur_transid.
1393 * Or it won't be committed again onto disk after later
1394 * insert_dir_item()
1395 */
1396 if (!ret)
1397 record_root_in_trans(trans, parent, 1);
1398 return ret;
1399}
1400
1401/*
Chris Masond352ac62008-09-29 15:18:18 -04001402 * new snapshots need to be created at a very specific time in the
Miao Xieaec80302013-03-04 09:44:29 +00001403 * transaction commit. This does the actual creation.
1404 *
1405 * Note:
1406 * If the error which may affect the commitment of the current transaction
1407 * happens, we should return the error number. If the error which just affect
1408 * the creation of the pending snapshots, just return 0.
Chris Masond352ac62008-09-29 15:18:18 -04001409 */
Chris Mason80b67942008-02-01 16:35:04 -05001410static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
Chris Mason3063d292008-01-08 15:46:30 -05001411 struct btrfs_fs_info *fs_info,
1412 struct btrfs_pending_snapshot *pending)
1413{
1414 struct btrfs_key key;
Chris Mason80b67942008-02-01 16:35:04 -05001415 struct btrfs_root_item *new_root_item;
Chris Mason3063d292008-01-08 15:46:30 -05001416 struct btrfs_root *tree_root = fs_info->tree_root;
1417 struct btrfs_root *root = pending->root;
Sage Weil6bdb72d2010-03-15 17:27:13 +00001418 struct btrfs_root *parent_root;
Liu Bo98c99422011-09-11 10:52:24 -04001419 struct btrfs_block_rsv *rsv;
Sage Weil6bdb72d2010-03-15 17:27:13 +00001420 struct inode *parent_inode;
Miao Xie42874b32012-09-06 04:03:32 -06001421 struct btrfs_path *path;
1422 struct btrfs_dir_item *dir_item;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001423 struct dentry *dentry;
Chris Mason3063d292008-01-08 15:46:30 -05001424 struct extent_buffer *tmp;
Chris Mason925baed2008-06-25 16:01:30 -04001425 struct extent_buffer *old;
Deepa Dinamani04b285f2016-02-06 23:57:21 -08001426 struct timespec cur_time;
Miao Xieaec80302013-03-04 09:44:29 +00001427 int ret = 0;
Yan, Zhengd68fc572010-05-16 10:49:58 -04001428 u64 to_reserve = 0;
Sage Weil6bdb72d2010-03-15 17:27:13 +00001429 u64 index = 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001430 u64 objectid;
Li Zefanb83cc962010-12-20 16:04:08 +08001431 u64 root_flags;
Alexander Block8ea05e32012-07-25 17:35:53 +02001432 uuid_le new_uuid;
Chris Mason3063d292008-01-08 15:46:30 -05001433
David Sterba8546b572015-11-10 18:54:03 +01001434 ASSERT(pending->path);
1435 path = pending->path;
Miao Xie42874b32012-09-06 04:03:32 -06001436
David Sterbab0c0ea62015-11-10 18:54:00 +01001437 ASSERT(pending->root_item);
1438 new_root_item = pending->root_item;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001439
Miao Xieaec80302013-03-04 09:44:29 +00001440 pending->error = btrfs_find_free_objectid(tree_root, &objectid);
1441 if (pending->error)
Miao Xie6fa97002012-09-06 04:00:32 -06001442 goto no_free_objectid;
Chris Mason3063d292008-01-08 15:46:30 -05001443
Qu Wenruod6726332015-04-20 10:09:06 +08001444 /*
1445 * Make qgroup to skip current new snapshot's qgroupid, as it is
1446 * accounted by later btrfs_qgroup_inherit().
1447 */
1448 btrfs_set_skip_qgroup(trans, objectid);
1449
Zhaolei147d2562015-08-06 20:58:11 +08001450 btrfs_reloc_pre_snapshot(pending, &to_reserve);
Yan, Zhengd68fc572010-05-16 10:49:58 -04001451
1452 if (to_reserve > 0) {
Miao Xieaec80302013-03-04 09:44:29 +00001453 pending->error = btrfs_block_rsv_add(root,
1454 &pending->block_rsv,
1455 to_reserve,
1456 BTRFS_RESERVE_NO_FLUSH);
1457 if (pending->error)
Qu Wenruod6726332015-04-20 10:09:06 +08001458 goto clear_skip_qgroup;
Yan, Zhengd68fc572010-05-16 10:49:58 -04001459 }
1460
Chris Mason3063d292008-01-08 15:46:30 -05001461 key.objectid = objectid;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001462 key.offset = (u64)-1;
1463 key.type = BTRFS_ROOT_ITEM_KEY;
Chris Mason3063d292008-01-08 15:46:30 -05001464
Miao Xie6fa97002012-09-06 04:00:32 -06001465 rsv = trans->block_rsv;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001466 trans->block_rsv = &pending->block_rsv;
Liu Bo2382c5c2013-02-22 04:33:36 +00001467 trans->bytes_reserved = trans->block_rsv->reserved;
Josef Bacik88d3a5a2016-01-13 13:21:20 -05001468 trace_btrfs_space_reservation(root->fs_info, "transaction",
1469 trans->transid,
1470 trans->bytes_reserved, 1);
Yan, Zhenga22285a2010-05-16 10:48:46 -04001471 dentry = pending->dentry;
Miao Xiee9662f72013-02-28 10:01:15 +00001472 parent_inode = pending->dir;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001473 parent_root = BTRFS_I(parent_inode)->root;
Qu Wenruo6426c7a2016-05-11 12:53:52 -07001474 record_root_in_trans(trans, parent_root, 0);
Yan, Zhenga22285a2010-05-16 10:48:46 -04001475
Deepa Dinamani04b285f2016-02-06 23:57:21 -08001476 cur_time = current_fs_time(parent_inode->i_sb);
1477
Sage Weil6bdb72d2010-03-15 17:27:13 +00001478 /*
1479 * insert the directory item
1480 */
Sage Weil6bdb72d2010-03-15 17:27:13 +00001481 ret = btrfs_set_inode_index(parent_inode, &index);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001482 BUG_ON(ret); /* -ENOMEM */
Miao Xie42874b32012-09-06 04:03:32 -06001483
1484 /* check if there is a file/dir which has the same name. */
1485 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1486 btrfs_ino(parent_inode),
1487 dentry->d_name.name,
1488 dentry->d_name.len, 0);
1489 if (dir_item != NULL && !IS_ERR(dir_item)) {
Chris Masonfe66a052012-02-20 08:40:56 -05001490 pending->error = -EEXIST;
Miao Xieaec80302013-03-04 09:44:29 +00001491 goto dir_item_existed;
Miao Xie42874b32012-09-06 04:03:32 -06001492 } else if (IS_ERR(dir_item)) {
1493 ret = PTR_ERR(dir_item);
Jeff Mahoney66642832016-06-10 18:19:25 -04001494 btrfs_abort_transaction(trans, ret);
Miao Xie8732d442012-09-17 23:52:38 -06001495 goto fail;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001496 }
Miao Xie42874b32012-09-06 04:03:32 -06001497 btrfs_release_path(path);
Sage Weil6bdb72d2010-03-15 17:27:13 +00001498
Chris Masone9993762011-06-17 16:14:09 -04001499 /*
1500 * pull in the delayed directory update
1501 * and the delayed inode item
1502 * otherwise we corrupt the FS during
1503 * snapshot
1504 */
1505 ret = btrfs_run_delayed_items(trans, root);
Miao Xie8732d442012-09-17 23:52:38 -06001506 if (ret) { /* Transaction aborted */
Jeff Mahoney66642832016-06-10 18:19:25 -04001507 btrfs_abort_transaction(trans, ret);
Miao Xie8732d442012-09-17 23:52:38 -06001508 goto fail;
1509 }
Chris Masone9993762011-06-17 16:14:09 -04001510
Qu Wenruo6426c7a2016-05-11 12:53:52 -07001511 record_root_in_trans(trans, root, 0);
Sage Weil6bdb72d2010-03-15 17:27:13 +00001512 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1513 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
Li Zefan08fe4db2011-03-28 02:01:25 +00001514 btrfs_check_and_init_root_item(new_root_item);
Sage Weil6bdb72d2010-03-15 17:27:13 +00001515
Li Zefanb83cc962010-12-20 16:04:08 +08001516 root_flags = btrfs_root_flags(new_root_item);
1517 if (pending->readonly)
1518 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1519 else
1520 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1521 btrfs_set_root_flags(new_root_item, root_flags);
1522
Alexander Block8ea05e32012-07-25 17:35:53 +02001523 btrfs_set_root_generation_v2(new_root_item,
1524 trans->transid);
1525 uuid_le_gen(&new_uuid);
1526 memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
1527 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1528 BTRFS_UUID_SIZE);
Stefan Behrens70023da2013-04-17 09:11:47 +00001529 if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1530 memset(new_root_item->received_uuid, 0,
1531 sizeof(new_root_item->received_uuid));
1532 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1533 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1534 btrfs_set_root_stransid(new_root_item, 0);
1535 btrfs_set_root_rtransid(new_root_item, 0);
1536 }
Qu Wenruo3cae2102013-07-16 11:19:18 +08001537 btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1538 btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
Alexander Block8ea05e32012-07-25 17:35:53 +02001539 btrfs_set_root_otransid(new_root_item, trans->transid);
Alexander Block8ea05e32012-07-25 17:35:53 +02001540
Chris Mason925baed2008-06-25 16:01:30 -04001541 old = btrfs_lock_root_node(root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001542 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001543 if (ret) {
1544 btrfs_tree_unlock(old);
1545 free_extent_buffer(old);
Jeff Mahoney66642832016-06-10 18:19:25 -04001546 btrfs_abort_transaction(trans, ret);
Miao Xie8732d442012-09-17 23:52:38 -06001547 goto fail;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001548 }
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001549
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001550 btrfs_set_lock_blocking(old);
Chris Mason3063d292008-01-08 15:46:30 -05001551
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001552 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001553 /* clean up in any case */
Chris Mason925baed2008-06-25 16:01:30 -04001554 btrfs_tree_unlock(old);
1555 free_extent_buffer(old);
Miao Xie8732d442012-09-17 23:52:38 -06001556 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04001557 btrfs_abort_transaction(trans, ret);
Miao Xie8732d442012-09-17 23:52:38 -06001558 goto fail;
1559 }
Liu Bof1ebcc72011-11-14 20:48:06 -05001560 /* see comments in should_cow_block() */
Miao Xie27cdeb72014-04-02 19:51:05 +08001561 set_bit(BTRFS_ROOT_FORCE_COW, &root->state);
Liu Bof1ebcc72011-11-14 20:48:06 -05001562 smp_wmb();
1563
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001564 btrfs_set_root_node(new_root_item, tmp);
Yan, Zhenga22285a2010-05-16 10:48:46 -04001565 /* record when the snapshot was created in key.offset */
1566 key.offset = trans->transid;
1567 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
Chris Mason925baed2008-06-25 16:01:30 -04001568 btrfs_tree_unlock(tmp);
Chris Mason3063d292008-01-08 15:46:30 -05001569 free_extent_buffer(tmp);
Miao Xie8732d442012-09-17 23:52:38 -06001570 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04001571 btrfs_abort_transaction(trans, ret);
Miao Xie8732d442012-09-17 23:52:38 -06001572 goto fail;
1573 }
Chris Mason0660b5a2008-11-17 20:37:39 -05001574
Yan, Zhenga22285a2010-05-16 10:48:46 -04001575 /*
1576 * insert root back/forward references
1577 */
1578 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1579 parent_root->root_key.objectid,
Li Zefan33345d012011-04-20 10:31:50 +08001580 btrfs_ino(parent_inode), index,
Yan, Zhenga22285a2010-05-16 10:48:46 -04001581 dentry->d_name.name, dentry->d_name.len);
Miao Xie8732d442012-09-17 23:52:38 -06001582 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04001583 btrfs_abort_transaction(trans, ret);
Miao Xie8732d442012-09-17 23:52:38 -06001584 goto fail;
1585 }
Yan, Zhenga22285a2010-05-16 10:48:46 -04001586
1587 key.offset = (u64)-1;
1588 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001589 if (IS_ERR(pending->snap)) {
1590 ret = PTR_ERR(pending->snap);
Jeff Mahoney66642832016-06-10 18:19:25 -04001591 btrfs_abort_transaction(trans, ret);
Miao Xie8732d442012-09-17 23:52:38 -06001592 goto fail;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001593 }
Yan, Zhengd68fc572010-05-16 10:49:58 -04001594
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001595 ret = btrfs_reloc_post_snapshot(trans, pending);
Miao Xie8732d442012-09-17 23:52:38 -06001596 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04001597 btrfs_abort_transaction(trans, ret);
Miao Xie8732d442012-09-17 23:52:38 -06001598 goto fail;
1599 }
Miao Xie361048f2012-09-06 04:00:57 -06001600
1601 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
Miao Xie8732d442012-09-17 23:52:38 -06001602 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04001603 btrfs_abort_transaction(trans, ret);
Miao Xie8732d442012-09-17 23:52:38 -06001604 goto fail;
1605 }
Miao Xie42874b32012-09-06 04:03:32 -06001606
Qu Wenruo6426c7a2016-05-11 12:53:52 -07001607 /*
1608 * Do special qgroup accounting for snapshot, as we do some qgroup
1609 * snapshot hack to do fast snapshot.
1610 * To co-operate with that hack, we do hack again.
1611 * Or snapshot will be greatly slowed down by a subtree qgroup rescan
1612 */
1613 ret = qgroup_account_snapshot(trans, root, parent_root,
1614 pending->inherit, objectid);
1615 if (ret < 0)
1616 goto fail;
1617
Miao Xie42874b32012-09-06 04:03:32 -06001618 ret = btrfs_insert_dir_item(trans, parent_root,
1619 dentry->d_name.name, dentry->d_name.len,
1620 parent_inode, &key,
1621 BTRFS_FT_DIR, index);
1622 /* We have check then name at the beginning, so it is impossible. */
Chris Mason9c520572012-12-17 14:26:57 -05001623 BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
Miao Xie8732d442012-09-17 23:52:38 -06001624 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04001625 btrfs_abort_transaction(trans, ret);
Miao Xie8732d442012-09-17 23:52:38 -06001626 goto fail;
1627 }
Miao Xie42874b32012-09-06 04:03:32 -06001628
1629 btrfs_i_size_write(parent_inode, parent_inode->i_size +
1630 dentry->d_name.len * 2);
Deepa Dinamani04b285f2016-02-06 23:57:21 -08001631 parent_inode->i_mtime = parent_inode->i_ctime =
1632 current_fs_time(parent_inode->i_sb);
Josef Bacikbe6aef62012-10-22 15:43:12 -04001633 ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode);
Stefan Behrensdd5f9612013-08-15 17:11:20 +02001634 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04001635 btrfs_abort_transaction(trans, ret);
Stefan Behrensdd5f9612013-08-15 17:11:20 +02001636 goto fail;
1637 }
1638 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, new_uuid.b,
1639 BTRFS_UUID_KEY_SUBVOL, objectid);
1640 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04001641 btrfs_abort_transaction(trans, ret);
Stefan Behrensdd5f9612013-08-15 17:11:20 +02001642 goto fail;
1643 }
1644 if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1645 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
1646 new_root_item->received_uuid,
1647 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1648 objectid);
1649 if (ret && ret != -EEXIST) {
Jeff Mahoney66642832016-06-10 18:19:25 -04001650 btrfs_abort_transaction(trans, ret);
Stefan Behrensdd5f9612013-08-15 17:11:20 +02001651 goto fail;
1652 }
1653 }
Qu Wenruod6726332015-04-20 10:09:06 +08001654
1655 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1656 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04001657 btrfs_abort_transaction(trans, ret);
Qu Wenruod6726332015-04-20 10:09:06 +08001658 goto fail;
1659 }
1660
Chris Mason3063d292008-01-08 15:46:30 -05001661fail:
Miao Xieaec80302013-03-04 09:44:29 +00001662 pending->error = ret;
1663dir_item_existed:
Liu Bo98c99422011-09-11 10:52:24 -04001664 trans->block_rsv = rsv;
Liu Bo2382c5c2013-02-22 04:33:36 +00001665 trans->bytes_reserved = 0;
Qu Wenruod6726332015-04-20 10:09:06 +08001666clear_skip_qgroup:
1667 btrfs_clear_skip_qgroup(trans);
Miao Xie6fa97002012-09-06 04:00:32 -06001668no_free_objectid:
1669 kfree(new_root_item);
David Sterbab0c0ea62015-11-10 18:54:00 +01001670 pending->root_item = NULL;
Miao Xie42874b32012-09-06 04:03:32 -06001671 btrfs_free_path(path);
David Sterba8546b572015-11-10 18:54:03 +01001672 pending->path = NULL;
1673
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001674 return ret;
Chris Mason3063d292008-01-08 15:46:30 -05001675}
1676
Chris Masond352ac62008-09-29 15:18:18 -04001677/*
1678 * create all the snapshots we've scheduled for creation
1679 */
Chris Mason80b67942008-02-01 16:35:04 -05001680static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1681 struct btrfs_fs_info *fs_info)
Chris Mason3063d292008-01-08 15:46:30 -05001682{
Miao Xieaec80302013-03-04 09:44:29 +00001683 struct btrfs_pending_snapshot *pending, *next;
Chris Mason3063d292008-01-08 15:46:30 -05001684 struct list_head *head = &trans->transaction->pending_snapshots;
Miao Xieaec80302013-03-04 09:44:29 +00001685 int ret = 0;
Chris Mason3de45862008-11-17 21:02:50 -05001686
Miao Xieaec80302013-03-04 09:44:29 +00001687 list_for_each_entry_safe(pending, next, head, list) {
1688 list_del(&pending->list);
1689 ret = create_pending_snapshot(trans, fs_info, pending);
1690 if (ret)
1691 break;
1692 }
1693 return ret;
Chris Mason3de45862008-11-17 21:02:50 -05001694}
1695
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001696static void update_super_roots(struct btrfs_root *root)
1697{
1698 struct btrfs_root_item *root_item;
1699 struct btrfs_super_block *super;
1700
David Sterba6c417612011-04-13 15:41:04 +02001701 super = root->fs_info->super_copy;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001702
1703 root_item = &root->fs_info->chunk_root->root_item;
1704 super->chunk_root = root_item->bytenr;
1705 super->chunk_root_generation = root_item->generation;
1706 super->chunk_root_level = root_item->level;
1707
1708 root_item = &root->fs_info->tree_root->root_item;
1709 super->root = root_item->bytenr;
1710 super->generation = root_item->generation;
1711 super->root_level = root_item->level;
Jeff Mahoney3cdde222016-06-09 21:38:35 -04001712 if (btrfs_test_opt(root->fs_info, SPACE_CACHE))
Josef Bacik0af3d002010-06-21 14:48:16 -04001713 super->cache_generation = root_item->generation;
Josef Bacikafcdd122016-09-02 15:40:02 -04001714 if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &root->fs_info->flags))
Stefan Behrens70f80172013-08-15 17:11:23 +02001715 super->uuid_tree_generation = root_item->generation;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001716}
1717
Chris Masonf36f3042009-07-30 10:04:48 -04001718int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1719{
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001720 struct btrfs_transaction *trans;
Chris Masonf36f3042009-07-30 10:04:48 -04001721 int ret = 0;
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001722
Josef Bacika4abeea2011-04-11 17:25:13 -04001723 spin_lock(&info->trans_lock);
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001724 trans = info->running_transaction;
1725 if (trans)
1726 ret = (trans->state >= TRANS_STATE_COMMIT_START);
Josef Bacika4abeea2011-04-11 17:25:13 -04001727 spin_unlock(&info->trans_lock);
Chris Masonf36f3042009-07-30 10:04:48 -04001728 return ret;
1729}
1730
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001731int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1732{
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001733 struct btrfs_transaction *trans;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001734 int ret = 0;
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001735
Josef Bacika4abeea2011-04-11 17:25:13 -04001736 spin_lock(&info->trans_lock);
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001737 trans = info->running_transaction;
1738 if (trans)
1739 ret = is_transaction_blocked(trans);
Josef Bacika4abeea2011-04-11 17:25:13 -04001740 spin_unlock(&info->trans_lock);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001741 return ret;
1742}
1743
Sage Weilbb9c12c2010-10-29 15:37:34 -04001744/*
1745 * wait for the current transaction commit to start and block subsequent
1746 * transaction joins
1747 */
1748static void wait_current_trans_commit_start(struct btrfs_root *root,
1749 struct btrfs_transaction *trans)
1750{
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001751 wait_event(root->fs_info->transaction_blocked_wait,
Josef Bacik501407a2013-06-10 16:47:23 -04001752 trans->state >= TRANS_STATE_COMMIT_START ||
1753 trans->aborted);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001754}
1755
1756/*
1757 * wait for the current transaction to start and then become unblocked.
1758 * caller holds ref.
1759 */
1760static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1761 struct btrfs_transaction *trans)
1762{
Li Zefan72d63ed2011-07-14 03:17:00 +00001763 wait_event(root->fs_info->transaction_wait,
Josef Bacik501407a2013-06-10 16:47:23 -04001764 trans->state >= TRANS_STATE_UNBLOCKED ||
1765 trans->aborted);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001766}
1767
1768/*
1769 * commit transactions asynchronously. once btrfs_commit_transaction_async
1770 * returns, any subsequent transaction will not be allowed to join.
1771 */
1772struct btrfs_async_commit {
1773 struct btrfs_trans_handle *newtrans;
1774 struct btrfs_root *root;
Miao Xie7892b5a2012-11-15 08:14:47 +00001775 struct work_struct work;
Sage Weilbb9c12c2010-10-29 15:37:34 -04001776};
1777
1778static void do_async_commit(struct work_struct *work)
1779{
1780 struct btrfs_async_commit *ac =
Miao Xie7892b5a2012-11-15 08:14:47 +00001781 container_of(work, struct btrfs_async_commit, work);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001782
Sage Weil6fc4e35482012-08-30 16:26:15 -06001783 /*
1784 * We've got freeze protection passed with the transaction.
1785 * Tell lockdep about it.
1786 */
Liu Bob1a06a42013-11-06 16:57:55 +08001787 if (ac->newtrans->type & __TRANS_FREEZABLE)
Oleg Nesterovbee91822015-07-19 23:48:20 +02001788 __sb_writers_acquired(ac->root->fs_info->sb, SB_FREEZE_FS);
Sage Weil6fc4e35482012-08-30 16:26:15 -06001789
Sage Weile209db72012-08-30 16:26:16 -06001790 current->journal_info = ac->newtrans;
1791
Sage Weilbb9c12c2010-10-29 15:37:34 -04001792 btrfs_commit_transaction(ac->newtrans, ac->root);
1793 kfree(ac);
1794}
1795
1796int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1797 struct btrfs_root *root,
1798 int wait_for_unblock)
1799{
1800 struct btrfs_async_commit *ac;
1801 struct btrfs_transaction *cur_trans;
1802
1803 ac = kmalloc(sizeof(*ac), GFP_NOFS);
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00001804 if (!ac)
1805 return -ENOMEM;
Sage Weilbb9c12c2010-10-29 15:37:34 -04001806
Miao Xie7892b5a2012-11-15 08:14:47 +00001807 INIT_WORK(&ac->work, do_async_commit);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001808 ac->root = root;
Josef Bacik7a7eaa42011-04-13 12:54:33 -04001809 ac->newtrans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00001810 if (IS_ERR(ac->newtrans)) {
1811 int err = PTR_ERR(ac->newtrans);
1812 kfree(ac);
1813 return err;
1814 }
Sage Weilbb9c12c2010-10-29 15:37:34 -04001815
1816 /* take transaction reference */
Sage Weilbb9c12c2010-10-29 15:37:34 -04001817 cur_trans = trans->transaction;
Josef Bacik13c5a932011-04-11 15:45:29 -04001818 atomic_inc(&cur_trans->use_count);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001819
1820 btrfs_end_transaction(trans, root);
Sage Weil6fc4e35482012-08-30 16:26:15 -06001821
1822 /*
1823 * Tell lockdep we've released the freeze rwsem, since the
1824 * async commit thread will be the one to unlock it.
1825 */
Liu Bob1a06a42013-11-06 16:57:55 +08001826 if (ac->newtrans->type & __TRANS_FREEZABLE)
Oleg Nesterovbee91822015-07-19 23:48:20 +02001827 __sb_writers_release(root->fs_info->sb, SB_FREEZE_FS);
Sage Weil6fc4e35482012-08-30 16:26:15 -06001828
Miao Xie7892b5a2012-11-15 08:14:47 +00001829 schedule_work(&ac->work);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001830
1831 /* wait for transaction to start and unblock */
Sage Weilbb9c12c2010-10-29 15:37:34 -04001832 if (wait_for_unblock)
1833 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1834 else
1835 wait_current_trans_commit_start(root, cur_trans);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001836
Sage Weil38e88052011-06-10 18:43:13 +00001837 if (current->journal_info == trans)
1838 current->journal_info = NULL;
1839
Josef Bacik724e2312013-09-30 11:36:38 -04001840 btrfs_put_transaction(cur_trans);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001841 return 0;
1842}
1843
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001844
1845static void cleanup_transaction(struct btrfs_trans_handle *trans,
Josef Bacik7b8b92a2012-05-31 15:52:43 -04001846 struct btrfs_root *root, int err)
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001847{
1848 struct btrfs_transaction *cur_trans = trans->transaction;
Liu Bof094ac32013-02-27 13:28:25 +00001849 DEFINE_WAIT(wait);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001850
1851 WARN_ON(trans->use_count > 1);
1852
Jeff Mahoney66642832016-06-10 18:19:25 -04001853 btrfs_abort_transaction(trans, err);
Josef Bacik7b8b92a2012-05-31 15:52:43 -04001854
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001855 spin_lock(&root->fs_info->trans_lock);
Liu Bo66b61352013-03-04 16:25:41 +00001856
Miao Xie25d8c282013-05-15 07:48:26 +00001857 /*
1858 * If the transaction is removed from the list, it means this
1859 * transaction has been committed successfully, so it is impossible
1860 * to call the cleanup function.
1861 */
1862 BUG_ON(list_empty(&cur_trans->list));
Liu Bo66b61352013-03-04 16:25:41 +00001863
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001864 list_del_init(&cur_trans->list);
Josef Bacikd7096fc2012-05-31 15:49:57 -04001865 if (cur_trans == root->fs_info->running_transaction) {
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001866 cur_trans->state = TRANS_STATE_COMMIT_DOING;
Liu Bof094ac32013-02-27 13:28:25 +00001867 spin_unlock(&root->fs_info->trans_lock);
1868 wait_event(cur_trans->writer_wait,
1869 atomic_read(&cur_trans->num_writers) == 1);
1870
1871 spin_lock(&root->fs_info->trans_lock);
Josef Bacikd7096fc2012-05-31 15:49:57 -04001872 }
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001873 spin_unlock(&root->fs_info->trans_lock);
1874
1875 btrfs_cleanup_one_transaction(trans->transaction, root);
1876
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001877 spin_lock(&root->fs_info->trans_lock);
1878 if (cur_trans == root->fs_info->running_transaction)
1879 root->fs_info->running_transaction = NULL;
1880 spin_unlock(&root->fs_info->trans_lock);
1881
Josef Bacike0228282013-09-20 22:26:29 -04001882 if (trans->type & __TRANS_FREEZABLE)
1883 sb_end_intwrite(root->fs_info->sb);
Josef Bacik724e2312013-09-30 11:36:38 -04001884 btrfs_put_transaction(cur_trans);
1885 btrfs_put_transaction(cur_trans);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001886
1887 trace_btrfs_transaction_commit(root);
1888
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001889 if (current->journal_info == trans)
1890 current->journal_info = NULL;
Wang Shilongc0af8f02014-02-19 19:24:18 +08001891 btrfs_scrub_cancel(root->fs_info);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001892
1893 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1894}
1895
Miao Xie82436612013-05-15 07:48:28 +00001896static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
1897{
Jeff Mahoney3cdde222016-06-09 21:38:35 -04001898 if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
Miao Xie6c255e62014-03-06 13:55:01 +08001899 return btrfs_start_delalloc_roots(fs_info, 1, -1);
Miao Xie82436612013-05-15 07:48:28 +00001900 return 0;
1901}
1902
1903static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
1904{
Jeff Mahoney3cdde222016-06-09 21:38:35 -04001905 if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
Filipe Manana578def72016-04-26 15:36:38 +01001906 btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1);
Miao Xie82436612013-05-15 07:48:28 +00001907}
1908
Josef Bacik50d9aa92014-11-21 14:52:38 -05001909static inline void
Josef Bacik161c35492015-09-24 16:17:39 -04001910btrfs_wait_pending_ordered(struct btrfs_transaction *cur_trans)
Josef Bacik50d9aa92014-11-21 14:52:38 -05001911{
Josef Bacik161c35492015-09-24 16:17:39 -04001912 wait_event(cur_trans->pending_wait,
1913 atomic_read(&cur_trans->pending_ordered) == 0);
Josef Bacik50d9aa92014-11-21 14:52:38 -05001914}
1915
Chris Mason79154b12007-03-22 15:59:16 -04001916int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1917 struct btrfs_root *root)
1918{
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001919 struct btrfs_transaction *cur_trans = trans->transaction;
Chris Mason8fd17792007-04-19 21:01:03 -04001920 struct btrfs_transaction *prev_trans = NULL;
Miao Xie25287e02012-10-25 09:31:03 +00001921 int ret;
Chris Mason79154b12007-03-22 15:59:16 -04001922
Miao Xie8d25a082013-01-15 06:27:25 +00001923 /* Stop the commit early if ->aborted is set */
1924 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
Miao Xie25287e02012-10-25 09:31:03 +00001925 ret = cur_trans->aborted;
Josef Bacike4a2bca2013-02-06 16:55:41 -05001926 btrfs_end_transaction(trans, root);
1927 return ret;
Miao Xie25287e02012-10-25 09:31:03 +00001928 }
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001929
Chris Mason56bec292009-03-13 10:10:06 -04001930 /* make a pass through all the delayed refs we have so far
1931 * any runnings procs may add more while we are here
1932 */
1933 ret = btrfs_run_delayed_refs(trans, root, 0);
Josef Bacike4a2bca2013-02-06 16:55:41 -05001934 if (ret) {
1935 btrfs_end_transaction(trans, root);
1936 return ret;
1937 }
Chris Mason56bec292009-03-13 10:10:06 -04001938
Josef Bacik0e721102012-06-26 16:13:18 -04001939 btrfs_trans_release_metadata(trans, root);
1940 trans->block_rsv = NULL;
1941
Chris Masonb7ec40d2009-03-12 20:12:45 -04001942 cur_trans = trans->transaction;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001943
Chris Mason56bec292009-03-13 10:10:06 -04001944 /*
1945 * set the flushing flag so procs in this transaction have to
1946 * start sending their work down.
1947 */
Chris Masonb7ec40d2009-03-12 20:12:45 -04001948 cur_trans->delayed_refs.flushing = 1;
Josef Bacik1be41b72013-06-12 13:56:06 -04001949 smp_wmb();
Chris Mason56bec292009-03-13 10:10:06 -04001950
Josef Bacikea658ba2012-09-11 16:57:25 -04001951 if (!list_empty(&trans->new_bgs))
1952 btrfs_create_pending_block_groups(trans, root);
1953
Chris Masonc3e69d52009-03-13 10:17:05 -04001954 ret = btrfs_run_delayed_refs(trans, root, 0);
Josef Bacike4a2bca2013-02-06 16:55:41 -05001955 if (ret) {
1956 btrfs_end_transaction(trans, root);
1957 return ret;
1958 }
Chris Mason56bec292009-03-13 10:10:06 -04001959
Josef Bacik3204d332015-09-24 10:46:10 -04001960 if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) {
Chris Mason1bbc6212015-04-06 12:46:08 -07001961 int run_it = 0;
1962
1963 /* this mutex is also taken before trying to set
1964 * block groups readonly. We need to make sure
1965 * that nobody has set a block group readonly
1966 * after a extents from that block group have been
1967 * allocated for cache files. btrfs_set_block_group_ro
1968 * will wait for the transaction to commit if it
Josef Bacik3204d332015-09-24 10:46:10 -04001969 * finds BTRFS_TRANS_DIRTY_BG_RUN set.
Chris Mason1bbc6212015-04-06 12:46:08 -07001970 *
Josef Bacik3204d332015-09-24 10:46:10 -04001971 * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure
1972 * only one process starts all the block group IO. It wouldn't
Chris Mason1bbc6212015-04-06 12:46:08 -07001973 * hurt to have more than one go through, but there's no
1974 * real advantage to it either.
1975 */
1976 mutex_lock(&root->fs_info->ro_block_group_mutex);
Josef Bacik3204d332015-09-24 10:46:10 -04001977 if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN,
1978 &cur_trans->flags))
Chris Mason1bbc6212015-04-06 12:46:08 -07001979 run_it = 1;
Chris Mason1bbc6212015-04-06 12:46:08 -07001980 mutex_unlock(&root->fs_info->ro_block_group_mutex);
1981
1982 if (run_it)
1983 ret = btrfs_start_dirty_block_groups(trans, root);
1984 }
1985 if (ret) {
1986 btrfs_end_transaction(trans, root);
1987 return ret;
1988 }
1989
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001990 spin_lock(&root->fs_info->trans_lock);
1991 if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
1992 spin_unlock(&root->fs_info->trans_lock);
Josef Bacik13c5a932011-04-11 15:45:29 -04001993 atomic_inc(&cur_trans->use_count);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001994 ret = btrfs_end_transaction(trans, root);
Chris Masonccd467d2007-06-28 15:57:36 -04001995
Li Zefanb9c83002011-07-14 03:17:14 +00001996 wait_for_commit(root, cur_trans);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001997
Liu Bob4924a02015-03-06 20:23:44 +08001998 if (unlikely(cur_trans->aborted))
1999 ret = cur_trans->aborted;
2000
Josef Bacik724e2312013-09-30 11:36:38 -04002001 btrfs_put_transaction(cur_trans);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04002002
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002003 return ret;
Chris Mason79154b12007-03-22 15:59:16 -04002004 }
Chris Mason4313b392008-01-03 09:08:48 -05002005
Miao Xie4a9d8bd2013-05-17 03:53:43 +00002006 cur_trans->state = TRANS_STATE_COMMIT_START;
Sage Weilbb9c12c2010-10-29 15:37:34 -04002007 wake_up(&root->fs_info->transaction_blocked_wait);
2008
Chris Masonccd467d2007-06-28 15:57:36 -04002009 if (cur_trans->list.prev != &root->fs_info->trans_list) {
2010 prev_trans = list_entry(cur_trans->list.prev,
2011 struct btrfs_transaction, list);
Miao Xie4a9d8bd2013-05-17 03:53:43 +00002012 if (prev_trans->state != TRANS_STATE_COMPLETED) {
Josef Bacik13c5a932011-04-11 15:45:29 -04002013 atomic_inc(&prev_trans->use_count);
Josef Bacika4abeea2011-04-11 17:25:13 -04002014 spin_unlock(&root->fs_info->trans_lock);
Chris Masonccd467d2007-06-28 15:57:36 -04002015
2016 wait_for_commit(root, prev_trans);
Filipe Manana1f9b8c82015-08-12 11:54:35 +01002017 ret = prev_trans->aborted;
Chris Masonccd467d2007-06-28 15:57:36 -04002018
Josef Bacik724e2312013-09-30 11:36:38 -04002019 btrfs_put_transaction(prev_trans);
Filipe Manana1f9b8c82015-08-12 11:54:35 +01002020 if (ret)
2021 goto cleanup_transaction;
Josef Bacika4abeea2011-04-11 17:25:13 -04002022 } else {
2023 spin_unlock(&root->fs_info->trans_lock);
Chris Masonccd467d2007-06-28 15:57:36 -04002024 }
Josef Bacika4abeea2011-04-11 17:25:13 -04002025 } else {
2026 spin_unlock(&root->fs_info->trans_lock);
Chris Masonccd467d2007-06-28 15:57:36 -04002027 }
Josef Bacik15ee9bc2007-08-10 16:22:09 -04002028
Miao Xie0860adf2013-05-15 07:48:27 +00002029 extwriter_counter_dec(cur_trans, trans->type);
2030
Miao Xie82436612013-05-15 07:48:28 +00002031 ret = btrfs_start_delalloc_flush(root->fs_info);
2032 if (ret)
2033 goto cleanup_transaction;
2034
Chris Mason8d875f92014-08-12 10:47:42 -07002035 ret = btrfs_run_delayed_items(trans, root);
Miao Xie581227d2013-05-15 07:48:30 +00002036 if (ret)
2037 goto cleanup_transaction;
Chris Mason89573b92009-03-12 20:12:45 -04002038
Miao Xie581227d2013-05-15 07:48:30 +00002039 wait_event(cur_trans->writer_wait,
2040 extwriter_counter_read(cur_trans) == 0);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04002041
Miao Xie581227d2013-05-15 07:48:30 +00002042 /* some pending stuffs might be added after the previous flush. */
Chris Mason8d875f92014-08-12 10:47:42 -07002043 ret = btrfs_run_delayed_items(trans, root);
Miao Xieca469632012-11-01 07:33:14 +00002044 if (ret)
2045 goto cleanup_transaction;
2046
Miao Xie82436612013-05-15 07:48:28 +00002047 btrfs_wait_delalloc_flush(root->fs_info);
Wang Shilongcb7ab022013-12-04 21:16:53 +08002048
Josef Bacik161c35492015-09-24 16:17:39 -04002049 btrfs_wait_pending_ordered(cur_trans);
Josef Bacik50d9aa92014-11-21 14:52:38 -05002050
Wang Shilongcb7ab022013-12-04 21:16:53 +08002051 btrfs_scrub_pause(root);
Chris Mason75857172011-06-13 20:00:16 -04002052 /*
Josef Baciked0ca142011-06-14 16:22:15 -04002053 * Ok now we need to make sure to block out any other joins while we
2054 * commit the transaction. We could have started a join before setting
Miao Xie4a9d8bd2013-05-17 03:53:43 +00002055 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
Josef Baciked0ca142011-06-14 16:22:15 -04002056 */
2057 spin_lock(&root->fs_info->trans_lock);
Miao Xie4a9d8bd2013-05-17 03:53:43 +00002058 cur_trans->state = TRANS_STATE_COMMIT_DOING;
Josef Baciked0ca142011-06-14 16:22:15 -04002059 spin_unlock(&root->fs_info->trans_lock);
2060 wait_event(cur_trans->writer_wait,
2061 atomic_read(&cur_trans->num_writers) == 1);
2062
Miao Xie2cba30f2013-01-15 06:29:12 +00002063 /* ->aborted might be set after the previous check, so check it */
2064 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
2065 ret = cur_trans->aborted;
Wang Shilong6cf7f772014-02-19 19:24:16 +08002066 goto scrub_continue;
Miao Xie2cba30f2013-01-15 06:29:12 +00002067 }
Chris Masone038dca2011-06-17 14:16:13 -04002068 /*
Chris Mason75857172011-06-13 20:00:16 -04002069 * the reloc mutex makes sure that we stop
2070 * the balancing code from coming in and moving
2071 * extents around in the middle of the commit
2072 */
2073 mutex_lock(&root->fs_info->reloc_mutex);
2074
Miao Xie42874b32012-09-06 04:03:32 -06002075 /*
2076 * We needn't worry about the delayed items because we will
2077 * deal with them in create_pending_snapshot(), which is the
2078 * core function of the snapshot creation.
2079 */
2080 ret = create_pending_snapshots(trans, root->fs_info);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002081 if (ret) {
2082 mutex_unlock(&root->fs_info->reloc_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08002083 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002084 }
Chris Mason3063d292008-01-08 15:46:30 -05002085
Miao Xie42874b32012-09-06 04:03:32 -06002086 /*
2087 * We insert the dir indexes of the snapshots and update the inode
2088 * of the snapshots' parents after the snapshot creation, so there
2089 * are some delayed items which are not dealt with. Now deal with
2090 * them.
2091 *
2092 * We needn't worry that this operation will corrupt the snapshots,
2093 * because all the tree which are snapshoted will be forced to COW
2094 * the nodes and leaves.
2095 */
2096 ret = btrfs_run_delayed_items(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002097 if (ret) {
2098 mutex_unlock(&root->fs_info->reloc_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08002099 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002100 }
Miao Xie16cdcec2011-04-22 18:12:22 +08002101
Chris Mason56bec292009-03-13 10:10:06 -04002102 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002103 if (ret) {
2104 mutex_unlock(&root->fs_info->reloc_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08002105 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002106 }
Chris Mason56bec292009-03-13 10:10:06 -04002107
Qu Wenruo0ed47922015-04-16 16:55:08 +08002108 /* Reocrd old roots for later qgroup accounting */
2109 ret = btrfs_qgroup_prepare_account_extents(trans, root->fs_info);
2110 if (ret) {
2111 mutex_unlock(&root->fs_info->reloc_mutex);
2112 goto scrub_continue;
2113 }
2114
Chris Masone9993762011-06-17 16:14:09 -04002115 /*
2116 * make sure none of the code above managed to slip in a
2117 * delayed item
2118 */
2119 btrfs_assert_delayed_root_empty(root);
2120
Chris Mason2c90e5d2007-04-02 10:50:19 -04002121 WARN_ON(cur_trans != trans->transaction);
Chris Masondc17ff82008-01-08 15:46:30 -05002122
Chris Masone02119d2008-09-05 16:13:11 -04002123 /* btrfs_commit_tree_roots is responsible for getting the
2124 * various roots consistent with each other. Every pointer
2125 * in the tree of tree roots has to point to the most up to date
2126 * root for every subvolume and other tree. So, we have to keep
2127 * the tree logging code from jumping in and changing any
2128 * of the trees.
2129 *
2130 * At this point in the commit, there can't be any tree-log
2131 * writers, but a little lower down we drop the trans mutex
2132 * and let new people in. By holding the tree_log_mutex
2133 * from now until after the super is written, we avoid races
2134 * with the tree-log code.
2135 */
2136 mutex_lock(&root->fs_info->tree_log_mutex);
Zheng Yan1a40e232008-09-26 10:09:34 -04002137
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002138 ret = commit_fs_roots(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002139 if (ret) {
2140 mutex_unlock(&root->fs_info->tree_log_mutex);
David Sterba871383b2012-04-02 18:31:37 +02002141 mutex_unlock(&root->fs_info->reloc_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08002142 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002143 }
Chris Mason54aa1f42007-06-22 14:16:25 -04002144
Qu Wenruo3818aea2014-01-13 13:36:06 +08002145 /*
David Sterba7e1876a2014-02-05 15:26:17 +01002146 * Since the transaction is done, we can apply the pending changes
2147 * before the next transaction.
Qu Wenruo3818aea2014-01-13 13:36:06 +08002148 */
David Sterba572d9ab2014-02-05 15:26:17 +01002149 btrfs_apply_pending_changes(root->fs_info);
Qu Wenruo3818aea2014-01-13 13:36:06 +08002150
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002151 /* commit_fs_roots gets rid of all the tree log roots, it is now
Chris Masone02119d2008-09-05 16:13:11 -04002152 * safe to free the root of tree log roots
2153 */
2154 btrfs_free_log_root_tree(trans, root->fs_info);
2155
Qu Wenruo0ed47922015-04-16 16:55:08 +08002156 /*
2157 * Since fs roots are all committed, we can get a quite accurate
2158 * new_roots. So let's do quota accounting.
2159 */
2160 ret = btrfs_qgroup_account_extents(trans, root->fs_info);
2161 if (ret < 0) {
2162 mutex_unlock(&root->fs_info->tree_log_mutex);
2163 mutex_unlock(&root->fs_info->reloc_mutex);
2164 goto scrub_continue;
2165 }
2166
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002167 ret = commit_cowonly_roots(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002168 if (ret) {
2169 mutex_unlock(&root->fs_info->tree_log_mutex);
David Sterba871383b2012-04-02 18:31:37 +02002170 mutex_unlock(&root->fs_info->reloc_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08002171 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002172 }
Chris Mason54aa1f42007-06-22 14:16:25 -04002173
Miao Xie2cba30f2013-01-15 06:29:12 +00002174 /*
2175 * The tasks which save the space cache and inode cache may also
2176 * update ->aborted, check it.
2177 */
2178 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
2179 ret = cur_trans->aborted;
2180 mutex_unlock(&root->fs_info->tree_log_mutex);
2181 mutex_unlock(&root->fs_info->reloc_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08002182 goto scrub_continue;
Miao Xie2cba30f2013-01-15 06:29:12 +00002183 }
2184
Yan Zheng11833d62009-09-11 16:11:19 -04002185 btrfs_prepare_extent_commit(trans, root);
2186
Chris Mason78fae272007-03-25 11:35:08 -04002187 cur_trans = root->fs_info->running_transaction;
Chris Mason5f39d392007-10-15 16:14:19 -04002188
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002189 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
2190 root->fs_info->tree_root->node);
Josef Bacik9e351cc2014-03-13 15:42:13 -04002191 list_add_tail(&root->fs_info->tree_root->dirty_list,
2192 &cur_trans->switch_commits);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002193
2194 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
2195 root->fs_info->chunk_root->node);
Josef Bacik9e351cc2014-03-13 15:42:13 -04002196 list_add_tail(&root->fs_info->chunk_root->dirty_list,
2197 &cur_trans->switch_commits);
2198
2199 switch_commit_roots(cur_trans, root->fs_info);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002200
Jan Schmidtedf39272012-06-28 18:04:55 +02002201 assert_qgroups_uptodate(trans);
Josef Bacikce93ec52014-11-17 15:45:48 -05002202 ASSERT(list_empty(&cur_trans->dirty_bgs));
Chris Mason1bbc6212015-04-06 12:46:08 -07002203 ASSERT(list_empty(&cur_trans->io_bgs));
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002204 update_super_roots(root);
Chris Masone02119d2008-09-05 16:13:11 -04002205
Josef Bacik60e7cd32013-09-30 14:10:43 -04002206 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
2207 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
David Sterba6c417612011-04-13 15:41:04 +02002208 memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
2209 sizeof(*root->fs_info->super_copy));
Chris Masonccd467d2007-06-28 15:57:36 -04002210
Miao Xie935e5cc2014-09-03 21:35:33 +08002211 btrfs_update_commit_device_size(root->fs_info);
Miao Xiece7213c2014-09-03 21:35:34 +08002212 btrfs_update_commit_device_bytes_used(root, cur_trans);
Miao Xie935e5cc2014-09-03 21:35:33 +08002213
Josef Bacikafcdd122016-09-02 15:40:02 -04002214 clear_bit(BTRFS_FS_LOG1_ERR, &root->fs_info->flags);
2215 clear_bit(BTRFS_FS_LOG2_ERR, &root->fs_info->flags);
Filipe Manana656f30d2014-09-26 12:25:56 +01002216
Filipe Manana4fbcdf62015-05-20 14:01:54 +01002217 btrfs_trans_release_chunk_metadata(trans);
2218
Josef Bacika4abeea2011-04-11 17:25:13 -04002219 spin_lock(&root->fs_info->trans_lock);
Miao Xie4a9d8bd2013-05-17 03:53:43 +00002220 cur_trans->state = TRANS_STATE_UNBLOCKED;
Josef Bacika4abeea2011-04-11 17:25:13 -04002221 root->fs_info->running_transaction = NULL;
Josef Bacika4abeea2011-04-11 17:25:13 -04002222 spin_unlock(&root->fs_info->trans_lock);
Chris Mason75857172011-06-13 20:00:16 -04002223 mutex_unlock(&root->fs_info->reloc_mutex);
Chris Masonb7ec40d2009-03-12 20:12:45 -04002224
Chris Masonf9295742008-07-17 12:54:14 -04002225 wake_up(&root->fs_info->transaction_wait);
Chris Masone6dcd2d2008-07-17 12:53:50 -04002226
Chris Mason79154b12007-03-22 15:59:16 -04002227 ret = btrfs_write_and_wait_transaction(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002228 if (ret) {
Anand Jain34d97002016-03-16 16:43:06 +08002229 btrfs_handle_fs_error(root->fs_info, ret,
David Sterba08748812013-03-12 14:46:08 +00002230 "Error while writing out transaction");
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002231 mutex_unlock(&root->fs_info->tree_log_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08002232 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002233 }
2234
2235 ret = write_ctree_super(trans, root, 0);
2236 if (ret) {
2237 mutex_unlock(&root->fs_info->tree_log_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08002238 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01002239 }
Chris Mason4313b392008-01-03 09:08:48 -05002240
Chris Masone02119d2008-09-05 16:13:11 -04002241 /*
2242 * the super is written, we can safely allow the tree-loggers
2243 * to go about their business
2244 */
2245 mutex_unlock(&root->fs_info->tree_log_mutex);
2246
Yan Zheng11833d62009-09-11 16:11:19 -04002247 btrfs_finish_extent_commit(trans, root);
Chris Mason4313b392008-01-03 09:08:48 -05002248
Josef Bacik3204d332015-09-24 10:46:10 -04002249 if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags))
Zhao Lei13212b52015-02-12 14:18:17 +08002250 btrfs_clear_space_info_full(root->fs_info);
2251
Josef Bacik15ee9bc2007-08-10 16:22:09 -04002252 root->fs_info->last_trans_committed = cur_trans->transid;
Miao Xie4a9d8bd2013-05-17 03:53:43 +00002253 /*
2254 * We needn't acquire the lock here because there is no other task
2255 * which can change it.
2256 */
2257 cur_trans->state = TRANS_STATE_COMPLETED;
Chris Mason2c90e5d2007-04-02 10:50:19 -04002258 wake_up(&cur_trans->commit_wait);
Chris Mason3de45862008-11-17 21:02:50 -05002259
Josef Bacika4abeea2011-04-11 17:25:13 -04002260 spin_lock(&root->fs_info->trans_lock);
Josef Bacik13c5a932011-04-11 15:45:29 -04002261 list_del_init(&cur_trans->list);
Josef Bacika4abeea2011-04-11 17:25:13 -04002262 spin_unlock(&root->fs_info->trans_lock);
2263
Josef Bacik724e2312013-09-30 11:36:38 -04002264 btrfs_put_transaction(cur_trans);
2265 btrfs_put_transaction(cur_trans);
Josef Bacik58176a92007-08-29 15:47:34 -04002266
Miao Xie0860adf2013-05-15 07:48:27 +00002267 if (trans->type & __TRANS_FREEZABLE)
Miao Xie354aa0f2012-09-20 01:54:00 -06002268 sb_end_intwrite(root->fs_info->sb);
Jan Karab2b5ef52012-06-12 16:20:45 +02002269
liubo1abe9b82011-03-24 11:18:59 +00002270 trace_btrfs_transaction_commit(root);
2271
Arne Jansena2de7332011-03-08 14:14:00 +01002272 btrfs_scrub_continue(root);
2273
Josef Bacik9ed74f22009-09-11 16:12:44 -04002274 if (current->journal_info == trans)
2275 current->journal_info = NULL;
2276
Chris Mason2c90e5d2007-04-02 10:50:19 -04002277 kmem_cache_free(btrfs_trans_handle_cachep, trans);
Yan, Zheng24bbcf02009-11-12 09:36:34 +00002278
Wang Xiaoguang9e7cc912016-08-01 13:28:08 +08002279 /*
2280 * If fs has been frozen, we can not handle delayed iputs, otherwise
2281 * it'll result in deadlock about SB_FREEZE_FS.
2282 */
Zhao Lei8a733012015-07-15 11:48:14 +08002283 if (current != root->fs_info->transaction_kthread &&
Wang Xiaoguang9e7cc912016-08-01 13:28:08 +08002284 current != root->fs_info->cleaner_kthread &&
2285 !root->fs_info->fs_frozen)
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}