blob: e4bfac8d54b8827800fde80cfe84ba94ed96060a [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"
Chris Mason79154b12007-03-22 15:59:16 -040033
Chris Mason0f7d52f2007-04-09 10:42:37 -040034#define BTRFS_ROOT_TRANS_TAG 0
35
Jeff Mahoney49b25e02012-03-01 17:24:58 +010036void put_transaction(struct btrfs_transaction *transaction)
Chris Mason79154b12007-03-22 15:59:16 -040037{
Josef Bacik13c5a932011-04-11 15:45:29 -040038 WARN_ON(atomic_read(&transaction->use_count) == 0);
39 if (atomic_dec_and_test(&transaction->use_count)) {
Josef Bacika4abeea2011-04-11 17:25:13 -040040 BUG_ON(!list_empty(&transaction->list));
Arne Jansen00f04b82011-09-14 12:37:00 +020041 WARN_ON(transaction->delayed_refs.root.rb_node);
Chris Mason2c90e5d2007-04-02 10:50:19 -040042 memset(transaction, 0, sizeof(*transaction));
43 kmem_cache_free(btrfs_transaction_cachep, transaction);
Chris Mason78fae272007-03-25 11:35:08 -040044 }
Chris Mason79154b12007-03-22 15:59:16 -040045}
46
Josef Bacik817d52f2009-07-13 21:29:25 -040047static noinline void switch_commit_root(struct btrfs_root *root)
48{
Josef Bacik817d52f2009-07-13 21:29:25 -040049 free_extent_buffer(root->commit_root);
50 root->commit_root = btrfs_root_node(root);
Josef Bacik817d52f2009-07-13 21:29:25 -040051}
52
Chris Masond352ac62008-09-29 15:18:18 -040053/*
54 * either allocate a new transaction or hop into the existing one
55 */
Josef Bacika4abeea2011-04-11 17:25:13 -040056static noinline int join_transaction(struct btrfs_root *root, int nofail)
Chris Mason79154b12007-03-22 15:59:16 -040057{
58 struct btrfs_transaction *cur_trans;
Jan Schmidt19ae4e82012-05-20 15:42:19 +020059 struct btrfs_fs_info *fs_info = root->fs_info;
Josef Bacika4abeea2011-04-11 17:25:13 -040060
Jan Schmidt19ae4e82012-05-20 15:42:19 +020061 spin_lock(&fs_info->trans_lock);
Chris Masond43317d2011-11-06 03:26:19 -050062loop:
Jeff Mahoney49b25e02012-03-01 17:24:58 +010063 /* The file system has been taken offline. No new transactions. */
Jan Schmidt19ae4e82012-05-20 15:42:19 +020064 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
65 spin_unlock(&fs_info->trans_lock);
Jeff Mahoney49b25e02012-03-01 17:24:58 +010066 return -EROFS;
67 }
68
Jan Schmidt19ae4e82012-05-20 15:42:19 +020069 if (fs_info->trans_no_join) {
Josef Bacika4abeea2011-04-11 17:25:13 -040070 if (!nofail) {
Jan Schmidt19ae4e82012-05-20 15:42:19 +020071 spin_unlock(&fs_info->trans_lock);
Josef Bacika4abeea2011-04-11 17:25:13 -040072 return -EBUSY;
73 }
74 }
75
Jan Schmidt19ae4e82012-05-20 15:42:19 +020076 cur_trans = fs_info->running_transaction;
Josef Bacika4abeea2011-04-11 17:25:13 -040077 if (cur_trans) {
David Sterba871383b2012-04-02 18:31:37 +020078 if (cur_trans->aborted) {
Jan Schmidt19ae4e82012-05-20 15:42:19 +020079 spin_unlock(&fs_info->trans_lock);
Jeff Mahoney49b25e02012-03-01 17:24:58 +010080 return cur_trans->aborted;
David Sterba871383b2012-04-02 18:31:37 +020081 }
Josef Bacika4abeea2011-04-11 17:25:13 -040082 atomic_inc(&cur_trans->use_count);
Josef Bacik13c5a932011-04-11 15:45:29 -040083 atomic_inc(&cur_trans->num_writers);
Josef Bacik15ee9bc2007-08-10 16:22:09 -040084 cur_trans->num_joined++;
Jan Schmidt19ae4e82012-05-20 15:42:19 +020085 spin_unlock(&fs_info->trans_lock);
Josef Bacika4abeea2011-04-11 17:25:13 -040086 return 0;
Chris Mason79154b12007-03-22 15:59:16 -040087 }
Jan Schmidt19ae4e82012-05-20 15:42:19 +020088 spin_unlock(&fs_info->trans_lock);
Josef Bacika4abeea2011-04-11 17:25:13 -040089
90 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS);
91 if (!cur_trans)
92 return -ENOMEM;
Chris Masond43317d2011-11-06 03:26:19 -050093
Jan Schmidt19ae4e82012-05-20 15:42:19 +020094 spin_lock(&fs_info->trans_lock);
95 if (fs_info->running_transaction) {
Chris Masond43317d2011-11-06 03:26:19 -050096 /*
97 * someone started a transaction after we unlocked. Make sure
98 * to redo the trans_no_join checks above
99 */
Josef Bacika4abeea2011-04-11 17:25:13 -0400100 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200101 cur_trans = fs_info->running_transaction;
Chris Masond43317d2011-11-06 03:26:19 -0500102 goto loop;
Dan Carpentere4b50e12012-06-19 13:30:11 +0300103 } else if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
104 spin_unlock(&fs_info->trans_lock);
Josef Bacik7b8b92a2012-05-31 15:52:43 -0400105 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
106 return -EROFS;
Josef Bacika4abeea2011-04-11 17:25:13 -0400107 }
Chris Masond43317d2011-11-06 03:26:19 -0500108
Josef Bacika4abeea2011-04-11 17:25:13 -0400109 atomic_set(&cur_trans->num_writers, 1);
110 cur_trans->num_joined = 0;
111 init_waitqueue_head(&cur_trans->writer_wait);
112 init_waitqueue_head(&cur_trans->commit_wait);
113 cur_trans->in_commit = 0;
114 cur_trans->blocked = 0;
115 /*
116 * One for this trans handle, one so it will live on until we
117 * commit the transaction.
118 */
119 atomic_set(&cur_trans->use_count, 2);
120 cur_trans->commit_done = 0;
121 cur_trans->start_time = get_seconds();
122
123 cur_trans->delayed_refs.root = RB_ROOT;
124 cur_trans->delayed_refs.num_entries = 0;
125 cur_trans->delayed_refs.num_heads_ready = 0;
126 cur_trans->delayed_refs.num_heads = 0;
127 cur_trans->delayed_refs.flushing = 0;
128 cur_trans->delayed_refs.run_delayed_start = 0;
Jan Schmidt20b297d2012-05-20 15:43:53 +0200129
130 /*
131 * although the tree mod log is per file system and not per transaction,
132 * the log must never go across transaction boundaries.
133 */
134 smp_mb();
135 if (!list_empty(&fs_info->tree_mod_seq_list)) {
136 printk(KERN_ERR "btrfs: tree_mod_seq_list not empty when "
137 "creating a fresh transaction\n");
138 WARN_ON(1);
139 }
140 if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) {
141 printk(KERN_ERR "btrfs: tree_mod_log rb tree not empty when "
142 "creating a fresh transaction\n");
143 WARN_ON(1);
144 }
145 atomic_set(&fs_info->tree_mod_seq, 0);
146
Josef Bacika4abeea2011-04-11 17:25:13 -0400147 spin_lock_init(&cur_trans->commit_lock);
148 spin_lock_init(&cur_trans->delayed_refs.lock);
149
150 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200151 list_add_tail(&cur_trans->list, &fs_info->trans_list);
Josef Bacika4abeea2011-04-11 17:25:13 -0400152 extent_io_tree_init(&cur_trans->dirty_pages,
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200153 fs_info->btree_inode->i_mapping);
154 fs_info->generation++;
155 cur_trans->transid = fs_info->generation;
156 fs_info->running_transaction = cur_trans;
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100157 cur_trans->aborted = 0;
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200158 spin_unlock(&fs_info->trans_lock);
Josef Bacik15ee9bc2007-08-10 16:22:09 -0400159
Chris Mason79154b12007-03-22 15:59:16 -0400160 return 0;
161}
162
Chris Masond352ac62008-09-29 15:18:18 -0400163/*
Chris Masond3977122009-01-05 21:25:51 -0500164 * this does all the record keeping required to make sure that a reference
165 * counted root is properly recorded in a given transaction. This is required
166 * to make sure the old root from before we joined the transaction is deleted
167 * when the transaction commits
Chris Masond352ac62008-09-29 15:18:18 -0400168 */
Chris Mason75857172011-06-13 20:00:16 -0400169static int record_root_in_trans(struct btrfs_trans_handle *trans,
Josef Bacika4abeea2011-04-11 17:25:13 -0400170 struct btrfs_root *root)
Chris Mason6702ed42007-08-07 16:15:09 -0400171{
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400172 if (root->ref_cows && root->last_trans < trans->transid) {
Chris Mason6702ed42007-08-07 16:15:09 -0400173 WARN_ON(root == root->fs_info->extent_root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400174 WARN_ON(root->commit_root != root->node);
Yan Zheng31153d82008-07-28 15:32:19 -0400175
Chris Mason75857172011-06-13 20:00:16 -0400176 /*
177 * see below for in_trans_setup usage rules
178 * we have the reloc mutex held now, so there
179 * is only one writer in this function
180 */
181 root->in_trans_setup = 1;
182
183 /* make sure readers find in_trans_setup before
184 * they find our root->last_trans update
185 */
186 smp_wmb();
187
Josef Bacika4abeea2011-04-11 17:25:13 -0400188 spin_lock(&root->fs_info->fs_roots_radix_lock);
189 if (root->last_trans == trans->transid) {
190 spin_unlock(&root->fs_info->fs_roots_radix_lock);
191 return 0;
192 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400193 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
194 (unsigned long)root->root_key.objectid,
195 BTRFS_ROOT_TRANS_TAG);
Josef Bacika4abeea2011-04-11 17:25:13 -0400196 spin_unlock(&root->fs_info->fs_roots_radix_lock);
Chris Mason75857172011-06-13 20:00:16 -0400197 root->last_trans = trans->transid;
198
199 /* this is pretty tricky. We don't want to
200 * take the relocation lock in btrfs_record_root_in_trans
201 * unless we're really doing the first setup for this root in
202 * this transaction.
203 *
204 * Normally we'd use root->last_trans as a flag to decide
205 * if we want to take the expensive mutex.
206 *
207 * But, we have to set root->last_trans before we
208 * init the relocation root, otherwise, we trip over warnings
209 * in ctree.c. The solution used here is to flag ourselves
210 * with root->in_trans_setup. When this is 1, we're still
211 * fixing up the reloc trees and everyone must wait.
212 *
213 * When this is zero, they can trust root->last_trans and fly
214 * through btrfs_record_root_in_trans without having to take the
215 * lock. smp_wmb() makes sure that all the writes above are
216 * done before we pop in the zero below
217 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400218 btrfs_init_reloc_root(trans, root);
Chris Mason75857172011-06-13 20:00:16 -0400219 smp_wmb();
220 root->in_trans_setup = 0;
Chris Mason6702ed42007-08-07 16:15:09 -0400221 }
222 return 0;
223}
224
Chris Mason75857172011-06-13 20:00:16 -0400225
226int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
227 struct btrfs_root *root)
228{
229 if (!root->ref_cows)
230 return 0;
231
232 /*
233 * see record_root_in_trans for comments about in_trans_setup usage
234 * and barriers
235 */
236 smp_rmb();
237 if (root->last_trans == trans->transid &&
238 !root->in_trans_setup)
239 return 0;
240
241 mutex_lock(&root->fs_info->reloc_mutex);
242 record_root_in_trans(trans, root);
243 mutex_unlock(&root->fs_info->reloc_mutex);
244
245 return 0;
246}
247
Chris Masond352ac62008-09-29 15:18:18 -0400248/* wait for commit against the current transaction to become unblocked
249 * when this is done, it is safe to start a new transaction, but the current
250 * transaction might not be fully on disk.
251 */
Chris Mason37d1aee2008-07-31 10:48:37 -0400252static void wait_current_trans(struct btrfs_root *root)
Chris Mason79154b12007-03-22 15:59:16 -0400253{
Chris Masonf9295742008-07-17 12:54:14 -0400254 struct btrfs_transaction *cur_trans;
Chris Mason79154b12007-03-22 15:59:16 -0400255
Josef Bacika4abeea2011-04-11 17:25:13 -0400256 spin_lock(&root->fs_info->trans_lock);
Chris Masonf9295742008-07-17 12:54:14 -0400257 cur_trans = root->fs_info->running_transaction;
Chris Mason37d1aee2008-07-31 10:48:37 -0400258 if (cur_trans && cur_trans->blocked) {
Josef Bacik13c5a932011-04-11 15:45:29 -0400259 atomic_inc(&cur_trans->use_count);
Josef Bacika4abeea2011-04-11 17:25:13 -0400260 spin_unlock(&root->fs_info->trans_lock);
Li Zefan72d63ed2011-07-14 03:17:00 +0000261
262 wait_event(root->fs_info->transaction_wait,
263 !cur_trans->blocked);
Chris Masonf9295742008-07-17 12:54:14 -0400264 put_transaction(cur_trans);
Josef Bacika4abeea2011-04-11 17:25:13 -0400265 } else {
266 spin_unlock(&root->fs_info->trans_lock);
Chris Masonf9295742008-07-17 12:54:14 -0400267 }
Chris Mason37d1aee2008-07-31 10:48:37 -0400268}
269
Josef Bacik249ac1e2009-11-10 21:23:48 -0500270enum btrfs_trans_type {
271 TRANS_START,
272 TRANS_JOIN,
273 TRANS_USERSPACE,
Josef Bacik0af3d002010-06-21 14:48:16 -0400274 TRANS_JOIN_NOLOCK,
Josef Bacik60376ce2012-09-14 10:34:40 -0400275 TRANS_JOIN_FREEZE,
Josef Bacik249ac1e2009-11-10 21:23:48 -0500276};
277
Yan, Zhenga22285a2010-05-16 10:48:46 -0400278static int may_wait_transaction(struct btrfs_root *root, int type)
Chris Mason37d1aee2008-07-31 10:48:37 -0400279{
Josef Bacika4abeea2011-04-11 17:25:13 -0400280 if (root->fs_info->log_root_recovering)
281 return 0;
282
283 if (type == TRANS_USERSPACE)
Yan, Zhenga22285a2010-05-16 10:48:46 -0400284 return 1;
Josef Bacika4abeea2011-04-11 17:25:13 -0400285
286 if (type == TRANS_START &&
287 !atomic_read(&root->fs_info->open_ioctl_trans))
288 return 1;
289
Yan, Zhenga22285a2010-05-16 10:48:46 -0400290 return 0;
291}
292
293static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
Miao Xie8407aa42012-09-07 01:43:32 -0600294 u64 num_items, int type,
295 int noflush)
Yan, Zhenga22285a2010-05-16 10:48:46 -0400296{
297 struct btrfs_trans_handle *h;
298 struct btrfs_transaction *cur_trans;
Josef Bacikb5009942011-06-07 15:07:51 -0400299 u64 num_bytes = 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400300 int ret;
Arne Jansenc5567232011-09-14 15:44:05 +0200301 u64 qgroup_reserved = 0;
liuboacce9522011-01-06 19:30:25 +0800302
303 if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
304 return ERR_PTR(-EROFS);
Josef Bacik2a1eb462011-04-13 15:15:59 -0400305
306 if (current->journal_info) {
307 WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK);
308 h = current->journal_info;
309 h->use_count++;
310 h->orig_rsv = h->block_rsv;
311 h->block_rsv = NULL;
312 goto got_it;
313 }
Josef Bacikb5009942011-06-07 15:07:51 -0400314
315 /*
316 * Do the reservation before we join the transaction so we can do all
317 * the appropriate flushing if need be.
318 */
319 if (num_items > 0 && root != root->fs_info->chunk_root) {
Arne Jansenc5567232011-09-14 15:44:05 +0200320 if (root->fs_info->quota_enabled &&
321 is_fstree(root->root_key.objectid)) {
322 qgroup_reserved = num_items * root->leafsize;
323 ret = btrfs_qgroup_reserve(root, qgroup_reserved);
324 if (ret)
325 return ERR_PTR(ret);
326 }
327
Josef Bacikb5009942011-06-07 15:07:51 -0400328 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
Miao Xie8407aa42012-09-07 01:43:32 -0600329 if (noflush)
330 ret = btrfs_block_rsv_add_noflush(root,
331 &root->fs_info->trans_block_rsv,
332 num_bytes);
333 else
334 ret = btrfs_block_rsv_add(root,
335 &root->fs_info->trans_block_rsv,
336 num_bytes);
Josef Bacikb5009942011-06-07 15:07:51 -0400337 if (ret)
338 return ERR_PTR(ret);
339 }
Yan, Zhenga22285a2010-05-16 10:48:46 -0400340again:
341 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
342 if (!h)
343 return ERR_PTR(-ENOMEM);
344
Josef Bacik60376ce2012-09-14 10:34:40 -0400345 if (!__sb_start_write(root->fs_info->sb, SB_FREEZE_FS, false)) {
346 if (type == TRANS_JOIN_FREEZE)
347 return ERR_PTR(-EPERM);
348 sb_start_intwrite(root->fs_info->sb);
349 }
Jan Karab2b5ef52012-06-12 16:20:45 +0200350
Yan, Zhenga22285a2010-05-16 10:48:46 -0400351 if (may_wait_transaction(root, type))
Chris Mason37d1aee2008-07-31 10:48:37 -0400352 wait_current_trans(root);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400353
Josef Bacika4abeea2011-04-11 17:25:13 -0400354 do {
355 ret = join_transaction(root, type == TRANS_JOIN_NOLOCK);
356 if (ret == -EBUSY)
357 wait_current_trans(root);
358 } while (ret == -EBUSY);
359
Tsutomu Itohdb5b4932011-03-23 08:14:16 +0000360 if (ret < 0) {
Jan Karab2b5ef52012-06-12 16:20:45 +0200361 sb_end_intwrite(root->fs_info->sb);
Yoshinori Sano6e8df2a2011-04-03 12:31:28 +0000362 kmem_cache_free(btrfs_trans_handle_cachep, h);
Tsutomu Itohdb5b4932011-03-23 08:14:16 +0000363 return ERR_PTR(ret);
364 }
Chris Mason0f7d52f2007-04-09 10:42:37 -0400365
Yan, Zhenga22285a2010-05-16 10:48:46 -0400366 cur_trans = root->fs_info->running_transaction;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400367
368 h->transid = cur_trans->transid;
369 h->transaction = cur_trans;
Chris Mason79154b12007-03-22 15:59:16 -0400370 h->blocks_used = 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400371 h->bytes_reserved = 0;
Arne Jansend13603e2011-09-13 11:40:09 +0200372 h->root = root;
Chris Mason56bec292009-03-13 10:10:06 -0400373 h->delayed_ref_updates = 0;
Josef Bacik2a1eb462011-04-13 15:15:59 -0400374 h->use_count = 1;
Josef Bacik0e721102012-06-26 16:13:18 -0400375 h->adding_csums = 0;
Yan, Zhengf0486c62010-05-16 10:46:25 -0400376 h->block_rsv = NULL;
Josef Bacik2a1eb462011-04-13 15:15:59 -0400377 h->orig_rsv = NULL;
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100378 h->aborted = 0;
Arne Jansenc5567232011-09-14 15:44:05 +0200379 h->qgroup_reserved = qgroup_reserved;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200380 h->delayed_ref_elem.seq = 0;
381 INIT_LIST_HEAD(&h->qgroup_ref_list);
Josef Bacikea658ba2012-09-11 16:57:25 -0400382 INIT_LIST_HEAD(&h->new_bgs);
Chris Masonb7ec40d2009-03-12 20:12:45 -0400383
Yan, Zhenga22285a2010-05-16 10:48:46 -0400384 smp_mb();
385 if (cur_trans->blocked && may_wait_transaction(root, type)) {
386 btrfs_commit_transaction(h, root);
387 goto again;
388 }
Josef Bacik9ed74f22009-09-11 16:12:44 -0400389
Josef Bacikb5009942011-06-07 15:07:51 -0400390 if (num_bytes) {
Josef Bacik8c2a3ca2012-01-10 10:31:31 -0500391 trace_btrfs_space_reservation(root->fs_info, "transaction",
Liu Bo2bcc0322012-03-29 09:57:44 -0400392 h->transid, num_bytes, 1);
Josef Bacikb5009942011-06-07 15:07:51 -0400393 h->block_rsv = &root->fs_info->trans_block_rsv;
394 h->bytes_reserved = num_bytes;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400395 }
396
Josef Bacik2a1eb462011-04-13 15:15:59 -0400397got_it:
Josef Bacika4abeea2011-04-11 17:25:13 -0400398 btrfs_record_root_in_trans(h, root);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400399
400 if (!current->journal_info && type != TRANS_USERSPACE)
401 current->journal_info = h;
Chris Mason79154b12007-03-22 15:59:16 -0400402 return h;
403}
404
Chris Masonf9295742008-07-17 12:54:14 -0400405struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
Yan, Zhenga22285a2010-05-16 10:48:46 -0400406 int num_items)
Chris Masonf9295742008-07-17 12:54:14 -0400407{
Miao Xie8407aa42012-09-07 01:43:32 -0600408 return start_transaction(root, num_items, TRANS_START, 0);
Chris Masonf9295742008-07-17 12:54:14 -0400409}
Miao Xie8407aa42012-09-07 01:43:32 -0600410
411struct btrfs_trans_handle *btrfs_start_transaction_noflush(
412 struct btrfs_root *root, int num_items)
413{
414 return start_transaction(root, num_items, TRANS_START, 1);
415}
416
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400417struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
Chris Masonf9295742008-07-17 12:54:14 -0400418{
Miao Xie8407aa42012-09-07 01:43:32 -0600419 return start_transaction(root, 0, TRANS_JOIN, 0);
Chris Masonf9295742008-07-17 12:54:14 -0400420}
421
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400422struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
Josef Bacik0af3d002010-06-21 14:48:16 -0400423{
Miao Xie8407aa42012-09-07 01:43:32 -0600424 return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0);
Josef Bacik0af3d002010-06-21 14:48:16 -0400425}
426
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400427struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
Sage Weil9ca9ee02008-08-04 10:41:27 -0400428{
Miao Xie8407aa42012-09-07 01:43:32 -0600429 return start_transaction(root, 0, TRANS_USERSPACE, 0);
Sage Weil9ca9ee02008-08-04 10:41:27 -0400430}
431
Josef Bacik60376ce2012-09-14 10:34:40 -0400432struct btrfs_trans_handle *btrfs_join_transaction_freeze(struct btrfs_root *root)
433{
434 return start_transaction(root, 0, TRANS_JOIN_FREEZE, 0);
435}
436
Chris Masond352ac62008-09-29 15:18:18 -0400437/* wait for a transaction commit to be fully complete */
Li Zefanb9c83002011-07-14 03:17:14 +0000438static noinline void wait_for_commit(struct btrfs_root *root,
Chris Mason89ce8a62008-06-25 16:01:31 -0400439 struct btrfs_transaction *commit)
440{
Li Zefan72d63ed2011-07-14 03:17:00 +0000441 wait_event(commit->commit_wait, commit->commit_done);
Chris Mason89ce8a62008-06-25 16:01:31 -0400442}
443
Sage Weil46204592010-10-29 15:41:32 -0400444int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
445{
446 struct btrfs_transaction *cur_trans = NULL, *t;
447 int ret;
448
Sage Weil46204592010-10-29 15:41:32 -0400449 ret = 0;
450 if (transid) {
451 if (transid <= root->fs_info->last_trans_committed)
Josef Bacika4abeea2011-04-11 17:25:13 -0400452 goto out;
Sage Weil46204592010-10-29 15:41:32 -0400453
454 /* find specified transaction */
Josef Bacika4abeea2011-04-11 17:25:13 -0400455 spin_lock(&root->fs_info->trans_lock);
Sage Weil46204592010-10-29 15:41:32 -0400456 list_for_each_entry(t, &root->fs_info->trans_list, list) {
457 if (t->transid == transid) {
458 cur_trans = t;
Josef Bacika4abeea2011-04-11 17:25:13 -0400459 atomic_inc(&cur_trans->use_count);
Sage Weil46204592010-10-29 15:41:32 -0400460 break;
461 }
462 if (t->transid > transid)
463 break;
464 }
Josef Bacika4abeea2011-04-11 17:25:13 -0400465 spin_unlock(&root->fs_info->trans_lock);
Sage Weil46204592010-10-29 15:41:32 -0400466 ret = -EINVAL;
467 if (!cur_trans)
Josef Bacika4abeea2011-04-11 17:25:13 -0400468 goto out; /* bad transid */
Sage Weil46204592010-10-29 15:41:32 -0400469 } else {
470 /* find newest transaction that is committing | committed */
Josef Bacika4abeea2011-04-11 17:25:13 -0400471 spin_lock(&root->fs_info->trans_lock);
Sage Weil46204592010-10-29 15:41:32 -0400472 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
473 list) {
474 if (t->in_commit) {
475 if (t->commit_done)
Josef Bacik3473f3c2011-06-09 10:15:17 -0400476 break;
Sage Weil46204592010-10-29 15:41:32 -0400477 cur_trans = t;
Josef Bacika4abeea2011-04-11 17:25:13 -0400478 atomic_inc(&cur_trans->use_count);
Sage Weil46204592010-10-29 15:41:32 -0400479 break;
480 }
481 }
Josef Bacika4abeea2011-04-11 17:25:13 -0400482 spin_unlock(&root->fs_info->trans_lock);
Sage Weil46204592010-10-29 15:41:32 -0400483 if (!cur_trans)
Josef Bacika4abeea2011-04-11 17:25:13 -0400484 goto out; /* nothing committing|committed */
Sage Weil46204592010-10-29 15:41:32 -0400485 }
486
Sage Weil46204592010-10-29 15:41:32 -0400487 wait_for_commit(root, cur_trans);
488
Sage Weil46204592010-10-29 15:41:32 -0400489 put_transaction(cur_trans);
490 ret = 0;
Josef Bacika4abeea2011-04-11 17:25:13 -0400491out:
Sage Weil46204592010-10-29 15:41:32 -0400492 return ret;
493}
494
Chris Mason37d1aee2008-07-31 10:48:37 -0400495void btrfs_throttle(struct btrfs_root *root)
496{
Josef Bacika4abeea2011-04-11 17:25:13 -0400497 if (!atomic_read(&root->fs_info->open_ioctl_trans))
Sage Weil9ca9ee02008-08-04 10:41:27 -0400498 wait_current_trans(root);
Chris Mason37d1aee2008-07-31 10:48:37 -0400499}
500
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400501static int should_end_transaction(struct btrfs_trans_handle *trans,
502 struct btrfs_root *root)
503{
504 int ret;
Josef Bacik36ba0222011-10-18 12:15:48 -0400505
506 ret = btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400507 return ret ? 1 : 0;
508}
509
510int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
511 struct btrfs_root *root)
512{
513 struct btrfs_transaction *cur_trans = trans->transaction;
514 int updates;
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100515 int err;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400516
Josef Bacika4abeea2011-04-11 17:25:13 -0400517 smp_mb();
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400518 if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
519 return 1;
520
521 updates = trans->delayed_ref_updates;
522 trans->delayed_ref_updates = 0;
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100523 if (updates) {
524 err = btrfs_run_delayed_refs(trans, root, updates);
525 if (err) /* Error code will also eval true */
526 return err;
527 }
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400528
529 return should_end_transaction(trans, root);
530}
531
Chris Mason89ce8a62008-06-25 16:01:31 -0400532static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
Josef Bacik0af3d002010-06-21 14:48:16 -0400533 struct btrfs_root *root, int throttle, int lock)
Chris Mason79154b12007-03-22 15:59:16 -0400534{
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400535 struct btrfs_transaction *cur_trans = trans->transaction;
Chris Masonab78c842008-07-29 16:15:18 -0400536 struct btrfs_fs_info *info = root->fs_info;
Chris Masonc3e69d52009-03-13 10:17:05 -0400537 int count = 0;
Dave Jones4edc2ca2012-04-12 16:03:56 -0400538 int err = 0;
Chris Masond6e4a422007-04-06 15:37:36 -0400539
Josef Bacik2a1eb462011-04-13 15:15:59 -0400540 if (--trans->use_count) {
541 trans->block_rsv = trans->orig_rsv;
542 return 0;
543 }
544
Jan Schmidtedf39272012-06-28 18:04:55 +0200545 /*
546 * do the qgroup accounting as early as possible
547 */
548 err = btrfs_delayed_refs_qgroup_accounting(trans, info);
549
Josef Bacikb24e03d2011-10-14 14:40:17 -0400550 btrfs_trans_release_metadata(trans, root);
Josef Bacik4c13d752011-08-30 11:31:29 -0400551 trans->block_rsv = NULL;
Arne Jansend13603e2011-09-13 11:40:09 +0200552 /*
553 * the same root has to be passed to start_transaction and
554 * end_transaction. Subvolume quota depends on this.
555 */
556 WARN_ON(trans->root != root);
Arne Jansenc5567232011-09-14 15:44:05 +0200557
558 if (trans->qgroup_reserved) {
559 btrfs_qgroup_free(root, trans->qgroup_reserved);
560 trans->qgroup_reserved = 0;
561 }
562
Josef Bacikea658ba2012-09-11 16:57:25 -0400563 if (!list_empty(&trans->new_bgs))
564 btrfs_create_pending_block_groups(trans, root);
565
Chris Mason203bf282012-01-06 15:23:57 -0500566 while (count < 2) {
Chris Masonc3e69d52009-03-13 10:17:05 -0400567 unsigned long cur = trans->delayed_ref_updates;
568 trans->delayed_ref_updates = 0;
569 if (cur &&
570 trans->transaction->delayed_refs.num_heads_ready > 64) {
571 trans->delayed_ref_updates = 0;
572 btrfs_run_delayed_refs(trans, root, cur);
573 } else {
574 break;
575 }
576 count++;
Chris Mason56bec292009-03-13 10:10:06 -0400577 }
Josef Bacik0e721102012-06-26 16:13:18 -0400578 btrfs_trans_release_metadata(trans, root);
579 trans->block_rsv = NULL;
Chris Mason56bec292009-03-13 10:10:06 -0400580
Josef Bacikea658ba2012-09-11 16:57:25 -0400581 if (!list_empty(&trans->new_bgs))
582 btrfs_create_pending_block_groups(trans, root);
583
Josef Bacika4abeea2011-04-11 17:25:13 -0400584 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
585 should_end_transaction(trans, root)) {
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400586 trans->transaction->blocked = 1;
Josef Bacika4abeea2011-04-11 17:25:13 -0400587 smp_wmb();
588 }
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400589
Josef Bacik0af3d002010-06-21 14:48:16 -0400590 if (lock && cur_trans->blocked && !cur_trans->in_commit) {
Josef Bacik81317fd2011-07-24 15:45:34 -0400591 if (throttle) {
592 /*
593 * We may race with somebody else here so end up having
594 * to call end_transaction on ourselves again, so inc
595 * our use_count.
596 */
597 trans->use_count++;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400598 return btrfs_commit_transaction(trans, root);
Josef Bacik81317fd2011-07-24 15:45:34 -0400599 } else {
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400600 wake_up_process(info->transaction_kthread);
Josef Bacik81317fd2011-07-24 15:45:34 -0400601 }
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400602 }
603
Josef Bacik6df78812012-09-05 08:08:30 -0600604 sb_end_intwrite(root->fs_info->sb);
605
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400606 WARN_ON(cur_trans != info->running_transaction);
Josef Bacik13c5a932011-04-11 15:45:29 -0400607 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
608 atomic_dec(&cur_trans->num_writers);
Chris Mason89ce8a62008-06-25 16:01:31 -0400609
Sage Weil99d16cb2010-10-29 15:37:34 -0400610 smp_mb();
Chris Mason79154b12007-03-22 15:59:16 -0400611 if (waitqueue_active(&cur_trans->writer_wait))
612 wake_up(&cur_trans->writer_wait);
Chris Mason79154b12007-03-22 15:59:16 -0400613 put_transaction(cur_trans);
Josef Bacik9ed74f22009-09-11 16:12:44 -0400614
615 if (current->journal_info == trans)
616 current->journal_info = NULL;
Chris Masonab78c842008-07-29 16:15:18 -0400617
Yan, Zheng24bbcf02009-11-12 09:36:34 +0000618 if (throttle)
619 btrfs_run_delayed_iputs(root);
620
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100621 if (trans->aborted ||
622 root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
Dave Jones4edc2ca2012-04-12 16:03:56 -0400623 err = -EIO;
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100624 }
Jan Schmidtedf39272012-06-28 18:04:55 +0200625 assert_qgroups_uptodate(trans);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100626
Dave Jones4edc2ca2012-04-12 16:03:56 -0400627 memset(trans, 0, sizeof(*trans));
628 kmem_cache_free(btrfs_trans_handle_cachep, trans);
629 return err;
Chris Mason79154b12007-03-22 15:59:16 -0400630}
631
Chris Mason89ce8a62008-06-25 16:01:31 -0400632int btrfs_end_transaction(struct btrfs_trans_handle *trans,
633 struct btrfs_root *root)
634{
Miao Xie16cdcec2011-04-22 18:12:22 +0800635 int ret;
636
637 ret = __btrfs_end_transaction(trans, root, 0, 1);
638 if (ret)
639 return ret;
640 return 0;
Chris Mason89ce8a62008-06-25 16:01:31 -0400641}
642
643int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
644 struct btrfs_root *root)
645{
Miao Xie16cdcec2011-04-22 18:12:22 +0800646 int ret;
647
648 ret = __btrfs_end_transaction(trans, root, 1, 1);
649 if (ret)
650 return ret;
651 return 0;
Josef Bacik0af3d002010-06-21 14:48:16 -0400652}
653
654int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans,
655 struct btrfs_root *root)
656{
Miao Xie16cdcec2011-04-22 18:12:22 +0800657 int ret;
658
659 ret = __btrfs_end_transaction(trans, root, 0, 0);
660 if (ret)
661 return ret;
662 return 0;
663}
664
665int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans,
666 struct btrfs_root *root)
667{
668 return __btrfs_end_transaction(trans, root, 1, 1);
Chris Mason89ce8a62008-06-25 16:01:31 -0400669}
670
Chris Masond352ac62008-09-29 15:18:18 -0400671/*
672 * when btree blocks are allocated, they have some corresponding bits set for
673 * them in one of two extent_io trees. This is used to make sure all of
Chris Mason690587d2009-10-13 13:29:19 -0400674 * those extents are sent to disk but does not wait on them
Chris Masond352ac62008-09-29 15:18:18 -0400675 */
Chris Mason690587d2009-10-13 13:29:19 -0400676int btrfs_write_marked_extents(struct btrfs_root *root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000677 struct extent_io_tree *dirty_pages, int mark)
Chris Mason79154b12007-03-22 15:59:16 -0400678{
Chris Mason777e6bd2008-08-15 15:34:15 -0400679 int err = 0;
Chris Mason7c4452b2007-04-28 09:29:35 -0400680 int werr = 0;
Josef Bacik17283662011-09-26 13:58:47 -0400681 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
Chris Mason777e6bd2008-08-15 15:34:15 -0400682 u64 start = 0;
Chris Mason5f39d392007-10-15 16:14:19 -0400683 u64 end;
Chris Mason7c4452b2007-04-28 09:29:35 -0400684
Josef Bacik17283662011-09-26 13:58:47 -0400685 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
686 mark)) {
687 convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, mark,
688 GFP_NOFS);
689 err = filemap_fdatawrite_range(mapping, start, end);
690 if (err)
691 werr = err;
692 cond_resched();
693 start = end + 1;
Chris Mason7c4452b2007-04-28 09:29:35 -0400694 }
Chris Mason690587d2009-10-13 13:29:19 -0400695 if (err)
696 werr = err;
697 return werr;
698}
699
700/*
701 * when btree blocks are allocated, they have some corresponding bits set for
702 * them in one of two extent_io trees. This is used to make sure all of
703 * those extents are on disk for transaction or log commit. We wait
704 * on all the pages and clear them from the dirty pages state tree
705 */
706int btrfs_wait_marked_extents(struct btrfs_root *root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000707 struct extent_io_tree *dirty_pages, int mark)
Chris Mason690587d2009-10-13 13:29:19 -0400708{
Chris Mason690587d2009-10-13 13:29:19 -0400709 int err = 0;
710 int werr = 0;
Josef Bacik17283662011-09-26 13:58:47 -0400711 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
Chris Mason690587d2009-10-13 13:29:19 -0400712 u64 start = 0;
713 u64 end;
Chris Mason690587d2009-10-13 13:29:19 -0400714
Josef Bacik17283662011-09-26 13:58:47 -0400715 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
716 EXTENT_NEED_WAIT)) {
717 clear_extent_bits(dirty_pages, start, end, EXTENT_NEED_WAIT, GFP_NOFS);
718 err = filemap_fdatawait_range(mapping, start, end);
719 if (err)
720 werr = err;
721 cond_resched();
722 start = end + 1;
Chris Mason777e6bd2008-08-15 15:34:15 -0400723 }
Chris Mason7c4452b2007-04-28 09:29:35 -0400724 if (err)
725 werr = err;
726 return werr;
Chris Mason79154b12007-03-22 15:59:16 -0400727}
728
Chris Mason690587d2009-10-13 13:29:19 -0400729/*
730 * when btree blocks are allocated, they have some corresponding bits set for
731 * them in one of two extent_io trees. This is used to make sure all of
732 * those extents are on disk for transaction or log commit
733 */
734int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000735 struct extent_io_tree *dirty_pages, int mark)
Chris Mason690587d2009-10-13 13:29:19 -0400736{
737 int ret;
738 int ret2;
739
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000740 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
741 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
Chris Masonbf0da8c2011-11-04 12:29:37 -0400742
743 if (ret)
744 return ret;
745 if (ret2)
746 return ret2;
747 return 0;
Chris Mason690587d2009-10-13 13:29:19 -0400748}
749
Chris Masond0c803c2008-09-11 16:17:57 -0400750int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
751 struct btrfs_root *root)
752{
753 if (!trans || !trans->transaction) {
754 struct inode *btree_inode;
755 btree_inode = root->fs_info->btree_inode;
756 return filemap_write_and_wait(btree_inode->i_mapping);
757 }
758 return btrfs_write_and_wait_marked_extents(root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000759 &trans->transaction->dirty_pages,
760 EXTENT_DIRTY);
Chris Masond0c803c2008-09-11 16:17:57 -0400761}
762
Chris Masond352ac62008-09-29 15:18:18 -0400763/*
764 * this is used to update the root pointer in the tree of tree roots.
765 *
766 * But, in the case of the extent allocation tree, updating the root
767 * pointer may allocate blocks which may change the root of the extent
768 * allocation tree.
769 *
770 * So, this loops and repeats and makes sure the cowonly root didn't
771 * change while the root pointer was being updated in the metadata.
772 */
Chris Mason0b86a832008-03-24 15:01:56 -0400773static int update_cowonly_root(struct btrfs_trans_handle *trans,
774 struct btrfs_root *root)
775{
776 int ret;
777 u64 old_root_bytenr;
Yan, Zheng86b9f2e2009-11-12 09:36:50 +0000778 u64 old_root_used;
Chris Mason0b86a832008-03-24 15:01:56 -0400779 struct btrfs_root *tree_root = root->fs_info->tree_root;
780
Yan, Zheng86b9f2e2009-11-12 09:36:50 +0000781 old_root_used = btrfs_root_used(&root->root_item);
Chris Mason0b86a832008-03-24 15:01:56 -0400782 btrfs_write_dirty_block_groups(trans, root);
Chris Mason56bec292009-03-13 10:10:06 -0400783
Chris Masond3977122009-01-05 21:25:51 -0500784 while (1) {
Chris Mason0b86a832008-03-24 15:01:56 -0400785 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
Yan, Zheng86b9f2e2009-11-12 09:36:50 +0000786 if (old_root_bytenr == root->node->start &&
787 old_root_used == btrfs_root_used(&root->root_item))
Chris Mason0b86a832008-03-24 15:01:56 -0400788 break;
Chris Mason87ef2bb2008-10-30 11:23:27 -0400789
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400790 btrfs_set_root_node(&root->root_item, root->node);
Chris Mason0b86a832008-03-24 15:01:56 -0400791 ret = btrfs_update_root(trans, tree_root,
792 &root->root_key,
793 &root->root_item);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100794 if (ret)
795 return ret;
Chris Mason56bec292009-03-13 10:10:06 -0400796
Yan, Zheng86b9f2e2009-11-12 09:36:50 +0000797 old_root_used = btrfs_root_used(&root->root_item);
Yan Zheng4a8c9a62009-07-22 10:07:05 -0400798 ret = btrfs_write_dirty_block_groups(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100799 if (ret)
800 return ret;
Chris Mason0b86a832008-03-24 15:01:56 -0400801 }
Yan Zheng276e6802009-07-30 09:40:40 -0400802
803 if (root != root->fs_info->extent_root)
804 switch_commit_root(root);
805
Chris Mason0b86a832008-03-24 15:01:56 -0400806 return 0;
807}
808
Chris Masond352ac62008-09-29 15:18:18 -0400809/*
810 * update all the cowonly tree roots on disk
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100811 *
812 * The error handling in this function may not be obvious. Any of the
813 * failures will cause the file system to go offline. We still need
814 * to clean up the delayed refs.
Chris Masond352ac62008-09-29 15:18:18 -0400815 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400816static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
817 struct btrfs_root *root)
Chris Mason79154b12007-03-22 15:59:16 -0400818{
Chris Mason79154b12007-03-22 15:59:16 -0400819 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Mason0b86a832008-03-24 15:01:56 -0400820 struct list_head *next;
Yan Zheng84234f32008-10-29 14:49:05 -0400821 struct extent_buffer *eb;
Chris Mason56bec292009-03-13 10:10:06 -0400822 int ret;
Yan Zheng84234f32008-10-29 14:49:05 -0400823
Chris Mason56bec292009-03-13 10:10:06 -0400824 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100825 if (ret)
826 return ret;
Chris Mason87ef2bb2008-10-30 11:23:27 -0400827
Yan Zheng84234f32008-10-29 14:49:05 -0400828 eb = btrfs_lock_root_node(fs_info->tree_root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100829 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
830 0, &eb);
Yan Zheng84234f32008-10-29 14:49:05 -0400831 btrfs_tree_unlock(eb);
832 free_extent_buffer(eb);
Chris Mason79154b12007-03-22 15:59:16 -0400833
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100834 if (ret)
835 return ret;
836
Chris Mason56bec292009-03-13 10:10:06 -0400837 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100838 if (ret)
839 return ret;
Chris Mason87ef2bb2008-10-30 11:23:27 -0400840
Stefan Behrens733f4fb2012-05-25 16:06:10 +0200841 ret = btrfs_run_dev_stats(trans, root->fs_info);
842 BUG_ON(ret);
843
Jan Schmidt546adb02012-06-14 16:37:44 +0200844 ret = btrfs_run_qgroups(trans, root->fs_info);
845 BUG_ON(ret);
846
847 /* run_qgroups might have added some more refs */
848 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
849 BUG_ON(ret);
850
Chris Masond3977122009-01-05 21:25:51 -0500851 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
Chris Mason0b86a832008-03-24 15:01:56 -0400852 next = fs_info->dirty_cowonly_roots.next;
853 list_del_init(next);
854 root = list_entry(next, struct btrfs_root, dirty_list);
Chris Mason87ef2bb2008-10-30 11:23:27 -0400855
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100856 ret = update_cowonly_root(trans, root);
857 if (ret)
858 return ret;
Chris Mason79154b12007-03-22 15:59:16 -0400859 }
Yan Zheng276e6802009-07-30 09:40:40 -0400860
861 down_write(&fs_info->extent_commit_sem);
862 switch_commit_root(fs_info->extent_root);
863 up_write(&fs_info->extent_commit_sem);
864
Chris Mason79154b12007-03-22 15:59:16 -0400865 return 0;
866}
867
Chris Masond352ac62008-09-29 15:18:18 -0400868/*
869 * dead roots are old snapshots that need to be deleted. This allocates
870 * a dirty root struct and adds it into the list of dead roots that need to
871 * be deleted
872 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400873int btrfs_add_dead_root(struct btrfs_root *root)
Chris Mason5eda7b52007-06-22 14:16:25 -0400874{
Josef Bacika4abeea2011-04-11 17:25:13 -0400875 spin_lock(&root->fs_info->trans_lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400876 list_add(&root->root_list, &root->fs_info->dead_roots);
Josef Bacika4abeea2011-04-11 17:25:13 -0400877 spin_unlock(&root->fs_info->trans_lock);
Chris Mason5eda7b52007-06-22 14:16:25 -0400878 return 0;
879}
880
Chris Masond352ac62008-09-29 15:18:18 -0400881/*
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400882 * update all the cowonly tree roots on disk
Chris Masond352ac62008-09-29 15:18:18 -0400883 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400884static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
885 struct btrfs_root *root)
Chris Mason0f7d52f2007-04-09 10:42:37 -0400886{
Chris Mason0f7d52f2007-04-09 10:42:37 -0400887 struct btrfs_root *gang[8];
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400888 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Mason0f7d52f2007-04-09 10:42:37 -0400889 int i;
890 int ret;
Chris Mason54aa1f42007-06-22 14:16:25 -0400891 int err = 0;
892
Josef Bacika4abeea2011-04-11 17:25:13 -0400893 spin_lock(&fs_info->fs_roots_radix_lock);
Chris Masond3977122009-01-05 21:25:51 -0500894 while (1) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400895 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
896 (void **)gang, 0,
Chris Mason0f7d52f2007-04-09 10:42:37 -0400897 ARRAY_SIZE(gang),
898 BTRFS_ROOT_TRANS_TAG);
899 if (ret == 0)
900 break;
901 for (i = 0; i < ret; i++) {
902 root = gang[i];
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400903 radix_tree_tag_clear(&fs_info->fs_roots_radix,
904 (unsigned long)root->root_key.objectid,
905 BTRFS_ROOT_TRANS_TAG);
Josef Bacika4abeea2011-04-11 17:25:13 -0400906 spin_unlock(&fs_info->fs_roots_radix_lock);
Yan Zheng31153d82008-07-28 15:32:19 -0400907
Chris Masone02119d2008-09-05 16:13:11 -0400908 btrfs_free_log(trans, root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400909 btrfs_update_reloc_root(trans, root);
Yan, Zhengd68fc572010-05-16 10:49:58 -0400910 btrfs_orphan_commit_root(trans, root);
Chris Masone02119d2008-09-05 16:13:11 -0400911
Li Zefan82d59022011-04-20 10:33:24 +0800912 btrfs_save_ino_cache(root, trans);
913
Liu Bof1ebcc72011-11-14 20:48:06 -0500914 /* see comments in should_cow_block() */
915 root->force_cow = 0;
916 smp_wmb();
917
Yan Zheng978d9102009-06-15 20:01:02 -0400918 if (root->commit_root != root->node) {
Li Zefan581bb052011-04-20 10:06:11 +0800919 mutex_lock(&root->fs_commit_mutex);
Josef Bacik817d52f2009-07-13 21:29:25 -0400920 switch_commit_root(root);
Li Zefan581bb052011-04-20 10:06:11 +0800921 btrfs_unpin_free_ino(root);
922 mutex_unlock(&root->fs_commit_mutex);
923
Yan Zheng978d9102009-06-15 20:01:02 -0400924 btrfs_set_root_node(&root->root_item,
925 root->node);
926 }
Chris Mason9f3a7422007-08-07 15:52:19 -0400927
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400928 err = btrfs_update_root(trans, fs_info->tree_root,
Chris Mason0f7d52f2007-04-09 10:42:37 -0400929 &root->root_key,
930 &root->root_item);
Josef Bacika4abeea2011-04-11 17:25:13 -0400931 spin_lock(&fs_info->fs_roots_radix_lock);
Chris Mason54aa1f42007-06-22 14:16:25 -0400932 if (err)
933 break;
Chris Mason0f7d52f2007-04-09 10:42:37 -0400934 }
935 }
Josef Bacika4abeea2011-04-11 17:25:13 -0400936 spin_unlock(&fs_info->fs_roots_radix_lock);
Chris Mason54aa1f42007-06-22 14:16:25 -0400937 return err;
Chris Mason0f7d52f2007-04-09 10:42:37 -0400938}
939
Chris Masond352ac62008-09-29 15:18:18 -0400940/*
941 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
942 * otherwise every leaf in the btree is read and defragged.
943 */
Chris Masone9d0b132007-08-10 14:06:19 -0400944int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
945{
946 struct btrfs_fs_info *info = root->fs_info;
Chris Masone9d0b132007-08-10 14:06:19 -0400947 struct btrfs_trans_handle *trans;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400948 int ret;
Chris Masond3c2fdcf2007-09-17 10:58:06 -0400949 unsigned long nr;
Chris Masone9d0b132007-08-10 14:06:19 -0400950
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400951 if (xchg(&root->defrag_running, 1))
Chris Masone9d0b132007-08-10 14:06:19 -0400952 return 0;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400953
Chris Mason6b800532007-10-15 16:17:34 -0400954 while (1) {
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400955 trans = btrfs_start_transaction(root, 0);
956 if (IS_ERR(trans))
957 return PTR_ERR(trans);
958
Chris Masone9d0b132007-08-10 14:06:19 -0400959 ret = btrfs_defrag_leaves(trans, root, cacheonly);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400960
Chris Masond3c2fdcf2007-09-17 10:58:06 -0400961 nr = trans->blocks_used;
Chris Masone9d0b132007-08-10 14:06:19 -0400962 btrfs_end_transaction(trans, root);
Chris Masond3c2fdcf2007-09-17 10:58:06 -0400963 btrfs_btree_balance_dirty(info->tree_root, nr);
Chris Masone9d0b132007-08-10 14:06:19 -0400964 cond_resched();
965
David Sterba7841cb22011-05-31 18:07:27 +0200966 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
Chris Masone9d0b132007-08-10 14:06:19 -0400967 break;
968 }
969 root->defrag_running = 0;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400970 return ret;
Chris Masone9d0b132007-08-10 14:06:19 -0400971}
972
Chris Masond352ac62008-09-29 15:18:18 -0400973/*
974 * new snapshots need to be created at a very specific time in the
975 * transaction commit. This does the actual creation
976 */
Chris Mason80b67942008-02-01 16:35:04 -0500977static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
Chris Mason3063d292008-01-08 15:46:30 -0500978 struct btrfs_fs_info *fs_info,
979 struct btrfs_pending_snapshot *pending)
980{
981 struct btrfs_key key;
Chris Mason80b67942008-02-01 16:35:04 -0500982 struct btrfs_root_item *new_root_item;
Chris Mason3063d292008-01-08 15:46:30 -0500983 struct btrfs_root *tree_root = fs_info->tree_root;
984 struct btrfs_root *root = pending->root;
Sage Weil6bdb72d2010-03-15 17:27:13 +0000985 struct btrfs_root *parent_root;
Liu Bo98c99422011-09-11 10:52:24 -0400986 struct btrfs_block_rsv *rsv;
Sage Weil6bdb72d2010-03-15 17:27:13 +0000987 struct inode *parent_inode;
Miao Xie42874b32012-09-06 04:03:32 -0600988 struct btrfs_path *path;
989 struct btrfs_dir_item *dir_item;
Josef Bacik6a912212010-11-20 09:48:00 +0000990 struct dentry *parent;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400991 struct dentry *dentry;
Chris Mason3063d292008-01-08 15:46:30 -0500992 struct extent_buffer *tmp;
Chris Mason925baed2008-06-25 16:01:30 -0400993 struct extent_buffer *old;
Alexander Block8ea05e32012-07-25 17:35:53 +0200994 struct timespec cur_time = CURRENT_TIME;
Chris Mason3063d292008-01-08 15:46:30 -0500995 int ret;
Yan, Zhengd68fc572010-05-16 10:49:58 -0400996 u64 to_reserve = 0;
Sage Weil6bdb72d2010-03-15 17:27:13 +0000997 u64 index = 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400998 u64 objectid;
Li Zefanb83cc962010-12-20 16:04:08 +0800999 u64 root_flags;
Alexander Block8ea05e32012-07-25 17:35:53 +02001000 uuid_le new_uuid;
Chris Mason3063d292008-01-08 15:46:30 -05001001
Miao Xie42874b32012-09-06 04:03:32 -06001002 path = btrfs_alloc_path();
1003 if (!path) {
1004 ret = pending->error = -ENOMEM;
1005 goto path_alloc_fail;
1006 }
1007
Chris Mason80b67942008-02-01 16:35:04 -05001008 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
1009 if (!new_root_item) {
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001010 ret = pending->error = -ENOMEM;
Miao Xie6fa97002012-09-06 04:00:32 -06001011 goto root_item_alloc_fail;
Chris Mason80b67942008-02-01 16:35:04 -05001012 }
Yan, Zhenga22285a2010-05-16 10:48:46 -04001013
Li Zefan581bb052011-04-20 10:06:11 +08001014 ret = btrfs_find_free_objectid(tree_root, &objectid);
Yan, Zhenga22285a2010-05-16 10:48:46 -04001015 if (ret) {
1016 pending->error = ret;
Miao Xie6fa97002012-09-06 04:00:32 -06001017 goto no_free_objectid;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001018 }
Chris Mason3063d292008-01-08 15:46:30 -05001019
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001020 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
Yan, Zhengd68fc572010-05-16 10:49:58 -04001021
1022 if (to_reserve > 0) {
Miao Xie62f30c52011-11-10 20:45:05 -05001023 ret = btrfs_block_rsv_add_noflush(root, &pending->block_rsv,
1024 to_reserve);
Yan, Zhengd68fc572010-05-16 10:49:58 -04001025 if (ret) {
1026 pending->error = ret;
Miao Xie6fa97002012-09-06 04:00:32 -06001027 goto no_free_objectid;
Yan, Zhengd68fc572010-05-16 10:49:58 -04001028 }
1029 }
1030
Arne Jansen6f72c7e2011-09-14 15:58:21 +02001031 ret = btrfs_qgroup_inherit(trans, fs_info, root->root_key.objectid,
1032 objectid, pending->inherit);
Arne Jansen6f72c7e2011-09-14 15:58:21 +02001033 if (ret) {
1034 pending->error = ret;
Miao Xie6fa97002012-09-06 04:00:32 -06001035 goto no_free_objectid;
Arne Jansen6f72c7e2011-09-14 15:58:21 +02001036 }
1037
Chris Mason3063d292008-01-08 15:46:30 -05001038 key.objectid = objectid;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001039 key.offset = (u64)-1;
1040 key.type = BTRFS_ROOT_ITEM_KEY;
Chris Mason3063d292008-01-08 15:46:30 -05001041
Miao Xie6fa97002012-09-06 04:00:32 -06001042 rsv = trans->block_rsv;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001043 trans->block_rsv = &pending->block_rsv;
Sage Weil6bdb72d2010-03-15 17:27:13 +00001044
Yan, Zhenga22285a2010-05-16 10:48:46 -04001045 dentry = pending->dentry;
Josef Bacik6a912212010-11-20 09:48:00 +00001046 parent = dget_parent(dentry);
1047 parent_inode = parent->d_inode;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001048 parent_root = BTRFS_I(parent_inode)->root;
Chris Mason75857172011-06-13 20:00:16 -04001049 record_root_in_trans(trans, parent_root);
Yan, Zhenga22285a2010-05-16 10:48:46 -04001050
Sage Weil6bdb72d2010-03-15 17:27:13 +00001051 /*
1052 * insert the directory item
1053 */
Sage Weil6bdb72d2010-03-15 17:27:13 +00001054 ret = btrfs_set_inode_index(parent_inode, &index);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001055 BUG_ON(ret); /* -ENOMEM */
Miao Xie42874b32012-09-06 04:03:32 -06001056
1057 /* check if there is a file/dir which has the same name. */
1058 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1059 btrfs_ino(parent_inode),
1060 dentry->d_name.name,
1061 dentry->d_name.len, 0);
1062 if (dir_item != NULL && !IS_ERR(dir_item)) {
Chris Masonfe66a052012-02-20 08:40:56 -05001063 pending->error = -EEXIST;
Chris Masonfe66a052012-02-20 08:40:56 -05001064 goto fail;
Miao Xie42874b32012-09-06 04:03:32 -06001065 } else if (IS_ERR(dir_item)) {
1066 ret = PTR_ERR(dir_item);
Miao Xie6fa97002012-09-06 04:00:32 -06001067 goto abort_trans;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001068 }
Miao Xie42874b32012-09-06 04:03:32 -06001069 btrfs_release_path(path);
Sage Weil6bdb72d2010-03-15 17:27:13 +00001070
Chris Masone9993762011-06-17 16:14:09 -04001071 /*
1072 * pull in the delayed directory update
1073 * and the delayed inode item
1074 * otherwise we corrupt the FS during
1075 * snapshot
1076 */
1077 ret = btrfs_run_delayed_items(trans, root);
Miao Xie6fa97002012-09-06 04:00:32 -06001078 if (ret) /* Transaction aborted */
1079 goto abort_trans;
Chris Masone9993762011-06-17 16:14:09 -04001080
Chris Mason75857172011-06-13 20:00:16 -04001081 record_root_in_trans(trans, root);
Sage Weil6bdb72d2010-03-15 17:27:13 +00001082 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1083 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
Li Zefan08fe4db2011-03-28 02:01:25 +00001084 btrfs_check_and_init_root_item(new_root_item);
Sage Weil6bdb72d2010-03-15 17:27:13 +00001085
Li Zefanb83cc962010-12-20 16:04:08 +08001086 root_flags = btrfs_root_flags(new_root_item);
1087 if (pending->readonly)
1088 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1089 else
1090 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1091 btrfs_set_root_flags(new_root_item, root_flags);
1092
Alexander Block8ea05e32012-07-25 17:35:53 +02001093 btrfs_set_root_generation_v2(new_root_item,
1094 trans->transid);
1095 uuid_le_gen(&new_uuid);
1096 memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
1097 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1098 BTRFS_UUID_SIZE);
1099 new_root_item->otime.sec = cpu_to_le64(cur_time.tv_sec);
Dan Carpenterdadd1102012-07-30 02:10:44 -06001100 new_root_item->otime.nsec = cpu_to_le32(cur_time.tv_nsec);
Alexander Block8ea05e32012-07-25 17:35:53 +02001101 btrfs_set_root_otransid(new_root_item, trans->transid);
1102 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1103 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1104 btrfs_set_root_stransid(new_root_item, 0);
1105 btrfs_set_root_rtransid(new_root_item, 0);
1106
Chris Mason925baed2008-06-25 16:01:30 -04001107 old = btrfs_lock_root_node(root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001108 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001109 if (ret) {
1110 btrfs_tree_unlock(old);
1111 free_extent_buffer(old);
Miao Xie6fa97002012-09-06 04:00:32 -06001112 goto abort_trans;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001113 }
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001114
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001115 btrfs_set_lock_blocking(old);
Chris Mason3063d292008-01-08 15:46:30 -05001116
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001117 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001118 /* clean up in any case */
Chris Mason925baed2008-06-25 16:01:30 -04001119 btrfs_tree_unlock(old);
1120 free_extent_buffer(old);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001121 if (ret)
Miao Xie6fa97002012-09-06 04:00:32 -06001122 goto abort_trans;
Chris Mason3063d292008-01-08 15:46:30 -05001123
Liu Bof1ebcc72011-11-14 20:48:06 -05001124 /* see comments in should_cow_block() */
1125 root->force_cow = 1;
1126 smp_wmb();
1127
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001128 btrfs_set_root_node(new_root_item, tmp);
Yan, Zhenga22285a2010-05-16 10:48:46 -04001129 /* record when the snapshot was created in key.offset */
1130 key.offset = trans->transid;
1131 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
Chris Mason925baed2008-06-25 16:01:30 -04001132 btrfs_tree_unlock(tmp);
Chris Mason3063d292008-01-08 15:46:30 -05001133 free_extent_buffer(tmp);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001134 if (ret)
Miao Xie6fa97002012-09-06 04:00:32 -06001135 goto abort_trans;
Chris Mason0660b5a2008-11-17 20:37:39 -05001136
Yan, Zhenga22285a2010-05-16 10:48:46 -04001137 /*
1138 * insert root back/forward references
1139 */
1140 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1141 parent_root->root_key.objectid,
Li Zefan33345d012011-04-20 10:31:50 +08001142 btrfs_ino(parent_inode), index,
Yan, Zhenga22285a2010-05-16 10:48:46 -04001143 dentry->d_name.name, dentry->d_name.len);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001144 if (ret)
Miao Xie6fa97002012-09-06 04:00:32 -06001145 goto abort_trans;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001146
1147 key.offset = (u64)-1;
1148 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001149 if (IS_ERR(pending->snap)) {
1150 ret = PTR_ERR(pending->snap);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001151 goto abort_trans;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001152 }
Yan, Zhengd68fc572010-05-16 10:49:58 -04001153
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001154 ret = btrfs_reloc_post_snapshot(trans, pending);
1155 if (ret)
1156 goto abort_trans;
Miao Xie361048f2012-09-06 04:00:57 -06001157
1158 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1159 if (ret)
1160 goto abort_trans;
Miao Xie42874b32012-09-06 04:03:32 -06001161
1162 ret = btrfs_insert_dir_item(trans, parent_root,
1163 dentry->d_name.name, dentry->d_name.len,
1164 parent_inode, &key,
1165 BTRFS_FT_DIR, index);
1166 /* We have check then name at the beginning, so it is impossible. */
1167 BUG_ON(ret == -EEXIST);
1168 if (ret)
1169 goto abort_trans;
1170
1171 btrfs_i_size_write(parent_inode, parent_inode->i_size +
1172 dentry->d_name.len * 2);
1173 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1174 ret = btrfs_update_inode(trans, parent_root, parent_inode);
1175 if (ret)
1176 goto abort_trans;
Chris Mason3063d292008-01-08 15:46:30 -05001177fail:
Miao Xie6fa97002012-09-06 04:00:32 -06001178 dput(parent);
Liu Bo98c99422011-09-11 10:52:24 -04001179 trans->block_rsv = rsv;
Miao Xie6fa97002012-09-06 04:00:32 -06001180no_free_objectid:
1181 kfree(new_root_item);
1182root_item_alloc_fail:
Miao Xie42874b32012-09-06 04:03:32 -06001183 btrfs_free_path(path);
1184path_alloc_fail:
Yan, Zhenga22285a2010-05-16 10:48:46 -04001185 btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001186 return ret;
1187
1188abort_trans:
1189 btrfs_abort_transaction(trans, root, ret);
1190 goto fail;
Chris Mason3063d292008-01-08 15:46:30 -05001191}
1192
Chris Masond352ac62008-09-29 15:18:18 -04001193/*
1194 * create all the snapshots we've scheduled for creation
1195 */
Chris Mason80b67942008-02-01 16:35:04 -05001196static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1197 struct btrfs_fs_info *fs_info)
Chris Mason3063d292008-01-08 15:46:30 -05001198{
1199 struct btrfs_pending_snapshot *pending;
1200 struct list_head *head = &trans->transaction->pending_snapshots;
Chris Mason3de45862008-11-17 21:02:50 -05001201
Chris Masonfe66a052012-02-20 08:40:56 -05001202 list_for_each_entry(pending, head, list)
1203 create_pending_snapshot(trans, fs_info, pending);
Chris Mason3de45862008-11-17 21:02:50 -05001204 return 0;
1205}
1206
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001207static void update_super_roots(struct btrfs_root *root)
1208{
1209 struct btrfs_root_item *root_item;
1210 struct btrfs_super_block *super;
1211
David Sterba6c417612011-04-13 15:41:04 +02001212 super = root->fs_info->super_copy;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001213
1214 root_item = &root->fs_info->chunk_root->root_item;
1215 super->chunk_root = root_item->bytenr;
1216 super->chunk_root_generation = root_item->generation;
1217 super->chunk_root_level = root_item->level;
1218
1219 root_item = &root->fs_info->tree_root->root_item;
1220 super->root = root_item->bytenr;
1221 super->generation = root_item->generation;
1222 super->root_level = root_item->level;
Josef Bacik73bc1872011-10-03 14:07:49 -04001223 if (btrfs_test_opt(root, SPACE_CACHE))
Josef Bacik0af3d002010-06-21 14:48:16 -04001224 super->cache_generation = root_item->generation;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001225}
1226
Chris Masonf36f3042009-07-30 10:04:48 -04001227int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1228{
1229 int ret = 0;
Josef Bacika4abeea2011-04-11 17:25:13 -04001230 spin_lock(&info->trans_lock);
Chris Masonf36f3042009-07-30 10:04:48 -04001231 if (info->running_transaction)
1232 ret = info->running_transaction->in_commit;
Josef Bacika4abeea2011-04-11 17:25:13 -04001233 spin_unlock(&info->trans_lock);
Chris Masonf36f3042009-07-30 10:04:48 -04001234 return ret;
1235}
1236
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001237int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1238{
1239 int ret = 0;
Josef Bacika4abeea2011-04-11 17:25:13 -04001240 spin_lock(&info->trans_lock);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001241 if (info->running_transaction)
1242 ret = info->running_transaction->blocked;
Josef Bacika4abeea2011-04-11 17:25:13 -04001243 spin_unlock(&info->trans_lock);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001244 return ret;
1245}
1246
Sage Weilbb9c12c2010-10-29 15:37:34 -04001247/*
1248 * wait for the current transaction commit to start and block subsequent
1249 * transaction joins
1250 */
1251static void wait_current_trans_commit_start(struct btrfs_root *root,
1252 struct btrfs_transaction *trans)
1253{
Li Zefan72d63ed2011-07-14 03:17:00 +00001254 wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001255}
1256
1257/*
1258 * wait for the current transaction to start and then become unblocked.
1259 * caller holds ref.
1260 */
1261static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1262 struct btrfs_transaction *trans)
1263{
Li Zefan72d63ed2011-07-14 03:17:00 +00001264 wait_event(root->fs_info->transaction_wait,
1265 trans->commit_done || (trans->in_commit && !trans->blocked));
Sage Weilbb9c12c2010-10-29 15:37:34 -04001266}
1267
1268/*
1269 * commit transactions asynchronously. once btrfs_commit_transaction_async
1270 * returns, any subsequent transaction will not be allowed to join.
1271 */
1272struct btrfs_async_commit {
1273 struct btrfs_trans_handle *newtrans;
1274 struct btrfs_root *root;
1275 struct delayed_work work;
1276};
1277
1278static void do_async_commit(struct work_struct *work)
1279{
1280 struct btrfs_async_commit *ac =
1281 container_of(work, struct btrfs_async_commit, work.work);
1282
Sage Weil6fc4e35482012-08-30 16:26:15 -06001283 /*
1284 * We've got freeze protection passed with the transaction.
1285 * Tell lockdep about it.
1286 */
1287 rwsem_acquire_read(
1288 &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1289 0, 1, _THIS_IP_);
1290
Sage Weile209db72012-08-30 16:26:16 -06001291 current->journal_info = ac->newtrans;
1292
Sage Weilbb9c12c2010-10-29 15:37:34 -04001293 btrfs_commit_transaction(ac->newtrans, ac->root);
1294 kfree(ac);
1295}
1296
1297int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1298 struct btrfs_root *root,
1299 int wait_for_unblock)
1300{
1301 struct btrfs_async_commit *ac;
1302 struct btrfs_transaction *cur_trans;
1303
1304 ac = kmalloc(sizeof(*ac), GFP_NOFS);
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00001305 if (!ac)
1306 return -ENOMEM;
Sage Weilbb9c12c2010-10-29 15:37:34 -04001307
1308 INIT_DELAYED_WORK(&ac->work, do_async_commit);
1309 ac->root = root;
Josef Bacik7a7eaa42011-04-13 12:54:33 -04001310 ac->newtrans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00001311 if (IS_ERR(ac->newtrans)) {
1312 int err = PTR_ERR(ac->newtrans);
1313 kfree(ac);
1314 return err;
1315 }
Sage Weilbb9c12c2010-10-29 15:37:34 -04001316
1317 /* take transaction reference */
Sage Weilbb9c12c2010-10-29 15:37:34 -04001318 cur_trans = trans->transaction;
Josef Bacik13c5a932011-04-11 15:45:29 -04001319 atomic_inc(&cur_trans->use_count);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001320
1321 btrfs_end_transaction(trans, root);
Sage Weil6fc4e35482012-08-30 16:26:15 -06001322
1323 /*
1324 * Tell lockdep we've released the freeze rwsem, since the
1325 * async commit thread will be the one to unlock it.
1326 */
1327 rwsem_release(&root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1328 1, _THIS_IP_);
1329
Sage Weilbb9c12c2010-10-29 15:37:34 -04001330 schedule_delayed_work(&ac->work, 0);
1331
1332 /* wait for transaction to start and unblock */
Sage Weilbb9c12c2010-10-29 15:37:34 -04001333 if (wait_for_unblock)
1334 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1335 else
1336 wait_current_trans_commit_start(root, cur_trans);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001337
Sage Weil38e88052011-06-10 18:43:13 +00001338 if (current->journal_info == trans)
1339 current->journal_info = NULL;
1340
1341 put_transaction(cur_trans);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001342 return 0;
1343}
1344
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001345
1346static void cleanup_transaction(struct btrfs_trans_handle *trans,
Josef Bacik7b8b92a2012-05-31 15:52:43 -04001347 struct btrfs_root *root, int err)
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001348{
1349 struct btrfs_transaction *cur_trans = trans->transaction;
1350
1351 WARN_ON(trans->use_count > 1);
1352
Josef Bacik7b8b92a2012-05-31 15:52:43 -04001353 btrfs_abort_transaction(trans, root, err);
1354
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001355 spin_lock(&root->fs_info->trans_lock);
1356 list_del_init(&cur_trans->list);
Josef Bacikd7096fc2012-05-31 15:49:57 -04001357 if (cur_trans == root->fs_info->running_transaction) {
1358 root->fs_info->running_transaction = NULL;
1359 root->fs_info->trans_no_join = 0;
1360 }
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001361 spin_unlock(&root->fs_info->trans_lock);
1362
1363 btrfs_cleanup_one_transaction(trans->transaction, root);
1364
1365 put_transaction(cur_trans);
1366 put_transaction(cur_trans);
1367
1368 trace_btrfs_transaction_commit(root);
1369
1370 btrfs_scrub_continue(root);
1371
1372 if (current->journal_info == trans)
1373 current->journal_info = NULL;
1374
1375 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1376}
1377
Sage Weilbb9c12c2010-10-29 15:37:34 -04001378/*
1379 * btrfs_transaction state sequence:
1380 * in_commit = 0, blocked = 0 (initial)
1381 * in_commit = 1, blocked = 1
1382 * blocked = 0
1383 * commit_done = 1
1384 */
Chris Mason79154b12007-03-22 15:59:16 -04001385int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1386 struct btrfs_root *root)
1387{
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001388 unsigned long joined = 0;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001389 struct btrfs_transaction *cur_trans = trans->transaction;
Chris Mason8fd17792007-04-19 21:01:03 -04001390 struct btrfs_transaction *prev_trans = NULL;
Chris Mason79154b12007-03-22 15:59:16 -04001391 DEFINE_WAIT(wait);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001392 int ret = -EIO;
Chris Mason89573b92009-03-12 20:12:45 -04001393 int should_grow = 0;
1394 unsigned long now = get_seconds();
Sage Weildccae992009-04-02 16:59:01 -04001395 int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
Chris Mason79154b12007-03-22 15:59:16 -04001396
Chris Mason5a3f23d2009-03-31 13:27:11 -04001397 btrfs_run_ordered_operations(root, 0);
1398
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001399 if (cur_trans->aborted)
1400 goto cleanup_transaction;
1401
Chris Mason56bec292009-03-13 10:10:06 -04001402 /* make a pass through all the delayed refs we have so far
1403 * any runnings procs may add more while we are here
1404 */
1405 ret = btrfs_run_delayed_refs(trans, root, 0);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001406 if (ret)
1407 goto cleanup_transaction;
Chris Mason56bec292009-03-13 10:10:06 -04001408
Josef Bacik0e721102012-06-26 16:13:18 -04001409 btrfs_trans_release_metadata(trans, root);
1410 trans->block_rsv = NULL;
1411
Chris Masonb7ec40d2009-03-12 20:12:45 -04001412 cur_trans = trans->transaction;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001413
Chris Mason56bec292009-03-13 10:10:06 -04001414 /*
1415 * set the flushing flag so procs in this transaction have to
1416 * start sending their work down.
1417 */
Chris Masonb7ec40d2009-03-12 20:12:45 -04001418 cur_trans->delayed_refs.flushing = 1;
Chris Mason56bec292009-03-13 10:10:06 -04001419
Josef Bacikea658ba2012-09-11 16:57:25 -04001420 if (!list_empty(&trans->new_bgs))
1421 btrfs_create_pending_block_groups(trans, root);
1422
Chris Masonc3e69d52009-03-13 10:17:05 -04001423 ret = btrfs_run_delayed_refs(trans, root, 0);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001424 if (ret)
1425 goto cleanup_transaction;
Chris Mason56bec292009-03-13 10:10:06 -04001426
Josef Bacika4abeea2011-04-11 17:25:13 -04001427 spin_lock(&cur_trans->commit_lock);
Chris Masonb7ec40d2009-03-12 20:12:45 -04001428 if (cur_trans->in_commit) {
Josef Bacika4abeea2011-04-11 17:25:13 -04001429 spin_unlock(&cur_trans->commit_lock);
Josef Bacik13c5a932011-04-11 15:45:29 -04001430 atomic_inc(&cur_trans->use_count);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001431 ret = btrfs_end_transaction(trans, root);
Chris Masonccd467d2007-06-28 15:57:36 -04001432
Li Zefanb9c83002011-07-14 03:17:14 +00001433 wait_for_commit(root, cur_trans);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001434
Chris Mason79154b12007-03-22 15:59:16 -04001435 put_transaction(cur_trans);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001436
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001437 return ret;
Chris Mason79154b12007-03-22 15:59:16 -04001438 }
Chris Mason4313b392008-01-03 09:08:48 -05001439
Chris Mason2c90e5d2007-04-02 10:50:19 -04001440 trans->transaction->in_commit = 1;
Chris Masonf9295742008-07-17 12:54:14 -04001441 trans->transaction->blocked = 1;
Josef Bacika4abeea2011-04-11 17:25:13 -04001442 spin_unlock(&cur_trans->commit_lock);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001443 wake_up(&root->fs_info->transaction_blocked_wait);
1444
Josef Bacika4abeea2011-04-11 17:25:13 -04001445 spin_lock(&root->fs_info->trans_lock);
Chris Masonccd467d2007-06-28 15:57:36 -04001446 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1447 prev_trans = list_entry(cur_trans->list.prev,
1448 struct btrfs_transaction, list);
1449 if (!prev_trans->commit_done) {
Josef Bacik13c5a932011-04-11 15:45:29 -04001450 atomic_inc(&prev_trans->use_count);
Josef Bacika4abeea2011-04-11 17:25:13 -04001451 spin_unlock(&root->fs_info->trans_lock);
Chris Masonccd467d2007-06-28 15:57:36 -04001452
1453 wait_for_commit(root, prev_trans);
Chris Masonccd467d2007-06-28 15:57:36 -04001454
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001455 put_transaction(prev_trans);
Josef Bacika4abeea2011-04-11 17:25:13 -04001456 } else {
1457 spin_unlock(&root->fs_info->trans_lock);
Chris Masonccd467d2007-06-28 15:57:36 -04001458 }
Josef Bacika4abeea2011-04-11 17:25:13 -04001459 } else {
1460 spin_unlock(&root->fs_info->trans_lock);
Chris Masonccd467d2007-06-28 15:57:36 -04001461 }
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001462
Chris Masone39e64a2012-07-23 15:23:45 -04001463 if (!btrfs_test_opt(root, SSD) &&
1464 (now < cur_trans->start_time || now - cur_trans->start_time < 1))
Chris Mason89573b92009-03-12 20:12:45 -04001465 should_grow = 1;
1466
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001467 do {
Yan Zheng7ea394f2008-08-05 13:05:02 -04001468 int snap_pending = 0;
Josef Bacika4abeea2011-04-11 17:25:13 -04001469
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001470 joined = cur_trans->num_joined;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001471 if (!list_empty(&trans->transaction->pending_snapshots))
1472 snap_pending = 1;
1473
Chris Mason2c90e5d2007-04-02 10:50:19 -04001474 WARN_ON(cur_trans != trans->transaction);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001475
Sage Weil0bdb1db2010-02-19 14:13:50 -08001476 if (flush_on_commit || snap_pending) {
Yan, Zheng24bbcf02009-11-12 09:36:34 +00001477 btrfs_start_delalloc_inodes(root, 1);
Jeff Mahoney143bede2012-03-01 14:56:26 +01001478 btrfs_wait_ordered_extents(root, 0, 1);
Yan Zheng7ea394f2008-08-05 13:05:02 -04001479 }
1480
Miao Xie16cdcec2011-04-22 18:12:22 +08001481 ret = btrfs_run_delayed_items(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001482 if (ret)
1483 goto cleanup_transaction;
Miao Xie16cdcec2011-04-22 18:12:22 +08001484
Chris Mason5a3f23d2009-03-31 13:27:11 -04001485 /*
Jan Schmidtedf39272012-06-28 18:04:55 +02001486 * running the delayed items may have added new refs. account
1487 * them now so that they hinder processing of more delayed refs
1488 * as little as possible.
1489 */
1490 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
1491
1492 /*
Chris Mason5a3f23d2009-03-31 13:27:11 -04001493 * rename don't use btrfs_join_transaction, so, once we
1494 * set the transaction to blocked above, we aren't going
1495 * to get any new ordered operations. We can safely run
1496 * it here and no for sure that nothing new will be added
1497 * to the list
1498 */
1499 btrfs_run_ordered_operations(root, 1);
1500
Chris Masoned3b3d3142010-05-25 10:12:41 -04001501 prepare_to_wait(&cur_trans->writer_wait, &wait,
1502 TASK_UNINTERRUPTIBLE);
1503
Josef Bacik13c5a932011-04-11 15:45:29 -04001504 if (atomic_read(&cur_trans->num_writers) > 1)
Sage Weil99d16cb2010-10-29 15:37:34 -04001505 schedule_timeout(MAX_SCHEDULE_TIMEOUT);
1506 else if (should_grow)
1507 schedule_timeout(1);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001508
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001509 finish_wait(&cur_trans->writer_wait, &wait);
Josef Bacik13c5a932011-04-11 15:45:29 -04001510 } while (atomic_read(&cur_trans->num_writers) > 1 ||
Chris Mason89573b92009-03-12 20:12:45 -04001511 (should_grow && cur_trans->num_joined != joined));
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001512
Chris Mason75857172011-06-13 20:00:16 -04001513 /*
Josef Baciked0ca142011-06-14 16:22:15 -04001514 * Ok now we need to make sure to block out any other joins while we
1515 * commit the transaction. We could have started a join before setting
1516 * no_join so make sure to wait for num_writers to == 1 again.
1517 */
1518 spin_lock(&root->fs_info->trans_lock);
1519 root->fs_info->trans_no_join = 1;
1520 spin_unlock(&root->fs_info->trans_lock);
1521 wait_event(cur_trans->writer_wait,
1522 atomic_read(&cur_trans->num_writers) == 1);
1523
Chris Masone038dca2011-06-17 14:16:13 -04001524 /*
Chris Mason75857172011-06-13 20:00:16 -04001525 * the reloc mutex makes sure that we stop
1526 * the balancing code from coming in and moving
1527 * extents around in the middle of the commit
1528 */
1529 mutex_lock(&root->fs_info->reloc_mutex);
1530
Miao Xie42874b32012-09-06 04:03:32 -06001531 /*
1532 * We needn't worry about the delayed items because we will
1533 * deal with them in create_pending_snapshot(), which is the
1534 * core function of the snapshot creation.
1535 */
1536 ret = create_pending_snapshots(trans, root->fs_info);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001537 if (ret) {
1538 mutex_unlock(&root->fs_info->reloc_mutex);
1539 goto cleanup_transaction;
1540 }
Chris Mason3063d292008-01-08 15:46:30 -05001541
Miao Xie42874b32012-09-06 04:03:32 -06001542 /*
1543 * We insert the dir indexes of the snapshots and update the inode
1544 * of the snapshots' parents after the snapshot creation, so there
1545 * are some delayed items which are not dealt with. Now deal with
1546 * them.
1547 *
1548 * We needn't worry that this operation will corrupt the snapshots,
1549 * because all the tree which are snapshoted will be forced to COW
1550 * the nodes and leaves.
1551 */
1552 ret = btrfs_run_delayed_items(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001553 if (ret) {
1554 mutex_unlock(&root->fs_info->reloc_mutex);
1555 goto cleanup_transaction;
1556 }
Miao Xie16cdcec2011-04-22 18:12:22 +08001557
Chris Mason56bec292009-03-13 10:10:06 -04001558 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001559 if (ret) {
1560 mutex_unlock(&root->fs_info->reloc_mutex);
1561 goto cleanup_transaction;
1562 }
Chris Mason56bec292009-03-13 10:10:06 -04001563
Chris Masone9993762011-06-17 16:14:09 -04001564 /*
1565 * make sure none of the code above managed to slip in a
1566 * delayed item
1567 */
1568 btrfs_assert_delayed_root_empty(root);
1569
Chris Mason2c90e5d2007-04-02 10:50:19 -04001570 WARN_ON(cur_trans != trans->transaction);
Chris Masondc17ff82008-01-08 15:46:30 -05001571
Arne Jansena2de7332011-03-08 14:14:00 +01001572 btrfs_scrub_pause(root);
Chris Masone02119d2008-09-05 16:13:11 -04001573 /* btrfs_commit_tree_roots is responsible for getting the
1574 * various roots consistent with each other. Every pointer
1575 * in the tree of tree roots has to point to the most up to date
1576 * root for every subvolume and other tree. So, we have to keep
1577 * the tree logging code from jumping in and changing any
1578 * of the trees.
1579 *
1580 * At this point in the commit, there can't be any tree-log
1581 * writers, but a little lower down we drop the trans mutex
1582 * and let new people in. By holding the tree_log_mutex
1583 * from now until after the super is written, we avoid races
1584 * with the tree-log code.
1585 */
1586 mutex_lock(&root->fs_info->tree_log_mutex);
Zheng Yan1a40e232008-09-26 10:09:34 -04001587
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001588 ret = commit_fs_roots(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001589 if (ret) {
1590 mutex_unlock(&root->fs_info->tree_log_mutex);
David Sterba871383b2012-04-02 18:31:37 +02001591 mutex_unlock(&root->fs_info->reloc_mutex);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001592 goto cleanup_transaction;
1593 }
Chris Mason54aa1f42007-06-22 14:16:25 -04001594
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001595 /* commit_fs_roots gets rid of all the tree log roots, it is now
Chris Masone02119d2008-09-05 16:13:11 -04001596 * safe to free the root of tree log roots
1597 */
1598 btrfs_free_log_root_tree(trans, root->fs_info);
1599
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001600 ret = commit_cowonly_roots(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001601 if (ret) {
1602 mutex_unlock(&root->fs_info->tree_log_mutex);
David Sterba871383b2012-04-02 18:31:37 +02001603 mutex_unlock(&root->fs_info->reloc_mutex);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001604 goto cleanup_transaction;
1605 }
Chris Mason54aa1f42007-06-22 14:16:25 -04001606
Yan Zheng11833d62009-09-11 16:11:19 -04001607 btrfs_prepare_extent_commit(trans, root);
1608
Chris Mason78fae272007-03-25 11:35:08 -04001609 cur_trans = root->fs_info->running_transaction;
Chris Mason5f39d392007-10-15 16:14:19 -04001610
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001611 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1612 root->fs_info->tree_root->node);
Josef Bacik817d52f2009-07-13 21:29:25 -04001613 switch_commit_root(root->fs_info->tree_root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001614
1615 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1616 root->fs_info->chunk_root->node);
Josef Bacik817d52f2009-07-13 21:29:25 -04001617 switch_commit_root(root->fs_info->chunk_root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001618
Jan Schmidtedf39272012-06-28 18:04:55 +02001619 assert_qgroups_uptodate(trans);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001620 update_super_roots(root);
Chris Masone02119d2008-09-05 16:13:11 -04001621
1622 if (!root->fs_info->log_root_recovering) {
David Sterba6c417612011-04-13 15:41:04 +02001623 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
1624 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
Chris Masone02119d2008-09-05 16:13:11 -04001625 }
1626
David Sterba6c417612011-04-13 15:41:04 +02001627 memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
1628 sizeof(*root->fs_info->super_copy));
Chris Masonccd467d2007-06-28 15:57:36 -04001629
Chris Masonf9295742008-07-17 12:54:14 -04001630 trans->transaction->blocked = 0;
Josef Bacika4abeea2011-04-11 17:25:13 -04001631 spin_lock(&root->fs_info->trans_lock);
1632 root->fs_info->running_transaction = NULL;
1633 root->fs_info->trans_no_join = 0;
1634 spin_unlock(&root->fs_info->trans_lock);
Chris Mason75857172011-06-13 20:00:16 -04001635 mutex_unlock(&root->fs_info->reloc_mutex);
Chris Masonb7ec40d2009-03-12 20:12:45 -04001636
Chris Masonf9295742008-07-17 12:54:14 -04001637 wake_up(&root->fs_info->transaction_wait);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001638
Chris Mason79154b12007-03-22 15:59:16 -04001639 ret = btrfs_write_and_wait_transaction(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001640 if (ret) {
1641 btrfs_error(root->fs_info, ret,
1642 "Error while writing out transaction.");
1643 mutex_unlock(&root->fs_info->tree_log_mutex);
1644 goto cleanup_transaction;
1645 }
1646
1647 ret = write_ctree_super(trans, root, 0);
1648 if (ret) {
1649 mutex_unlock(&root->fs_info->tree_log_mutex);
1650 goto cleanup_transaction;
1651 }
Chris Mason4313b392008-01-03 09:08:48 -05001652
Chris Masone02119d2008-09-05 16:13:11 -04001653 /*
1654 * the super is written, we can safely allow the tree-loggers
1655 * to go about their business
1656 */
1657 mutex_unlock(&root->fs_info->tree_log_mutex);
1658
Yan Zheng11833d62009-09-11 16:11:19 -04001659 btrfs_finish_extent_commit(trans, root);
Chris Mason4313b392008-01-03 09:08:48 -05001660
Chris Mason2c90e5d2007-04-02 10:50:19 -04001661 cur_trans->commit_done = 1;
Chris Masonb7ec40d2009-03-12 20:12:45 -04001662
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001663 root->fs_info->last_trans_committed = cur_trans->transid;
Josef Bacik817d52f2009-07-13 21:29:25 -04001664
Chris Mason2c90e5d2007-04-02 10:50:19 -04001665 wake_up(&cur_trans->commit_wait);
Chris Mason3de45862008-11-17 21:02:50 -05001666
Josef Bacika4abeea2011-04-11 17:25:13 -04001667 spin_lock(&root->fs_info->trans_lock);
Josef Bacik13c5a932011-04-11 15:45:29 -04001668 list_del_init(&cur_trans->list);
Josef Bacika4abeea2011-04-11 17:25:13 -04001669 spin_unlock(&root->fs_info->trans_lock);
1670
Chris Mason79154b12007-03-22 15:59:16 -04001671 put_transaction(cur_trans);
Chris Mason78fae272007-03-25 11:35:08 -04001672 put_transaction(cur_trans);
Josef Bacik58176a92007-08-29 15:47:34 -04001673
Jan Karab2b5ef52012-06-12 16:20:45 +02001674 sb_end_intwrite(root->fs_info->sb);
1675
liubo1abe9b82011-03-24 11:18:59 +00001676 trace_btrfs_transaction_commit(root);
1677
Arne Jansena2de7332011-03-08 14:14:00 +01001678 btrfs_scrub_continue(root);
1679
Josef Bacik9ed74f22009-09-11 16:12:44 -04001680 if (current->journal_info == trans)
1681 current->journal_info = NULL;
1682
Chris Mason2c90e5d2007-04-02 10:50:19 -04001683 kmem_cache_free(btrfs_trans_handle_cachep, trans);
Yan, Zheng24bbcf02009-11-12 09:36:34 +00001684
1685 if (current != root->fs_info->transaction_kthread)
1686 btrfs_run_delayed_iputs(root);
1687
Chris Mason79154b12007-03-22 15:59:16 -04001688 return ret;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001689
1690cleanup_transaction:
Josef Bacik0e721102012-06-26 16:13:18 -04001691 btrfs_trans_release_metadata(trans, root);
1692 trans->block_rsv = NULL;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001693 btrfs_printk(root->fs_info, "Skipping commit of aborted transaction.\n");
1694// WARN_ON(1);
1695 if (current->journal_info == trans)
1696 current->journal_info = NULL;
Josef Bacik7b8b92a2012-05-31 15:52:43 -04001697 cleanup_transaction(trans, root, ret);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001698
1699 return ret;
Chris Mason79154b12007-03-22 15:59:16 -04001700}
1701
Chris Masond352ac62008-09-29 15:18:18 -04001702/*
1703 * interface function to delete all the snapshots we have scheduled for deletion
1704 */
Chris Masone9d0b132007-08-10 14:06:19 -04001705int btrfs_clean_old_snapshots(struct btrfs_root *root)
1706{
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001707 LIST_HEAD(list);
1708 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone9d0b132007-08-10 14:06:19 -04001709
Josef Bacika4abeea2011-04-11 17:25:13 -04001710 spin_lock(&fs_info->trans_lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001711 list_splice_init(&fs_info->dead_roots, &list);
Josef Bacika4abeea2011-04-11 17:25:13 -04001712 spin_unlock(&fs_info->trans_lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001713
1714 while (!list_empty(&list)) {
Jeff Mahoney2c536792011-10-03 23:22:41 -04001715 int ret;
1716
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001717 root = list_entry(list.next, struct btrfs_root, root_list);
Yan, Zheng76dda932009-09-21 16:00:26 -04001718 list_del(&root->root_list);
1719
Miao Xie16cdcec2011-04-22 18:12:22 +08001720 btrfs_kill_all_delayed_nodes(root);
1721
Yan, Zheng76dda932009-09-21 16:00:26 -04001722 if (btrfs_header_backref_rev(root->node) <
1723 BTRFS_MIXED_BACKREF_REV)
Jeff Mahoney2c536792011-10-03 23:22:41 -04001724 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
Yan, Zheng76dda932009-09-21 16:00:26 -04001725 else
Jeff Mahoney2c536792011-10-03 23:22:41 -04001726 ret =btrfs_drop_snapshot(root, NULL, 1, 0);
1727 BUG_ON(ret < 0);
Chris Masone9d0b132007-08-10 14:06:19 -04001728 }
1729 return 0;
1730}