blob: 51dcec86757f071654bc3866123e65157ef0286b [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 Masond3c2fdc2007-09-17 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>
Chris Mason79154b12007-03-22 15:59:16 -040025#include "ctree.h"
26#include "disk-io.h"
27#include "transaction.h"
Chris Mason925baed2008-06-25 16:01:30 -040028#include "locking.h"
Chris Masone02119d2008-09-05 16:13:11 -040029#include "tree-log.h"
Li Zefan581bb052011-04-20 10:06:11 +080030#include "inode-map.h"
Chris Mason79154b12007-03-22 15:59:16 -040031
Chris Mason0f7d52f2007-04-09 10:42:37 -040032#define BTRFS_ROOT_TRANS_TAG 0
33
Chris Mason80b67942008-02-01 16:35:04 -050034static noinline void put_transaction(struct btrfs_transaction *transaction)
Chris Mason79154b12007-03-22 15:59:16 -040035{
Josef Bacik13c5a932011-04-11 15:45:29 -040036 WARN_ON(atomic_read(&transaction->use_count) == 0);
37 if (atomic_dec_and_test(&transaction->use_count)) {
Josef Bacika4abeea2011-04-11 17:25:13 -040038 BUG_ON(!list_empty(&transaction->list));
Chris Mason2c90e5d2007-04-02 10:50:19 -040039 memset(transaction, 0, sizeof(*transaction));
40 kmem_cache_free(btrfs_transaction_cachep, transaction);
Chris Mason78fae272007-03-25 11:35:08 -040041 }
Chris Mason79154b12007-03-22 15:59:16 -040042}
43
Josef Bacik817d52f2009-07-13 21:29:25 -040044static noinline void switch_commit_root(struct btrfs_root *root)
45{
Josef Bacik817d52f2009-07-13 21:29:25 -040046 free_extent_buffer(root->commit_root);
47 root->commit_root = btrfs_root_node(root);
Josef Bacik817d52f2009-07-13 21:29:25 -040048}
49
Chris Masond352ac62008-09-29 15:18:18 -040050/*
51 * either allocate a new transaction or hop into the existing one
52 */
Josef Bacika4abeea2011-04-11 17:25:13 -040053static noinline int join_transaction(struct btrfs_root *root, int nofail)
Chris Mason79154b12007-03-22 15:59:16 -040054{
55 struct btrfs_transaction *cur_trans;
Josef Bacika4abeea2011-04-11 17:25:13 -040056
57 spin_lock(&root->fs_info->trans_lock);
58 if (root->fs_info->trans_no_join) {
59 if (!nofail) {
60 spin_unlock(&root->fs_info->trans_lock);
61 return -EBUSY;
62 }
63 }
64
Chris Mason79154b12007-03-22 15:59:16 -040065 cur_trans = root->fs_info->running_transaction;
Josef Bacika4abeea2011-04-11 17:25:13 -040066 if (cur_trans) {
67 atomic_inc(&cur_trans->use_count);
Josef Bacik13c5a932011-04-11 15:45:29 -040068 atomic_inc(&cur_trans->num_writers);
Josef Bacik15ee9bc2007-08-10 16:22:09 -040069 cur_trans->num_joined++;
Josef Bacika4abeea2011-04-11 17:25:13 -040070 spin_unlock(&root->fs_info->trans_lock);
71 return 0;
Chris Mason79154b12007-03-22 15:59:16 -040072 }
Josef Bacika4abeea2011-04-11 17:25:13 -040073 spin_unlock(&root->fs_info->trans_lock);
74
75 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS);
76 if (!cur_trans)
77 return -ENOMEM;
78 spin_lock(&root->fs_info->trans_lock);
79 if (root->fs_info->running_transaction) {
80 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
81 cur_trans = root->fs_info->running_transaction;
82 atomic_inc(&cur_trans->use_count);
83 atomic_inc(&cur_trans->num_writers);
84 cur_trans->num_joined++;
85 spin_unlock(&root->fs_info->trans_lock);
86 return 0;
87 }
88 atomic_set(&cur_trans->num_writers, 1);
89 cur_trans->num_joined = 0;
90 init_waitqueue_head(&cur_trans->writer_wait);
91 init_waitqueue_head(&cur_trans->commit_wait);
92 cur_trans->in_commit = 0;
93 cur_trans->blocked = 0;
94 /*
95 * One for this trans handle, one so it will live on until we
96 * commit the transaction.
97 */
98 atomic_set(&cur_trans->use_count, 2);
99 cur_trans->commit_done = 0;
100 cur_trans->start_time = get_seconds();
101
102 cur_trans->delayed_refs.root = RB_ROOT;
103 cur_trans->delayed_refs.num_entries = 0;
104 cur_trans->delayed_refs.num_heads_ready = 0;
105 cur_trans->delayed_refs.num_heads = 0;
106 cur_trans->delayed_refs.flushing = 0;
107 cur_trans->delayed_refs.run_delayed_start = 0;
108 spin_lock_init(&cur_trans->commit_lock);
109 spin_lock_init(&cur_trans->delayed_refs.lock);
110
111 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
112 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
113 extent_io_tree_init(&cur_trans->dirty_pages,
Chris Masonff5714c2011-05-28 07:00:39 -0400114 root->fs_info->btree_inode->i_mapping);
Josef Bacika4abeea2011-04-11 17:25:13 -0400115 root->fs_info->generation++;
116 cur_trans->transid = root->fs_info->generation;
117 root->fs_info->running_transaction = cur_trans;
118 spin_unlock(&root->fs_info->trans_lock);
Josef Bacik15ee9bc2007-08-10 16:22:09 -0400119
Chris Mason79154b12007-03-22 15:59:16 -0400120 return 0;
121}
122
Chris Masond352ac62008-09-29 15:18:18 -0400123/*
Chris Masond3977122009-01-05 21:25:51 -0500124 * this does all the record keeping required to make sure that a reference
125 * counted root is properly recorded in a given transaction. This is required
126 * to make sure the old root from before we joined the transaction is deleted
127 * when the transaction commits
Chris Masond352ac62008-09-29 15:18:18 -0400128 */
Chris Mason75857172011-06-13 20:00:16 -0400129static int record_root_in_trans(struct btrfs_trans_handle *trans,
Josef Bacika4abeea2011-04-11 17:25:13 -0400130 struct btrfs_root *root)
Chris Mason6702ed42007-08-07 16:15:09 -0400131{
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400132 if (root->ref_cows && root->last_trans < trans->transid) {
Chris Mason6702ed42007-08-07 16:15:09 -0400133 WARN_ON(root == root->fs_info->extent_root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400134 WARN_ON(root->commit_root != root->node);
Yan Zheng31153d82008-07-28 15:32:19 -0400135
Chris Mason75857172011-06-13 20:00:16 -0400136 /*
137 * see below for in_trans_setup usage rules
138 * we have the reloc mutex held now, so there
139 * is only one writer in this function
140 */
141 root->in_trans_setup = 1;
142
143 /* make sure readers find in_trans_setup before
144 * they find our root->last_trans update
145 */
146 smp_wmb();
147
Josef Bacika4abeea2011-04-11 17:25:13 -0400148 spin_lock(&root->fs_info->fs_roots_radix_lock);
149 if (root->last_trans == trans->transid) {
150 spin_unlock(&root->fs_info->fs_roots_radix_lock);
151 return 0;
152 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400153 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
154 (unsigned long)root->root_key.objectid,
155 BTRFS_ROOT_TRANS_TAG);
Josef Bacika4abeea2011-04-11 17:25:13 -0400156 spin_unlock(&root->fs_info->fs_roots_radix_lock);
Chris Mason75857172011-06-13 20:00:16 -0400157 root->last_trans = trans->transid;
158
159 /* this is pretty tricky. We don't want to
160 * take the relocation lock in btrfs_record_root_in_trans
161 * unless we're really doing the first setup for this root in
162 * this transaction.
163 *
164 * Normally we'd use root->last_trans as a flag to decide
165 * if we want to take the expensive mutex.
166 *
167 * But, we have to set root->last_trans before we
168 * init the relocation root, otherwise, we trip over warnings
169 * in ctree.c. The solution used here is to flag ourselves
170 * with root->in_trans_setup. When this is 1, we're still
171 * fixing up the reloc trees and everyone must wait.
172 *
173 * When this is zero, they can trust root->last_trans and fly
174 * through btrfs_record_root_in_trans without having to take the
175 * lock. smp_wmb() makes sure that all the writes above are
176 * done before we pop in the zero below
177 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400178 btrfs_init_reloc_root(trans, root);
Chris Mason75857172011-06-13 20:00:16 -0400179 smp_wmb();
180 root->in_trans_setup = 0;
Chris Mason6702ed42007-08-07 16:15:09 -0400181 }
182 return 0;
183}
184
Chris Mason75857172011-06-13 20:00:16 -0400185
186int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
187 struct btrfs_root *root)
188{
189 if (!root->ref_cows)
190 return 0;
191
192 /*
193 * see record_root_in_trans for comments about in_trans_setup usage
194 * and barriers
195 */
196 smp_rmb();
197 if (root->last_trans == trans->transid &&
198 !root->in_trans_setup)
199 return 0;
200
201 mutex_lock(&root->fs_info->reloc_mutex);
202 record_root_in_trans(trans, root);
203 mutex_unlock(&root->fs_info->reloc_mutex);
204
205 return 0;
206}
207
Chris Masond352ac62008-09-29 15:18:18 -0400208/* wait for commit against the current transaction to become unblocked
209 * when this is done, it is safe to start a new transaction, but the current
210 * transaction might not be fully on disk.
211 */
Chris Mason37d1aee2008-07-31 10:48:37 -0400212static void wait_current_trans(struct btrfs_root *root)
Chris Mason79154b12007-03-22 15:59:16 -0400213{
Chris Masonf9295742008-07-17 12:54:14 -0400214 struct btrfs_transaction *cur_trans;
Chris Mason79154b12007-03-22 15:59:16 -0400215
Josef Bacika4abeea2011-04-11 17:25:13 -0400216 spin_lock(&root->fs_info->trans_lock);
Chris Masonf9295742008-07-17 12:54:14 -0400217 cur_trans = root->fs_info->running_transaction;
Chris Mason37d1aee2008-07-31 10:48:37 -0400218 if (cur_trans && cur_trans->blocked) {
Chris Masonf9295742008-07-17 12:54:14 -0400219 DEFINE_WAIT(wait);
Josef Bacik13c5a932011-04-11 15:45:29 -0400220 atomic_inc(&cur_trans->use_count);
Josef Bacika4abeea2011-04-11 17:25:13 -0400221 spin_unlock(&root->fs_info->trans_lock);
Chris Masond3977122009-01-05 21:25:51 -0500222 while (1) {
Chris Masonf9295742008-07-17 12:54:14 -0400223 prepare_to_wait(&root->fs_info->transaction_wait, &wait,
224 TASK_UNINTERRUPTIBLE);
Zhao Lei471fa172010-03-25 12:35:14 +0000225 if (!cur_trans->blocked)
Chris Masonf9295742008-07-17 12:54:14 -0400226 break;
Zhao Lei471fa172010-03-25 12:35:14 +0000227 schedule();
Chris Masonf9295742008-07-17 12:54:14 -0400228 }
Zhao Lei471fa172010-03-25 12:35:14 +0000229 finish_wait(&root->fs_info->transaction_wait, &wait);
Chris Masonf9295742008-07-17 12:54:14 -0400230 put_transaction(cur_trans);
Josef Bacika4abeea2011-04-11 17:25:13 -0400231 } else {
232 spin_unlock(&root->fs_info->trans_lock);
Chris Masonf9295742008-07-17 12:54:14 -0400233 }
Chris Mason37d1aee2008-07-31 10:48:37 -0400234}
235
Josef Bacik249ac1e2009-11-10 21:23:48 -0500236enum btrfs_trans_type {
237 TRANS_START,
238 TRANS_JOIN,
239 TRANS_USERSPACE,
Josef Bacik0af3d002010-06-21 14:48:16 -0400240 TRANS_JOIN_NOLOCK,
Josef Bacik249ac1e2009-11-10 21:23:48 -0500241};
242
Yan, Zhenga22285a2010-05-16 10:48:46 -0400243static int may_wait_transaction(struct btrfs_root *root, int type)
Chris Mason37d1aee2008-07-31 10:48:37 -0400244{
Josef Bacika4abeea2011-04-11 17:25:13 -0400245 if (root->fs_info->log_root_recovering)
246 return 0;
247
248 if (type == TRANS_USERSPACE)
Yan, Zhenga22285a2010-05-16 10:48:46 -0400249 return 1;
Josef Bacika4abeea2011-04-11 17:25:13 -0400250
251 if (type == TRANS_START &&
252 !atomic_read(&root->fs_info->open_ioctl_trans))
253 return 1;
254
Yan, Zhenga22285a2010-05-16 10:48:46 -0400255 return 0;
256}
257
258static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
259 u64 num_items, int type)
260{
261 struct btrfs_trans_handle *h;
262 struct btrfs_transaction *cur_trans;
Josef Bacik06d5a582011-04-05 11:57:27 -0400263 int retries = 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400264 int ret;
liuboacce9522011-01-06 19:30:25 +0800265
266 if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
267 return ERR_PTR(-EROFS);
Josef Bacik2a1eb462011-04-13 15:15:59 -0400268
269 if (current->journal_info) {
270 WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK);
271 h = current->journal_info;
272 h->use_count++;
273 h->orig_rsv = h->block_rsv;
274 h->block_rsv = NULL;
275 goto got_it;
276 }
Yan, Zhenga22285a2010-05-16 10:48:46 -0400277again:
278 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
279 if (!h)
280 return ERR_PTR(-ENOMEM);
281
Yan, Zhenga22285a2010-05-16 10:48:46 -0400282 if (may_wait_transaction(root, type))
Chris Mason37d1aee2008-07-31 10:48:37 -0400283 wait_current_trans(root);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400284
Josef Bacika4abeea2011-04-11 17:25:13 -0400285 do {
286 ret = join_transaction(root, type == TRANS_JOIN_NOLOCK);
287 if (ret == -EBUSY)
288 wait_current_trans(root);
289 } while (ret == -EBUSY);
290
Tsutomu Itohdb5b4932011-03-23 08:14:16 +0000291 if (ret < 0) {
Yoshinori Sano6e8df2a2011-04-03 12:31:28 +0000292 kmem_cache_free(btrfs_trans_handle_cachep, h);
Tsutomu Itohdb5b4932011-03-23 08:14:16 +0000293 return ERR_PTR(ret);
294 }
Chris Mason0f7d52f2007-04-09 10:42:37 -0400295
Yan, Zhenga22285a2010-05-16 10:48:46 -0400296 cur_trans = root->fs_info->running_transaction;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400297
298 h->transid = cur_trans->transid;
299 h->transaction = cur_trans;
Chris Mason79154b12007-03-22 15:59:16 -0400300 h->blocks_used = 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400301 h->bytes_reserved = 0;
Chris Mason56bec292009-03-13 10:10:06 -0400302 h->delayed_ref_updates = 0;
Josef Bacik2a1eb462011-04-13 15:15:59 -0400303 h->use_count = 1;
Yan, Zhengf0486c62010-05-16 10:46:25 -0400304 h->block_rsv = NULL;
Josef Bacik2a1eb462011-04-13 15:15:59 -0400305 h->orig_rsv = NULL;
Chris Masonb7ec40d2009-03-12 20:12:45 -0400306
Yan, Zhenga22285a2010-05-16 10:48:46 -0400307 smp_mb();
308 if (cur_trans->blocked && may_wait_transaction(root, type)) {
309 btrfs_commit_transaction(h, root);
310 goto again;
311 }
Josef Bacik9ed74f22009-09-11 16:12:44 -0400312
Yan, Zhenga22285a2010-05-16 10:48:46 -0400313 if (num_items > 0) {
Josef Bacik8bb8ab22010-10-15 16:52:49 -0400314 ret = btrfs_trans_reserve_metadata(h, root, num_items);
Josef Bacik06d5a582011-04-05 11:57:27 -0400315 if (ret == -EAGAIN && !retries) {
316 retries++;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400317 btrfs_commit_transaction(h, root);
318 goto again;
Josef Bacik06d5a582011-04-05 11:57:27 -0400319 } else if (ret == -EAGAIN) {
320 /*
321 * We have already retried and got EAGAIN, so really we
322 * don't have space, so set ret to -ENOSPC.
323 */
324 ret = -ENOSPC;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400325 }
Josef Bacik06d5a582011-04-05 11:57:27 -0400326
Yan, Zhenga22285a2010-05-16 10:48:46 -0400327 if (ret < 0) {
328 btrfs_end_transaction(h, root);
329 return ERR_PTR(ret);
330 }
331 }
332
Josef Bacik2a1eb462011-04-13 15:15:59 -0400333got_it:
Josef Bacika4abeea2011-04-11 17:25:13 -0400334 btrfs_record_root_in_trans(h, root);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400335
336 if (!current->journal_info && type != TRANS_USERSPACE)
337 current->journal_info = h;
Chris Mason79154b12007-03-22 15:59:16 -0400338 return h;
339}
340
Chris Masonf9295742008-07-17 12:54:14 -0400341struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
Yan, Zhenga22285a2010-05-16 10:48:46 -0400342 int num_items)
Chris Masonf9295742008-07-17 12:54:14 -0400343{
Yan, Zhenga22285a2010-05-16 10:48:46 -0400344 return start_transaction(root, num_items, TRANS_START);
Chris Masonf9295742008-07-17 12:54:14 -0400345}
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400346struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
Chris Masonf9295742008-07-17 12:54:14 -0400347{
Yan, Zhenga22285a2010-05-16 10:48:46 -0400348 return start_transaction(root, 0, TRANS_JOIN);
Chris Masonf9295742008-07-17 12:54:14 -0400349}
350
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400351struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
Josef Bacik0af3d002010-06-21 14:48:16 -0400352{
353 return start_transaction(root, 0, TRANS_JOIN_NOLOCK);
354}
355
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400356struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
Sage Weil9ca9ee02008-08-04 10:41:27 -0400357{
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400358 return start_transaction(root, 0, TRANS_USERSPACE);
Sage Weil9ca9ee02008-08-04 10:41:27 -0400359}
360
Chris Masond352ac62008-09-29 15:18:18 -0400361/* wait for a transaction commit to be fully complete */
Chris Mason89ce8a62008-06-25 16:01:31 -0400362static noinline int wait_for_commit(struct btrfs_root *root,
363 struct btrfs_transaction *commit)
364{
365 DEFINE_WAIT(wait);
Chris Masond3977122009-01-05 21:25:51 -0500366 while (!commit->commit_done) {
Chris Mason89ce8a62008-06-25 16:01:31 -0400367 prepare_to_wait(&commit->commit_wait, &wait,
368 TASK_UNINTERRUPTIBLE);
369 if (commit->commit_done)
370 break;
Chris Mason89ce8a62008-06-25 16:01:31 -0400371 schedule();
Chris Mason89ce8a62008-06-25 16:01:31 -0400372 }
Chris Mason89ce8a62008-06-25 16:01:31 -0400373 finish_wait(&commit->commit_wait, &wait);
374 return 0;
375}
376
Sage Weil46204592010-10-29 15:41:32 -0400377int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
378{
379 struct btrfs_transaction *cur_trans = NULL, *t;
380 int ret;
381
Sage Weil46204592010-10-29 15:41:32 -0400382 ret = 0;
383 if (transid) {
384 if (transid <= root->fs_info->last_trans_committed)
Josef Bacika4abeea2011-04-11 17:25:13 -0400385 goto out;
Sage Weil46204592010-10-29 15:41:32 -0400386
387 /* find specified transaction */
Josef Bacika4abeea2011-04-11 17:25:13 -0400388 spin_lock(&root->fs_info->trans_lock);
Sage Weil46204592010-10-29 15:41:32 -0400389 list_for_each_entry(t, &root->fs_info->trans_list, list) {
390 if (t->transid == transid) {
391 cur_trans = t;
Josef Bacika4abeea2011-04-11 17:25:13 -0400392 atomic_inc(&cur_trans->use_count);
Sage Weil46204592010-10-29 15:41:32 -0400393 break;
394 }
395 if (t->transid > transid)
396 break;
397 }
Josef Bacika4abeea2011-04-11 17:25:13 -0400398 spin_unlock(&root->fs_info->trans_lock);
Sage Weil46204592010-10-29 15:41:32 -0400399 ret = -EINVAL;
400 if (!cur_trans)
Josef Bacika4abeea2011-04-11 17:25:13 -0400401 goto out; /* bad transid */
Sage Weil46204592010-10-29 15:41:32 -0400402 } else {
403 /* find newest transaction that is committing | committed */
Josef Bacika4abeea2011-04-11 17:25:13 -0400404 spin_lock(&root->fs_info->trans_lock);
Sage Weil46204592010-10-29 15:41:32 -0400405 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
406 list) {
407 if (t->in_commit) {
408 if (t->commit_done)
Josef Bacik3473f3c2011-06-09 10:15:17 -0400409 break;
Sage Weil46204592010-10-29 15:41:32 -0400410 cur_trans = t;
Josef Bacika4abeea2011-04-11 17:25:13 -0400411 atomic_inc(&cur_trans->use_count);
Sage Weil46204592010-10-29 15:41:32 -0400412 break;
413 }
414 }
Josef Bacika4abeea2011-04-11 17:25:13 -0400415 spin_unlock(&root->fs_info->trans_lock);
Sage Weil46204592010-10-29 15:41:32 -0400416 if (!cur_trans)
Josef Bacika4abeea2011-04-11 17:25:13 -0400417 goto out; /* nothing committing|committed */
Sage Weil46204592010-10-29 15:41:32 -0400418 }
419
Sage Weil46204592010-10-29 15:41:32 -0400420 wait_for_commit(root, cur_trans);
421
Sage Weil46204592010-10-29 15:41:32 -0400422 put_transaction(cur_trans);
423 ret = 0;
Josef Bacika4abeea2011-04-11 17:25:13 -0400424out:
Sage Weil46204592010-10-29 15:41:32 -0400425 return ret;
426}
427
Chris Mason37d1aee2008-07-31 10:48:37 -0400428void btrfs_throttle(struct btrfs_root *root)
429{
Josef Bacika4abeea2011-04-11 17:25:13 -0400430 if (!atomic_read(&root->fs_info->open_ioctl_trans))
Sage Weil9ca9ee02008-08-04 10:41:27 -0400431 wait_current_trans(root);
Chris Mason37d1aee2008-07-31 10:48:37 -0400432}
433
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400434static int should_end_transaction(struct btrfs_trans_handle *trans,
435 struct btrfs_root *root)
436{
437 int ret;
438 ret = btrfs_block_rsv_check(trans, root,
439 &root->fs_info->global_block_rsv, 0, 5);
440 return ret ? 1 : 0;
441}
442
443int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
444 struct btrfs_root *root)
445{
446 struct btrfs_transaction *cur_trans = trans->transaction;
447 int updates;
448
Josef Bacika4abeea2011-04-11 17:25:13 -0400449 smp_mb();
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400450 if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
451 return 1;
452
453 updates = trans->delayed_ref_updates;
454 trans->delayed_ref_updates = 0;
455 if (updates)
456 btrfs_run_delayed_refs(trans, root, updates);
457
458 return should_end_transaction(trans, root);
459}
460
Chris Mason89ce8a62008-06-25 16:01:31 -0400461static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
Josef Bacik0af3d002010-06-21 14:48:16 -0400462 struct btrfs_root *root, int throttle, int lock)
Chris Mason79154b12007-03-22 15:59:16 -0400463{
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400464 struct btrfs_transaction *cur_trans = trans->transaction;
Chris Masonab78c842008-07-29 16:15:18 -0400465 struct btrfs_fs_info *info = root->fs_info;
Chris Masonc3e69d52009-03-13 10:17:05 -0400466 int count = 0;
Chris Masond6e4a422007-04-06 15:37:36 -0400467
Josef Bacik2a1eb462011-04-13 15:15:59 -0400468 if (--trans->use_count) {
469 trans->block_rsv = trans->orig_rsv;
470 return 0;
471 }
472
Chris Masonc3e69d52009-03-13 10:17:05 -0400473 while (count < 4) {
474 unsigned long cur = trans->delayed_ref_updates;
475 trans->delayed_ref_updates = 0;
476 if (cur &&
477 trans->transaction->delayed_refs.num_heads_ready > 64) {
478 trans->delayed_ref_updates = 0;
Chris Masonb7ec40d2009-03-12 20:12:45 -0400479
480 /*
481 * do a full flush if the transaction is trying
482 * to close
483 */
484 if (trans->transaction->delayed_refs.flushing)
485 cur = 0;
Chris Masonc3e69d52009-03-13 10:17:05 -0400486 btrfs_run_delayed_refs(trans, root, cur);
487 } else {
488 break;
489 }
490 count++;
Chris Mason56bec292009-03-13 10:10:06 -0400491 }
492
Yan, Zhenga22285a2010-05-16 10:48:46 -0400493 btrfs_trans_release_metadata(trans, root);
494
Josef Bacika4abeea2011-04-11 17:25:13 -0400495 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
496 should_end_transaction(trans, root)) {
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400497 trans->transaction->blocked = 1;
Josef Bacika4abeea2011-04-11 17:25:13 -0400498 smp_wmb();
499 }
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400500
Josef Bacik0af3d002010-06-21 14:48:16 -0400501 if (lock && cur_trans->blocked && !cur_trans->in_commit) {
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400502 if (throttle)
503 return btrfs_commit_transaction(trans, root);
504 else
505 wake_up_process(info->transaction_kthread);
506 }
507
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400508 WARN_ON(cur_trans != info->running_transaction);
Josef Bacik13c5a932011-04-11 15:45:29 -0400509 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
510 atomic_dec(&cur_trans->num_writers);
Chris Mason89ce8a62008-06-25 16:01:31 -0400511
Sage Weil99d16cb2010-10-29 15:37:34 -0400512 smp_mb();
Chris Mason79154b12007-03-22 15:59:16 -0400513 if (waitqueue_active(&cur_trans->writer_wait))
514 wake_up(&cur_trans->writer_wait);
Chris Mason79154b12007-03-22 15:59:16 -0400515 put_transaction(cur_trans);
Josef Bacik9ed74f22009-09-11 16:12:44 -0400516
517 if (current->journal_info == trans)
518 current->journal_info = NULL;
Chris Masond6025572007-03-30 14:27:56 -0400519 memset(trans, 0, sizeof(*trans));
Chris Mason2c90e5d2007-04-02 10:50:19 -0400520 kmem_cache_free(btrfs_trans_handle_cachep, trans);
Chris Masonab78c842008-07-29 16:15:18 -0400521
Yan, Zheng24bbcf02009-11-12 09:36:34 +0000522 if (throttle)
523 btrfs_run_delayed_iputs(root);
524
Chris Mason79154b12007-03-22 15:59:16 -0400525 return 0;
526}
527
Chris Mason89ce8a62008-06-25 16:01:31 -0400528int btrfs_end_transaction(struct btrfs_trans_handle *trans,
529 struct btrfs_root *root)
530{
Miao Xie16cdcec2011-04-22 18:12:22 +0800531 int ret;
532
533 ret = __btrfs_end_transaction(trans, root, 0, 1);
534 if (ret)
535 return ret;
536 return 0;
Chris Mason89ce8a62008-06-25 16:01:31 -0400537}
538
539int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
540 struct btrfs_root *root)
541{
Miao Xie16cdcec2011-04-22 18:12:22 +0800542 int ret;
543
544 ret = __btrfs_end_transaction(trans, root, 1, 1);
545 if (ret)
546 return ret;
547 return 0;
Josef Bacik0af3d002010-06-21 14:48:16 -0400548}
549
550int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans,
551 struct btrfs_root *root)
552{
Miao Xie16cdcec2011-04-22 18:12:22 +0800553 int ret;
554
555 ret = __btrfs_end_transaction(trans, root, 0, 0);
556 if (ret)
557 return ret;
558 return 0;
559}
560
561int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans,
562 struct btrfs_root *root)
563{
564 return __btrfs_end_transaction(trans, root, 1, 1);
Chris Mason89ce8a62008-06-25 16:01:31 -0400565}
566
Chris Masond352ac62008-09-29 15:18:18 -0400567/*
568 * when btree blocks are allocated, they have some corresponding bits set for
569 * them in one of two extent_io trees. This is used to make sure all of
Chris Mason690587d2009-10-13 13:29:19 -0400570 * those extents are sent to disk but does not wait on them
Chris Masond352ac62008-09-29 15:18:18 -0400571 */
Chris Mason690587d2009-10-13 13:29:19 -0400572int btrfs_write_marked_extents(struct btrfs_root *root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000573 struct extent_io_tree *dirty_pages, int mark)
Chris Mason79154b12007-03-22 15:59:16 -0400574{
Chris Mason7c4452b2007-04-28 09:29:35 -0400575 int ret;
Chris Mason777e6bd2008-08-15 15:34:15 -0400576 int err = 0;
Chris Mason7c4452b2007-04-28 09:29:35 -0400577 int werr = 0;
578 struct page *page;
Chris Mason7c4452b2007-04-28 09:29:35 -0400579 struct inode *btree_inode = root->fs_info->btree_inode;
Chris Mason777e6bd2008-08-15 15:34:15 -0400580 u64 start = 0;
Chris Mason5f39d392007-10-15 16:14:19 -0400581 u64 end;
582 unsigned long index;
Chris Mason7c4452b2007-04-28 09:29:35 -0400583
Chris Masond3977122009-01-05 21:25:51 -0500584 while (1) {
Chris Mason777e6bd2008-08-15 15:34:15 -0400585 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000586 mark);
Chris Mason5f39d392007-10-15 16:14:19 -0400587 if (ret)
Chris Mason7c4452b2007-04-28 09:29:35 -0400588 break;
Chris Masond3977122009-01-05 21:25:51 -0500589 while (start <= end) {
Chris Mason777e6bd2008-08-15 15:34:15 -0400590 cond_resched();
591
Chris Mason5f39d392007-10-15 16:14:19 -0400592 index = start >> PAGE_CACHE_SHIFT;
Chris Mason35ebb932007-10-30 16:56:53 -0400593 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
Chris Mason4bef0842008-09-08 11:18:08 -0400594 page = find_get_page(btree_inode->i_mapping, index);
Chris Mason7c4452b2007-04-28 09:29:35 -0400595 if (!page)
596 continue;
Chris Mason4bef0842008-09-08 11:18:08 -0400597
598 btree_lock_page_hook(page);
599 if (!page->mapping) {
600 unlock_page(page);
601 page_cache_release(page);
602 continue;
603 }
604
Chris Mason6702ed42007-08-07 16:15:09 -0400605 if (PageWriteback(page)) {
606 if (PageDirty(page))
607 wait_on_page_writeback(page);
608 else {
609 unlock_page(page);
610 page_cache_release(page);
611 continue;
612 }
613 }
Chris Mason7c4452b2007-04-28 09:29:35 -0400614 err = write_one_page(page, 0);
615 if (err)
616 werr = err;
617 page_cache_release(page);
618 }
619 }
Chris Mason690587d2009-10-13 13:29:19 -0400620 if (err)
621 werr = err;
622 return werr;
623}
624
625/*
626 * when btree blocks are allocated, they have some corresponding bits set for
627 * them in one of two extent_io trees. This is used to make sure all of
628 * those extents are on disk for transaction or log commit. We wait
629 * on all the pages and clear them from the dirty pages state tree
630 */
631int btrfs_wait_marked_extents(struct btrfs_root *root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000632 struct extent_io_tree *dirty_pages, int mark)
Chris Mason690587d2009-10-13 13:29:19 -0400633{
634 int ret;
635 int err = 0;
636 int werr = 0;
637 struct page *page;
638 struct inode *btree_inode = root->fs_info->btree_inode;
639 u64 start = 0;
640 u64 end;
641 unsigned long index;
642
Chris Masond3977122009-01-05 21:25:51 -0500643 while (1) {
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000644 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
645 mark);
Chris Mason777e6bd2008-08-15 15:34:15 -0400646 if (ret)
647 break;
648
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000649 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
Chris Masond3977122009-01-05 21:25:51 -0500650 while (start <= end) {
Chris Mason777e6bd2008-08-15 15:34:15 -0400651 index = start >> PAGE_CACHE_SHIFT;
652 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
653 page = find_get_page(btree_inode->i_mapping, index);
654 if (!page)
655 continue;
656 if (PageDirty(page)) {
Chris Mason4bef0842008-09-08 11:18:08 -0400657 btree_lock_page_hook(page);
658 wait_on_page_writeback(page);
Chris Mason777e6bd2008-08-15 15:34:15 -0400659 err = write_one_page(page, 0);
660 if (err)
661 werr = err;
662 }
Chris Mason105d9312008-11-18 12:13:12 -0500663 wait_on_page_writeback(page);
Chris Mason777e6bd2008-08-15 15:34:15 -0400664 page_cache_release(page);
665 cond_resched();
666 }
667 }
Chris Mason7c4452b2007-04-28 09:29:35 -0400668 if (err)
669 werr = err;
670 return werr;
Chris Mason79154b12007-03-22 15:59:16 -0400671}
672
Chris Mason690587d2009-10-13 13:29:19 -0400673/*
674 * when btree blocks are allocated, they have some corresponding bits set for
675 * them in one of two extent_io trees. This is used to make sure all of
676 * those extents are on disk for transaction or log commit
677 */
678int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000679 struct extent_io_tree *dirty_pages, int mark)
Chris Mason690587d2009-10-13 13:29:19 -0400680{
681 int ret;
682 int ret2;
683
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000684 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
685 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
Chris Mason690587d2009-10-13 13:29:19 -0400686 return ret || ret2;
687}
688
Chris Masond0c803c2008-09-11 16:17:57 -0400689int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
690 struct btrfs_root *root)
691{
692 if (!trans || !trans->transaction) {
693 struct inode *btree_inode;
694 btree_inode = root->fs_info->btree_inode;
695 return filemap_write_and_wait(btree_inode->i_mapping);
696 }
697 return btrfs_write_and_wait_marked_extents(root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000698 &trans->transaction->dirty_pages,
699 EXTENT_DIRTY);
Chris Masond0c803c2008-09-11 16:17:57 -0400700}
701
Chris Masond352ac62008-09-29 15:18:18 -0400702/*
703 * this is used to update the root pointer in the tree of tree roots.
704 *
705 * But, in the case of the extent allocation tree, updating the root
706 * pointer may allocate blocks which may change the root of the extent
707 * allocation tree.
708 *
709 * So, this loops and repeats and makes sure the cowonly root didn't
710 * change while the root pointer was being updated in the metadata.
711 */
Chris Mason0b86a832008-03-24 15:01:56 -0400712static int update_cowonly_root(struct btrfs_trans_handle *trans,
713 struct btrfs_root *root)
714{
715 int ret;
716 u64 old_root_bytenr;
Yan, Zheng86b9f2e2009-11-12 09:36:50 +0000717 u64 old_root_used;
Chris Mason0b86a832008-03-24 15:01:56 -0400718 struct btrfs_root *tree_root = root->fs_info->tree_root;
719
Yan, Zheng86b9f2e2009-11-12 09:36:50 +0000720 old_root_used = btrfs_root_used(&root->root_item);
Chris Mason0b86a832008-03-24 15:01:56 -0400721 btrfs_write_dirty_block_groups(trans, root);
Chris Mason56bec292009-03-13 10:10:06 -0400722
Chris Masond3977122009-01-05 21:25:51 -0500723 while (1) {
Chris Mason0b86a832008-03-24 15:01:56 -0400724 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
Yan, Zheng86b9f2e2009-11-12 09:36:50 +0000725 if (old_root_bytenr == root->node->start &&
726 old_root_used == btrfs_root_used(&root->root_item))
Chris Mason0b86a832008-03-24 15:01:56 -0400727 break;
Chris Mason87ef2bb2008-10-30 11:23:27 -0400728
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400729 btrfs_set_root_node(&root->root_item, root->node);
Chris Mason0b86a832008-03-24 15:01:56 -0400730 ret = btrfs_update_root(trans, tree_root,
731 &root->root_key,
732 &root->root_item);
733 BUG_ON(ret);
Chris Mason56bec292009-03-13 10:10:06 -0400734
Yan, Zheng86b9f2e2009-11-12 09:36:50 +0000735 old_root_used = btrfs_root_used(&root->root_item);
Yan Zheng4a8c9a62009-07-22 10:07:05 -0400736 ret = btrfs_write_dirty_block_groups(trans, root);
Chris Mason56bec292009-03-13 10:10:06 -0400737 BUG_ON(ret);
Chris Mason0b86a832008-03-24 15:01:56 -0400738 }
Yan Zheng276e6802009-07-30 09:40:40 -0400739
740 if (root != root->fs_info->extent_root)
741 switch_commit_root(root);
742
Chris Mason0b86a832008-03-24 15:01:56 -0400743 return 0;
744}
745
Chris Masond352ac62008-09-29 15:18:18 -0400746/*
747 * update all the cowonly tree roots on disk
748 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400749static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
750 struct btrfs_root *root)
Chris Mason79154b12007-03-22 15:59:16 -0400751{
Chris Mason79154b12007-03-22 15:59:16 -0400752 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Mason0b86a832008-03-24 15:01:56 -0400753 struct list_head *next;
Yan Zheng84234f32008-10-29 14:49:05 -0400754 struct extent_buffer *eb;
Chris Mason56bec292009-03-13 10:10:06 -0400755 int ret;
Yan Zheng84234f32008-10-29 14:49:05 -0400756
Chris Mason56bec292009-03-13 10:10:06 -0400757 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
758 BUG_ON(ret);
Chris Mason87ef2bb2008-10-30 11:23:27 -0400759
Yan Zheng84234f32008-10-29 14:49:05 -0400760 eb = btrfs_lock_root_node(fs_info->tree_root);
Chris Mason9fa8cfe2009-03-13 10:24:59 -0400761 btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb);
Yan Zheng84234f32008-10-29 14:49:05 -0400762 btrfs_tree_unlock(eb);
763 free_extent_buffer(eb);
Chris Mason79154b12007-03-22 15:59:16 -0400764
Chris Mason56bec292009-03-13 10:10:06 -0400765 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
766 BUG_ON(ret);
Chris Mason87ef2bb2008-10-30 11:23:27 -0400767
Chris Masond3977122009-01-05 21:25:51 -0500768 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
Chris Mason0b86a832008-03-24 15:01:56 -0400769 next = fs_info->dirty_cowonly_roots.next;
770 list_del_init(next);
771 root = list_entry(next, struct btrfs_root, dirty_list);
Chris Mason87ef2bb2008-10-30 11:23:27 -0400772
Chris Mason0b86a832008-03-24 15:01:56 -0400773 update_cowonly_root(trans, root);
Chris Mason79154b12007-03-22 15:59:16 -0400774 }
Yan Zheng276e6802009-07-30 09:40:40 -0400775
776 down_write(&fs_info->extent_commit_sem);
777 switch_commit_root(fs_info->extent_root);
778 up_write(&fs_info->extent_commit_sem);
779
Chris Mason79154b12007-03-22 15:59:16 -0400780 return 0;
781}
782
Chris Masond352ac62008-09-29 15:18:18 -0400783/*
784 * dead roots are old snapshots that need to be deleted. This allocates
785 * a dirty root struct and adds it into the list of dead roots that need to
786 * be deleted
787 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400788int btrfs_add_dead_root(struct btrfs_root *root)
Chris Mason5eda7b52007-06-22 14:16:25 -0400789{
Josef Bacika4abeea2011-04-11 17:25:13 -0400790 spin_lock(&root->fs_info->trans_lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400791 list_add(&root->root_list, &root->fs_info->dead_roots);
Josef Bacika4abeea2011-04-11 17:25:13 -0400792 spin_unlock(&root->fs_info->trans_lock);
Chris Mason5eda7b52007-06-22 14:16:25 -0400793 return 0;
794}
795
Chris Masond352ac62008-09-29 15:18:18 -0400796/*
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400797 * update all the cowonly tree roots on disk
Chris Masond352ac62008-09-29 15:18:18 -0400798 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400799static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
800 struct btrfs_root *root)
Chris Mason0f7d52f2007-04-09 10:42:37 -0400801{
Chris Mason0f7d52f2007-04-09 10:42:37 -0400802 struct btrfs_root *gang[8];
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400803 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Mason0f7d52f2007-04-09 10:42:37 -0400804 int i;
805 int ret;
Chris Mason54aa1f42007-06-22 14:16:25 -0400806 int err = 0;
807
Josef Bacika4abeea2011-04-11 17:25:13 -0400808 spin_lock(&fs_info->fs_roots_radix_lock);
Chris Masond3977122009-01-05 21:25:51 -0500809 while (1) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400810 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
811 (void **)gang, 0,
Chris Mason0f7d52f2007-04-09 10:42:37 -0400812 ARRAY_SIZE(gang),
813 BTRFS_ROOT_TRANS_TAG);
814 if (ret == 0)
815 break;
816 for (i = 0; i < ret; i++) {
817 root = gang[i];
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400818 radix_tree_tag_clear(&fs_info->fs_roots_radix,
819 (unsigned long)root->root_key.objectid,
820 BTRFS_ROOT_TRANS_TAG);
Josef Bacika4abeea2011-04-11 17:25:13 -0400821 spin_unlock(&fs_info->fs_roots_radix_lock);
Yan Zheng31153d82008-07-28 15:32:19 -0400822
Chris Masone02119d2008-09-05 16:13:11 -0400823 btrfs_free_log(trans, root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400824 btrfs_update_reloc_root(trans, root);
Yan, Zhengd68fc572010-05-16 10:49:58 -0400825 btrfs_orphan_commit_root(trans, root);
Chris Masone02119d2008-09-05 16:13:11 -0400826
Li Zefan82d59022011-04-20 10:33:24 +0800827 btrfs_save_ino_cache(root, trans);
828
Yan Zheng978d9102009-06-15 20:01:02 -0400829 if (root->commit_root != root->node) {
Li Zefan581bb052011-04-20 10:06:11 +0800830 mutex_lock(&root->fs_commit_mutex);
Josef Bacik817d52f2009-07-13 21:29:25 -0400831 switch_commit_root(root);
Li Zefan581bb052011-04-20 10:06:11 +0800832 btrfs_unpin_free_ino(root);
833 mutex_unlock(&root->fs_commit_mutex);
834
Yan Zheng978d9102009-06-15 20:01:02 -0400835 btrfs_set_root_node(&root->root_item,
836 root->node);
837 }
Chris Mason9f3a7422007-08-07 15:52:19 -0400838
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400839 err = btrfs_update_root(trans, fs_info->tree_root,
Chris Mason0f7d52f2007-04-09 10:42:37 -0400840 &root->root_key,
841 &root->root_item);
Josef Bacika4abeea2011-04-11 17:25:13 -0400842 spin_lock(&fs_info->fs_roots_radix_lock);
Chris Mason54aa1f42007-06-22 14:16:25 -0400843 if (err)
844 break;
Chris Mason0f7d52f2007-04-09 10:42:37 -0400845 }
846 }
Josef Bacika4abeea2011-04-11 17:25:13 -0400847 spin_unlock(&fs_info->fs_roots_radix_lock);
Chris Mason54aa1f42007-06-22 14:16:25 -0400848 return err;
Chris Mason0f7d52f2007-04-09 10:42:37 -0400849}
850
Chris Masond352ac62008-09-29 15:18:18 -0400851/*
852 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
853 * otherwise every leaf in the btree is read and defragged.
854 */
Chris Masone9d0b132007-08-10 14:06:19 -0400855int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
856{
857 struct btrfs_fs_info *info = root->fs_info;
Chris Masone9d0b132007-08-10 14:06:19 -0400858 struct btrfs_trans_handle *trans;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400859 int ret;
Chris Masond3c2fdc2007-09-17 10:58:06 -0400860 unsigned long nr;
Chris Masone9d0b132007-08-10 14:06:19 -0400861
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400862 if (xchg(&root->defrag_running, 1))
Chris Masone9d0b132007-08-10 14:06:19 -0400863 return 0;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400864
Chris Mason6b800532007-10-15 16:17:34 -0400865 while (1) {
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400866 trans = btrfs_start_transaction(root, 0);
867 if (IS_ERR(trans))
868 return PTR_ERR(trans);
869
Chris Masone9d0b132007-08-10 14:06:19 -0400870 ret = btrfs_defrag_leaves(trans, root, cacheonly);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400871
Chris Masond3c2fdc2007-09-17 10:58:06 -0400872 nr = trans->blocks_used;
Chris Masone9d0b132007-08-10 14:06:19 -0400873 btrfs_end_transaction(trans, root);
Chris Masond3c2fdc2007-09-17 10:58:06 -0400874 btrfs_btree_balance_dirty(info->tree_root, nr);
Chris Masone9d0b132007-08-10 14:06:19 -0400875 cond_resched();
876
David Sterba7841cb22011-05-31 18:07:27 +0200877 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
Chris Masone9d0b132007-08-10 14:06:19 -0400878 break;
879 }
880 root->defrag_running = 0;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400881 return ret;
Chris Masone9d0b132007-08-10 14:06:19 -0400882}
883
Chris Masond352ac62008-09-29 15:18:18 -0400884/*
885 * new snapshots need to be created at a very specific time in the
886 * transaction commit. This does the actual creation
887 */
Chris Mason80b67942008-02-01 16:35:04 -0500888static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
Chris Mason3063d292008-01-08 15:46:30 -0500889 struct btrfs_fs_info *fs_info,
890 struct btrfs_pending_snapshot *pending)
891{
892 struct btrfs_key key;
Chris Mason80b67942008-02-01 16:35:04 -0500893 struct btrfs_root_item *new_root_item;
Chris Mason3063d292008-01-08 15:46:30 -0500894 struct btrfs_root *tree_root = fs_info->tree_root;
895 struct btrfs_root *root = pending->root;
Sage Weil6bdb72d2010-03-15 17:27:13 +0000896 struct btrfs_root *parent_root;
897 struct inode *parent_inode;
Josef Bacik6a912212010-11-20 09:48:00 +0000898 struct dentry *parent;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400899 struct dentry *dentry;
Chris Mason3063d292008-01-08 15:46:30 -0500900 struct extent_buffer *tmp;
Chris Mason925baed2008-06-25 16:01:30 -0400901 struct extent_buffer *old;
Chris Mason3063d292008-01-08 15:46:30 -0500902 int ret;
Yan, Zhengd68fc572010-05-16 10:49:58 -0400903 u64 to_reserve = 0;
Sage Weil6bdb72d2010-03-15 17:27:13 +0000904 u64 index = 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400905 u64 objectid;
Li Zefanb83cc962010-12-20 16:04:08 +0800906 u64 root_flags;
Chris Mason3063d292008-01-08 15:46:30 -0500907
Chris Mason80b67942008-02-01 16:35:04 -0500908 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
909 if (!new_root_item) {
Yan, Zhenga22285a2010-05-16 10:48:46 -0400910 pending->error = -ENOMEM;
Chris Mason80b67942008-02-01 16:35:04 -0500911 goto fail;
912 }
Yan, Zhenga22285a2010-05-16 10:48:46 -0400913
Li Zefan581bb052011-04-20 10:06:11 +0800914 ret = btrfs_find_free_objectid(tree_root, &objectid);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400915 if (ret) {
916 pending->error = ret;
Chris Mason3063d292008-01-08 15:46:30 -0500917 goto fail;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400918 }
Chris Mason3063d292008-01-08 15:46:30 -0500919
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400920 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
Yan, Zhengd68fc572010-05-16 10:49:58 -0400921 btrfs_orphan_pre_snapshot(trans, pending, &to_reserve);
922
923 if (to_reserve > 0) {
924 ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv,
Josef Bacik8bb8ab22010-10-15 16:52:49 -0400925 to_reserve);
Yan, Zhengd68fc572010-05-16 10:49:58 -0400926 if (ret) {
927 pending->error = ret;
928 goto fail;
929 }
930 }
931
Chris Mason3063d292008-01-08 15:46:30 -0500932 key.objectid = objectid;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400933 key.offset = (u64)-1;
934 key.type = BTRFS_ROOT_ITEM_KEY;
Chris Mason3063d292008-01-08 15:46:30 -0500935
Yan, Zhenga22285a2010-05-16 10:48:46 -0400936 trans->block_rsv = &pending->block_rsv;
Sage Weil6bdb72d2010-03-15 17:27:13 +0000937
Yan, Zhenga22285a2010-05-16 10:48:46 -0400938 dentry = pending->dentry;
Josef Bacik6a912212010-11-20 09:48:00 +0000939 parent = dget_parent(dentry);
940 parent_inode = parent->d_inode;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400941 parent_root = BTRFS_I(parent_inode)->root;
Chris Mason75857172011-06-13 20:00:16 -0400942 record_root_in_trans(trans, parent_root);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400943
Sage Weil6bdb72d2010-03-15 17:27:13 +0000944 /*
945 * insert the directory item
946 */
Sage Weil6bdb72d2010-03-15 17:27:13 +0000947 ret = btrfs_set_inode_index(parent_inode, &index);
948 BUG_ON(ret);
949 ret = btrfs_insert_dir_item(trans, parent_root,
Yan, Zhenga22285a2010-05-16 10:48:46 -0400950 dentry->d_name.name, dentry->d_name.len,
Miao Xie16cdcec2011-04-22 18:12:22 +0800951 parent_inode, &key,
Yan, Zhenga22285a2010-05-16 10:48:46 -0400952 BTRFS_FT_DIR, index);
Sage Weil6bdb72d2010-03-15 17:27:13 +0000953 BUG_ON(ret);
954
Yan, Zhenga22285a2010-05-16 10:48:46 -0400955 btrfs_i_size_write(parent_inode, parent_inode->i_size +
956 dentry->d_name.len * 2);
Sage Weil6bdb72d2010-03-15 17:27:13 +0000957 ret = btrfs_update_inode(trans, parent_root, parent_inode);
958 BUG_ON(ret);
959
Chris Masone9993762011-06-17 16:14:09 -0400960 /*
961 * pull in the delayed directory update
962 * and the delayed inode item
963 * otherwise we corrupt the FS during
964 * snapshot
965 */
966 ret = btrfs_run_delayed_items(trans, root);
967 BUG_ON(ret);
968
Chris Mason75857172011-06-13 20:00:16 -0400969 record_root_in_trans(trans, root);
Sage Weil6bdb72d2010-03-15 17:27:13 +0000970 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
971 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
Li Zefan08fe4db2011-03-28 02:01:25 +0000972 btrfs_check_and_init_root_item(new_root_item);
Sage Weil6bdb72d2010-03-15 17:27:13 +0000973
Li Zefanb83cc962010-12-20 16:04:08 +0800974 root_flags = btrfs_root_flags(new_root_item);
975 if (pending->readonly)
976 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
977 else
978 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
979 btrfs_set_root_flags(new_root_item, root_flags);
980
Chris Mason925baed2008-06-25 16:01:30 -0400981 old = btrfs_lock_root_node(root);
Chris Mason9fa8cfe2009-03-13 10:24:59 -0400982 btrfs_cow_block(trans, root, old, NULL, 0, &old);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400983 btrfs_set_lock_blocking(old);
Chris Mason3063d292008-01-08 15:46:30 -0500984
Chris Mason925baed2008-06-25 16:01:30 -0400985 btrfs_copy_root(trans, root, old, &tmp, objectid);
986 btrfs_tree_unlock(old);
987 free_extent_buffer(old);
Chris Mason3063d292008-01-08 15:46:30 -0500988
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400989 btrfs_set_root_node(new_root_item, tmp);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400990 /* record when the snapshot was created in key.offset */
991 key.offset = trans->transid;
992 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
Chris Mason925baed2008-06-25 16:01:30 -0400993 btrfs_tree_unlock(tmp);
Chris Mason3063d292008-01-08 15:46:30 -0500994 free_extent_buffer(tmp);
Chris Mason0660b5a2008-11-17 20:37:39 -0500995 BUG_ON(ret);
996
Yan, Zhenga22285a2010-05-16 10:48:46 -0400997 /*
998 * insert root back/forward references
999 */
1000 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1001 parent_root->root_key.objectid,
Li Zefan33345d012011-04-20 10:31:50 +08001002 btrfs_ino(parent_inode), index,
Yan, Zhenga22285a2010-05-16 10:48:46 -04001003 dentry->d_name.name, dentry->d_name.len);
1004 BUG_ON(ret);
Josef Bacik6a912212010-11-20 09:48:00 +00001005 dput(parent);
Yan, Zhenga22285a2010-05-16 10:48:46 -04001006
1007 key.offset = (u64)-1;
1008 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1009 BUG_ON(IS_ERR(pending->snap));
Yan, Zhengd68fc572010-05-16 10:49:58 -04001010
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001011 btrfs_reloc_post_snapshot(trans, pending);
Yan, Zhengd68fc572010-05-16 10:49:58 -04001012 btrfs_orphan_post_snapshot(trans, pending);
Chris Mason3063d292008-01-08 15:46:30 -05001013fail:
Sage Weil6bdb72d2010-03-15 17:27:13 +00001014 kfree(new_root_item);
Yan, Zhenga22285a2010-05-16 10:48:46 -04001015 btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1);
1016 return 0;
Chris Mason3063d292008-01-08 15:46:30 -05001017}
1018
Chris Masond352ac62008-09-29 15:18:18 -04001019/*
1020 * create all the snapshots we've scheduled for creation
1021 */
Chris Mason80b67942008-02-01 16:35:04 -05001022static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1023 struct btrfs_fs_info *fs_info)
Chris Mason3063d292008-01-08 15:46:30 -05001024{
1025 struct btrfs_pending_snapshot *pending;
1026 struct list_head *head = &trans->transaction->pending_snapshots;
Chris Mason3de45862008-11-17 21:02:50 -05001027 int ret;
1028
Qinghuang Fengc6e30872009-01-21 10:59:08 -05001029 list_for_each_entry(pending, head, list) {
Chris Mason3de45862008-11-17 21:02:50 -05001030 ret = create_pending_snapshot(trans, fs_info, pending);
1031 BUG_ON(ret);
1032 }
1033 return 0;
1034}
1035
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001036static void update_super_roots(struct btrfs_root *root)
1037{
1038 struct btrfs_root_item *root_item;
1039 struct btrfs_super_block *super;
1040
1041 super = &root->fs_info->super_copy;
1042
1043 root_item = &root->fs_info->chunk_root->root_item;
1044 super->chunk_root = root_item->bytenr;
1045 super->chunk_root_generation = root_item->generation;
1046 super->chunk_root_level = root_item->level;
1047
1048 root_item = &root->fs_info->tree_root->root_item;
1049 super->root = root_item->bytenr;
1050 super->generation = root_item->generation;
1051 super->root_level = root_item->level;
Josef Bacik0af3d002010-06-21 14:48:16 -04001052 if (super->cache_generation != 0 || btrfs_test_opt(root, SPACE_CACHE))
1053 super->cache_generation = root_item->generation;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001054}
1055
Chris Masonf36f3042009-07-30 10:04:48 -04001056int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1057{
1058 int ret = 0;
Josef Bacika4abeea2011-04-11 17:25:13 -04001059 spin_lock(&info->trans_lock);
Chris Masonf36f3042009-07-30 10:04:48 -04001060 if (info->running_transaction)
1061 ret = info->running_transaction->in_commit;
Josef Bacika4abeea2011-04-11 17:25:13 -04001062 spin_unlock(&info->trans_lock);
Chris Masonf36f3042009-07-30 10:04:48 -04001063 return ret;
1064}
1065
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001066int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1067{
1068 int ret = 0;
Josef Bacika4abeea2011-04-11 17:25:13 -04001069 spin_lock(&info->trans_lock);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001070 if (info->running_transaction)
1071 ret = info->running_transaction->blocked;
Josef Bacika4abeea2011-04-11 17:25:13 -04001072 spin_unlock(&info->trans_lock);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001073 return ret;
1074}
1075
Sage Weilbb9c12c2010-10-29 15:37:34 -04001076/*
1077 * wait for the current transaction commit to start and block subsequent
1078 * transaction joins
1079 */
1080static void wait_current_trans_commit_start(struct btrfs_root *root,
1081 struct btrfs_transaction *trans)
1082{
1083 DEFINE_WAIT(wait);
1084
1085 if (trans->in_commit)
1086 return;
1087
1088 while (1) {
1089 prepare_to_wait(&root->fs_info->transaction_blocked_wait, &wait,
1090 TASK_UNINTERRUPTIBLE);
1091 if (trans->in_commit) {
1092 finish_wait(&root->fs_info->transaction_blocked_wait,
1093 &wait);
1094 break;
1095 }
Sage Weilbb9c12c2010-10-29 15:37:34 -04001096 schedule();
Sage Weilbb9c12c2010-10-29 15:37:34 -04001097 finish_wait(&root->fs_info->transaction_blocked_wait, &wait);
1098 }
1099}
1100
1101/*
1102 * wait for the current transaction to start and then become unblocked.
1103 * caller holds ref.
1104 */
1105static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1106 struct btrfs_transaction *trans)
1107{
1108 DEFINE_WAIT(wait);
1109
1110 if (trans->commit_done || (trans->in_commit && !trans->blocked))
1111 return;
1112
1113 while (1) {
1114 prepare_to_wait(&root->fs_info->transaction_wait, &wait,
1115 TASK_UNINTERRUPTIBLE);
1116 if (trans->commit_done ||
1117 (trans->in_commit && !trans->blocked)) {
1118 finish_wait(&root->fs_info->transaction_wait,
1119 &wait);
1120 break;
1121 }
Sage Weilbb9c12c2010-10-29 15:37:34 -04001122 schedule();
Sage Weilbb9c12c2010-10-29 15:37:34 -04001123 finish_wait(&root->fs_info->transaction_wait,
1124 &wait);
1125 }
1126}
1127
1128/*
1129 * commit transactions asynchronously. once btrfs_commit_transaction_async
1130 * returns, any subsequent transaction will not be allowed to join.
1131 */
1132struct btrfs_async_commit {
1133 struct btrfs_trans_handle *newtrans;
1134 struct btrfs_root *root;
1135 struct delayed_work work;
1136};
1137
1138static void do_async_commit(struct work_struct *work)
1139{
1140 struct btrfs_async_commit *ac =
1141 container_of(work, struct btrfs_async_commit, work.work);
1142
1143 btrfs_commit_transaction(ac->newtrans, ac->root);
1144 kfree(ac);
1145}
1146
1147int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1148 struct btrfs_root *root,
1149 int wait_for_unblock)
1150{
1151 struct btrfs_async_commit *ac;
1152 struct btrfs_transaction *cur_trans;
1153
1154 ac = kmalloc(sizeof(*ac), GFP_NOFS);
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00001155 if (!ac)
1156 return -ENOMEM;
Sage Weilbb9c12c2010-10-29 15:37:34 -04001157
1158 INIT_DELAYED_WORK(&ac->work, do_async_commit);
1159 ac->root = root;
Josef Bacik7a7eaa42011-04-13 12:54:33 -04001160 ac->newtrans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00001161 if (IS_ERR(ac->newtrans)) {
1162 int err = PTR_ERR(ac->newtrans);
1163 kfree(ac);
1164 return err;
1165 }
Sage Weilbb9c12c2010-10-29 15:37:34 -04001166
1167 /* take transaction reference */
Sage Weilbb9c12c2010-10-29 15:37:34 -04001168 cur_trans = trans->transaction;
Josef Bacik13c5a932011-04-11 15:45:29 -04001169 atomic_inc(&cur_trans->use_count);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001170
1171 btrfs_end_transaction(trans, root);
1172 schedule_delayed_work(&ac->work, 0);
1173
1174 /* wait for transaction to start and unblock */
Sage Weilbb9c12c2010-10-29 15:37:34 -04001175 if (wait_for_unblock)
1176 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1177 else
1178 wait_current_trans_commit_start(root, cur_trans);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001179
Sage Weil38e88052011-06-10 18:43:13 +00001180 if (current->journal_info == trans)
1181 current->journal_info = NULL;
1182
1183 put_transaction(cur_trans);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001184 return 0;
1185}
1186
1187/*
1188 * btrfs_transaction state sequence:
1189 * in_commit = 0, blocked = 0 (initial)
1190 * in_commit = 1, blocked = 1
1191 * blocked = 0
1192 * commit_done = 1
1193 */
Chris Mason79154b12007-03-22 15:59:16 -04001194int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1195 struct btrfs_root *root)
1196{
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001197 unsigned long joined = 0;
Chris Mason79154b12007-03-22 15:59:16 -04001198 struct btrfs_transaction *cur_trans;
Chris Mason8fd17792007-04-19 21:01:03 -04001199 struct btrfs_transaction *prev_trans = NULL;
Chris Mason79154b12007-03-22 15:59:16 -04001200 DEFINE_WAIT(wait);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001201 int ret;
Chris Mason89573b92009-03-12 20:12:45 -04001202 int should_grow = 0;
1203 unsigned long now = get_seconds();
Sage Weildccae992009-04-02 16:59:01 -04001204 int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
Chris Mason79154b12007-03-22 15:59:16 -04001205
Chris Mason5a3f23d2009-03-31 13:27:11 -04001206 btrfs_run_ordered_operations(root, 0);
1207
Chris Mason56bec292009-03-13 10:10:06 -04001208 /* make a pass through all the delayed refs we have so far
1209 * any runnings procs may add more while we are here
1210 */
1211 ret = btrfs_run_delayed_refs(trans, root, 0);
1212 BUG_ON(ret);
1213
Yan, Zhenga22285a2010-05-16 10:48:46 -04001214 btrfs_trans_release_metadata(trans, root);
1215
Chris Masonb7ec40d2009-03-12 20:12:45 -04001216 cur_trans = trans->transaction;
Chris Mason56bec292009-03-13 10:10:06 -04001217 /*
1218 * set the flushing flag so procs in this transaction have to
1219 * start sending their work down.
1220 */
Chris Masonb7ec40d2009-03-12 20:12:45 -04001221 cur_trans->delayed_refs.flushing = 1;
Chris Mason56bec292009-03-13 10:10:06 -04001222
Chris Masonc3e69d52009-03-13 10:17:05 -04001223 ret = btrfs_run_delayed_refs(trans, root, 0);
Chris Mason56bec292009-03-13 10:10:06 -04001224 BUG_ON(ret);
1225
Josef Bacika4abeea2011-04-11 17:25:13 -04001226 spin_lock(&cur_trans->commit_lock);
Chris Masonb7ec40d2009-03-12 20:12:45 -04001227 if (cur_trans->in_commit) {
Josef Bacika4abeea2011-04-11 17:25:13 -04001228 spin_unlock(&cur_trans->commit_lock);
Josef Bacik13c5a932011-04-11 15:45:29 -04001229 atomic_inc(&cur_trans->use_count);
Chris Mason79154b12007-03-22 15:59:16 -04001230 btrfs_end_transaction(trans, root);
Chris Masonccd467d2007-06-28 15:57:36 -04001231
Chris Mason79154b12007-03-22 15:59:16 -04001232 ret = wait_for_commit(root, cur_trans);
1233 BUG_ON(ret);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001234
Chris Mason79154b12007-03-22 15:59:16 -04001235 put_transaction(cur_trans);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001236
Chris Mason79154b12007-03-22 15:59:16 -04001237 return 0;
1238 }
Chris Mason4313b392008-01-03 09:08:48 -05001239
Chris Mason2c90e5d2007-04-02 10:50:19 -04001240 trans->transaction->in_commit = 1;
Chris Masonf9295742008-07-17 12:54:14 -04001241 trans->transaction->blocked = 1;
Josef Bacika4abeea2011-04-11 17:25:13 -04001242 spin_unlock(&cur_trans->commit_lock);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001243 wake_up(&root->fs_info->transaction_blocked_wait);
1244
Josef Bacika4abeea2011-04-11 17:25:13 -04001245 spin_lock(&root->fs_info->trans_lock);
Chris Masonccd467d2007-06-28 15:57:36 -04001246 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1247 prev_trans = list_entry(cur_trans->list.prev,
1248 struct btrfs_transaction, list);
1249 if (!prev_trans->commit_done) {
Josef Bacik13c5a932011-04-11 15:45:29 -04001250 atomic_inc(&prev_trans->use_count);
Josef Bacika4abeea2011-04-11 17:25:13 -04001251 spin_unlock(&root->fs_info->trans_lock);
Chris Masonccd467d2007-06-28 15:57:36 -04001252
1253 wait_for_commit(root, prev_trans);
Chris Masonccd467d2007-06-28 15:57:36 -04001254
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001255 put_transaction(prev_trans);
Josef Bacika4abeea2011-04-11 17:25:13 -04001256 } else {
1257 spin_unlock(&root->fs_info->trans_lock);
Chris Masonccd467d2007-06-28 15:57:36 -04001258 }
Josef Bacika4abeea2011-04-11 17:25:13 -04001259 } else {
1260 spin_unlock(&root->fs_info->trans_lock);
Chris Masonccd467d2007-06-28 15:57:36 -04001261 }
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001262
Chris Mason89573b92009-03-12 20:12:45 -04001263 if (now < cur_trans->start_time || now - cur_trans->start_time < 1)
1264 should_grow = 1;
1265
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001266 do {
Yan Zheng7ea394f2008-08-05 13:05:02 -04001267 int snap_pending = 0;
Josef Bacika4abeea2011-04-11 17:25:13 -04001268
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001269 joined = cur_trans->num_joined;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001270 if (!list_empty(&trans->transaction->pending_snapshots))
1271 snap_pending = 1;
1272
Chris Mason2c90e5d2007-04-02 10:50:19 -04001273 WARN_ON(cur_trans != trans->transaction);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001274
Sage Weil0bdb1db2010-02-19 14:13:50 -08001275 if (flush_on_commit || snap_pending) {
Yan, Zheng24bbcf02009-11-12 09:36:34 +00001276 btrfs_start_delalloc_inodes(root, 1);
1277 ret = btrfs_wait_ordered_extents(root, 0, 1);
Sage Weilebecd3d2009-07-24 13:17:44 -04001278 BUG_ON(ret);
Yan Zheng7ea394f2008-08-05 13:05:02 -04001279 }
1280
Miao Xie16cdcec2011-04-22 18:12:22 +08001281 ret = btrfs_run_delayed_items(trans, root);
1282 BUG_ON(ret);
1283
Chris Mason5a3f23d2009-03-31 13:27:11 -04001284 /*
1285 * rename don't use btrfs_join_transaction, so, once we
1286 * set the transaction to blocked above, we aren't going
1287 * to get any new ordered operations. We can safely run
1288 * it here and no for sure that nothing new will be added
1289 * to the list
1290 */
1291 btrfs_run_ordered_operations(root, 1);
1292
Chris Masoned3b3d3142010-05-25 10:12:41 -04001293 prepare_to_wait(&cur_trans->writer_wait, &wait,
1294 TASK_UNINTERRUPTIBLE);
1295
Josef Bacik13c5a932011-04-11 15:45:29 -04001296 if (atomic_read(&cur_trans->num_writers) > 1)
Sage Weil99d16cb2010-10-29 15:37:34 -04001297 schedule_timeout(MAX_SCHEDULE_TIMEOUT);
1298 else if (should_grow)
1299 schedule_timeout(1);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001300
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001301 finish_wait(&cur_trans->writer_wait, &wait);
Josef Bacik13c5a932011-04-11 15:45:29 -04001302 } while (atomic_read(&cur_trans->num_writers) > 1 ||
Chris Mason89573b92009-03-12 20:12:45 -04001303 (should_grow && cur_trans->num_joined != joined));
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001304
Chris Mason75857172011-06-13 20:00:16 -04001305 /*
Josef Baciked0ca142011-06-14 16:22:15 -04001306 * Ok now we need to make sure to block out any other joins while we
1307 * commit the transaction. We could have started a join before setting
1308 * no_join so make sure to wait for num_writers to == 1 again.
1309 */
1310 spin_lock(&root->fs_info->trans_lock);
1311 root->fs_info->trans_no_join = 1;
1312 spin_unlock(&root->fs_info->trans_lock);
1313 wait_event(cur_trans->writer_wait,
1314 atomic_read(&cur_trans->num_writers) == 1);
1315
Chris Masone038dca2011-06-17 14:16:13 -04001316 /*
Chris Mason75857172011-06-13 20:00:16 -04001317 * the reloc mutex makes sure that we stop
1318 * the balancing code from coming in and moving
1319 * extents around in the middle of the commit
1320 */
1321 mutex_lock(&root->fs_info->reloc_mutex);
1322
Chris Masone9993762011-06-17 16:14:09 -04001323 ret = btrfs_run_delayed_items(trans, root);
Chris Mason3063d292008-01-08 15:46:30 -05001324 BUG_ON(ret);
1325
Chris Masone9993762011-06-17 16:14:09 -04001326 ret = create_pending_snapshots(trans, root->fs_info);
Miao Xie16cdcec2011-04-22 18:12:22 +08001327 BUG_ON(ret);
1328
Chris Mason56bec292009-03-13 10:10:06 -04001329 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1330 BUG_ON(ret);
1331
Chris Masone9993762011-06-17 16:14:09 -04001332 /*
1333 * make sure none of the code above managed to slip in a
1334 * delayed item
1335 */
1336 btrfs_assert_delayed_root_empty(root);
1337
Chris Mason2c90e5d2007-04-02 10:50:19 -04001338 WARN_ON(cur_trans != trans->transaction);
Chris Masondc17ff82008-01-08 15:46:30 -05001339
Arne Jansena2de7332011-03-08 14:14:00 +01001340 btrfs_scrub_pause(root);
Chris Masone02119d2008-09-05 16:13:11 -04001341 /* btrfs_commit_tree_roots is responsible for getting the
1342 * various roots consistent with each other. Every pointer
1343 * in the tree of tree roots has to point to the most up to date
1344 * root for every subvolume and other tree. So, we have to keep
1345 * the tree logging code from jumping in and changing any
1346 * of the trees.
1347 *
1348 * At this point in the commit, there can't be any tree-log
1349 * writers, but a little lower down we drop the trans mutex
1350 * and let new people in. By holding the tree_log_mutex
1351 * from now until after the super is written, we avoid races
1352 * with the tree-log code.
1353 */
1354 mutex_lock(&root->fs_info->tree_log_mutex);
Zheng Yan1a40e232008-09-26 10:09:34 -04001355
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001356 ret = commit_fs_roots(trans, root);
Chris Mason54aa1f42007-06-22 14:16:25 -04001357 BUG_ON(ret);
1358
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001359 /* commit_fs_roots gets rid of all the tree log roots, it is now
Chris Masone02119d2008-09-05 16:13:11 -04001360 * safe to free the root of tree log roots
1361 */
1362 btrfs_free_log_root_tree(trans, root->fs_info);
1363
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001364 ret = commit_cowonly_roots(trans, root);
Chris Mason79154b12007-03-22 15:59:16 -04001365 BUG_ON(ret);
Chris Mason54aa1f42007-06-22 14:16:25 -04001366
Yan Zheng11833d62009-09-11 16:11:19 -04001367 btrfs_prepare_extent_commit(trans, root);
1368
Chris Mason78fae272007-03-25 11:35:08 -04001369 cur_trans = root->fs_info->running_transaction;
Chris Mason5f39d392007-10-15 16:14:19 -04001370
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001371 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1372 root->fs_info->tree_root->node);
Josef Bacik817d52f2009-07-13 21:29:25 -04001373 switch_commit_root(root->fs_info->tree_root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001374
1375 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1376 root->fs_info->chunk_root->node);
Josef Bacik817d52f2009-07-13 21:29:25 -04001377 switch_commit_root(root->fs_info->chunk_root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001378
1379 update_super_roots(root);
Chris Masone02119d2008-09-05 16:13:11 -04001380
1381 if (!root->fs_info->log_root_recovering) {
1382 btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
1383 btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0);
1384 }
1385
Chris Masona061fc82008-05-07 11:43:44 -04001386 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
1387 sizeof(root->fs_info->super_copy));
Chris Masonccd467d2007-06-28 15:57:36 -04001388
Chris Masonf9295742008-07-17 12:54:14 -04001389 trans->transaction->blocked = 0;
Josef Bacika4abeea2011-04-11 17:25:13 -04001390 spin_lock(&root->fs_info->trans_lock);
1391 root->fs_info->running_transaction = NULL;
1392 root->fs_info->trans_no_join = 0;
1393 spin_unlock(&root->fs_info->trans_lock);
Chris Mason75857172011-06-13 20:00:16 -04001394 mutex_unlock(&root->fs_info->reloc_mutex);
Chris Masonb7ec40d2009-03-12 20:12:45 -04001395
Chris Masonf9295742008-07-17 12:54:14 -04001396 wake_up(&root->fs_info->transaction_wait);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001397
Chris Mason79154b12007-03-22 15:59:16 -04001398 ret = btrfs_write_and_wait_transaction(trans, root);
1399 BUG_ON(ret);
Yan Zhenga512bbf2008-12-08 16:46:26 -05001400 write_ctree_super(trans, root, 0);
Chris Mason4313b392008-01-03 09:08:48 -05001401
Chris Masone02119d2008-09-05 16:13:11 -04001402 /*
1403 * the super is written, we can safely allow the tree-loggers
1404 * to go about their business
1405 */
1406 mutex_unlock(&root->fs_info->tree_log_mutex);
1407
Yan Zheng11833d62009-09-11 16:11:19 -04001408 btrfs_finish_extent_commit(trans, root);
Chris Mason4313b392008-01-03 09:08:48 -05001409
Chris Mason2c90e5d2007-04-02 10:50:19 -04001410 cur_trans->commit_done = 1;
Chris Masonb7ec40d2009-03-12 20:12:45 -04001411
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001412 root->fs_info->last_trans_committed = cur_trans->transid;
Josef Bacik817d52f2009-07-13 21:29:25 -04001413
Chris Mason2c90e5d2007-04-02 10:50:19 -04001414 wake_up(&cur_trans->commit_wait);
Chris Mason3de45862008-11-17 21:02:50 -05001415
Josef Bacika4abeea2011-04-11 17:25:13 -04001416 spin_lock(&root->fs_info->trans_lock);
Josef Bacik13c5a932011-04-11 15:45:29 -04001417 list_del_init(&cur_trans->list);
Josef Bacika4abeea2011-04-11 17:25:13 -04001418 spin_unlock(&root->fs_info->trans_lock);
1419
Chris Mason79154b12007-03-22 15:59:16 -04001420 put_transaction(cur_trans);
Chris Mason78fae272007-03-25 11:35:08 -04001421 put_transaction(cur_trans);
Josef Bacik58176a92007-08-29 15:47:34 -04001422
liubo1abe9b82011-03-24 11:18:59 +00001423 trace_btrfs_transaction_commit(root);
1424
Arne Jansena2de7332011-03-08 14:14:00 +01001425 btrfs_scrub_continue(root);
1426
Josef Bacik9ed74f22009-09-11 16:12:44 -04001427 if (current->journal_info == trans)
1428 current->journal_info = NULL;
1429
Chris Mason2c90e5d2007-04-02 10:50:19 -04001430 kmem_cache_free(btrfs_trans_handle_cachep, trans);
Yan, Zheng24bbcf02009-11-12 09:36:34 +00001431
1432 if (current != root->fs_info->transaction_kthread)
1433 btrfs_run_delayed_iputs(root);
1434
Chris Mason79154b12007-03-22 15:59:16 -04001435 return ret;
1436}
1437
Chris Masond352ac62008-09-29 15:18:18 -04001438/*
1439 * interface function to delete all the snapshots we have scheduled for deletion
1440 */
Chris Masone9d0b132007-08-10 14:06:19 -04001441int btrfs_clean_old_snapshots(struct btrfs_root *root)
1442{
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001443 LIST_HEAD(list);
1444 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone9d0b132007-08-10 14:06:19 -04001445
Josef Bacika4abeea2011-04-11 17:25:13 -04001446 spin_lock(&fs_info->trans_lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001447 list_splice_init(&fs_info->dead_roots, &list);
Josef Bacika4abeea2011-04-11 17:25:13 -04001448 spin_unlock(&fs_info->trans_lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001449
1450 while (!list_empty(&list)) {
1451 root = list_entry(list.next, struct btrfs_root, root_list);
Yan, Zheng76dda932009-09-21 16:00:26 -04001452 list_del(&root->root_list);
1453
Miao Xie16cdcec2011-04-22 18:12:22 +08001454 btrfs_kill_all_delayed_nodes(root);
1455
Yan, Zheng76dda932009-09-21 16:00:26 -04001456 if (btrfs_header_backref_rev(root->node) <
1457 BTRFS_MIXED_BACKREF_REV)
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001458 btrfs_drop_snapshot(root, NULL, 0);
Yan, Zheng76dda932009-09-21 16:00:26 -04001459 else
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001460 btrfs_drop_snapshot(root, NULL, 1);
Chris Masone9d0b132007-08-10 14:06:19 -04001461 }
1462 return 0;
1463}