blob: b83ed5e64a325553c85b0c7fff10c0a37821d8f5 [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"
Chris Mason79154b12007-03-22 15:59:16 -040030
Chris Mason0f7d52f2007-04-09 10:42:37 -040031#define BTRFS_ROOT_TRANS_TAG 0
32
Chris Mason80b67942008-02-01 16:35:04 -050033static noinline void put_transaction(struct btrfs_transaction *transaction)
Chris Mason79154b12007-03-22 15:59:16 -040034{
Josef Bacik13c5a932011-04-11 15:45:29 -040035 WARN_ON(atomic_read(&transaction->use_count) == 0);
36 if (atomic_dec_and_test(&transaction->use_count)) {
Chris Mason2c90e5d2007-04-02 10:50:19 -040037 memset(transaction, 0, sizeof(*transaction));
38 kmem_cache_free(btrfs_transaction_cachep, transaction);
Chris Mason78fae272007-03-25 11:35:08 -040039 }
Chris Mason79154b12007-03-22 15:59:16 -040040}
41
Josef Bacik817d52f2009-07-13 21:29:25 -040042static noinline void switch_commit_root(struct btrfs_root *root)
43{
Josef Bacik817d52f2009-07-13 21:29:25 -040044 free_extent_buffer(root->commit_root);
45 root->commit_root = btrfs_root_node(root);
Josef Bacik817d52f2009-07-13 21:29:25 -040046}
47
Chris Masond352ac62008-09-29 15:18:18 -040048/*
49 * either allocate a new transaction or hop into the existing one
50 */
Chris Mason80b67942008-02-01 16:35:04 -050051static noinline int join_transaction(struct btrfs_root *root)
Chris Mason79154b12007-03-22 15:59:16 -040052{
53 struct btrfs_transaction *cur_trans;
54 cur_trans = root->fs_info->running_transaction;
55 if (!cur_trans) {
Chris Mason2c90e5d2007-04-02 10:50:19 -040056 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
57 GFP_NOFS);
Tsutomu Itohdb5b4932011-03-23 08:14:16 +000058 if (!cur_trans)
59 return -ENOMEM;
Chris Mason0f7d52f2007-04-09 10:42:37 -040060 root->fs_info->generation++;
Josef Bacik13c5a932011-04-11 15:45:29 -040061 atomic_set(&cur_trans->num_writers, 1);
Josef Bacik15ee9bc2007-08-10 16:22:09 -040062 cur_trans->num_joined = 0;
Chris Mason0f7d52f2007-04-09 10:42:37 -040063 cur_trans->transid = root->fs_info->generation;
Chris Mason79154b12007-03-22 15:59:16 -040064 init_waitqueue_head(&cur_trans->writer_wait);
65 init_waitqueue_head(&cur_trans->commit_wait);
66 cur_trans->in_commit = 0;
Chris Masonf9295742008-07-17 12:54:14 -040067 cur_trans->blocked = 0;
Josef Bacik13c5a932011-04-11 15:45:29 -040068 atomic_set(&cur_trans->use_count, 1);
Chris Mason79154b12007-03-22 15:59:16 -040069 cur_trans->commit_done = 0;
Chris Mason08607c12007-06-08 15:33:54 -040070 cur_trans->start_time = get_seconds();
Chris Mason56bec292009-03-13 10:10:06 -040071
Eric Paris6bef4d32010-02-23 19:43:04 +000072 cur_trans->delayed_refs.root = RB_ROOT;
Chris Mason56bec292009-03-13 10:10:06 -040073 cur_trans->delayed_refs.num_entries = 0;
Chris Masonc3e69d52009-03-13 10:17:05 -040074 cur_trans->delayed_refs.num_heads_ready = 0;
75 cur_trans->delayed_refs.num_heads = 0;
Chris Mason56bec292009-03-13 10:10:06 -040076 cur_trans->delayed_refs.flushing = 0;
Chris Masonc3e69d52009-03-13 10:17:05 -040077 cur_trans->delayed_refs.run_delayed_start = 0;
Chris Mason56bec292009-03-13 10:10:06 -040078 spin_lock_init(&cur_trans->delayed_refs.lock);
79
Chris Mason3063d292008-01-08 15:46:30 -050080 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
Chris Mason8fd17792007-04-19 21:01:03 -040081 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
Chris Masond1310b22008-01-24 16:13:08 -050082 extent_io_tree_init(&cur_trans->dirty_pages,
Chris Mason5f39d392007-10-15 16:14:19 -040083 root->fs_info->btree_inode->i_mapping,
84 GFP_NOFS);
Chris Mason48ec2cf2008-06-09 09:35:50 -040085 spin_lock(&root->fs_info->new_trans_lock);
86 root->fs_info->running_transaction = cur_trans;
87 spin_unlock(&root->fs_info->new_trans_lock);
Josef Bacik15ee9bc2007-08-10 16:22:09 -040088 } else {
Josef Bacik13c5a932011-04-11 15:45:29 -040089 atomic_inc(&cur_trans->num_writers);
Josef Bacik15ee9bc2007-08-10 16:22:09 -040090 cur_trans->num_joined++;
Chris Mason79154b12007-03-22 15:59:16 -040091 }
Josef Bacik15ee9bc2007-08-10 16:22:09 -040092
Chris Mason79154b12007-03-22 15:59:16 -040093 return 0;
94}
95
Chris Masond352ac62008-09-29 15:18:18 -040096/*
Chris Masond3977122009-01-05 21:25:51 -050097 * this does all the record keeping required to make sure that a reference
98 * counted root is properly recorded in a given transaction. This is required
99 * to make sure the old root from before we joined the transaction is deleted
100 * when the transaction commits
Chris Masond352ac62008-09-29 15:18:18 -0400101 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400102static noinline int record_root_in_trans(struct btrfs_trans_handle *trans,
103 struct btrfs_root *root)
Chris Mason6702ed42007-08-07 16:15:09 -0400104{
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400105 if (root->ref_cows && root->last_trans < trans->transid) {
Chris Mason6702ed42007-08-07 16:15:09 -0400106 WARN_ON(root == root->fs_info->extent_root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400107 WARN_ON(root->commit_root != root->node);
Yan Zheng31153d82008-07-28 15:32:19 -0400108
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400109 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
110 (unsigned long)root->root_key.objectid,
111 BTRFS_ROOT_TRANS_TAG);
112 root->last_trans = trans->transid;
113 btrfs_init_reloc_root(trans, root);
Chris Mason6702ed42007-08-07 16:15:09 -0400114 }
115 return 0;
116}
117
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400118int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
119 struct btrfs_root *root)
120{
121 if (!root->ref_cows)
122 return 0;
123
124 mutex_lock(&root->fs_info->trans_mutex);
125 if (root->last_trans == trans->transid) {
126 mutex_unlock(&root->fs_info->trans_mutex);
127 return 0;
128 }
129
130 record_root_in_trans(trans, root);
131 mutex_unlock(&root->fs_info->trans_mutex);
132 return 0;
133}
134
Chris Masond352ac62008-09-29 15:18:18 -0400135/* wait for commit against the current transaction to become unblocked
136 * when this is done, it is safe to start a new transaction, but the current
137 * transaction might not be fully on disk.
138 */
Chris Mason37d1aee2008-07-31 10:48:37 -0400139static void wait_current_trans(struct btrfs_root *root)
Chris Mason79154b12007-03-22 15:59:16 -0400140{
Chris Masonf9295742008-07-17 12:54:14 -0400141 struct btrfs_transaction *cur_trans;
Chris Mason79154b12007-03-22 15:59:16 -0400142
Chris Masonf9295742008-07-17 12:54:14 -0400143 cur_trans = root->fs_info->running_transaction;
Chris Mason37d1aee2008-07-31 10:48:37 -0400144 if (cur_trans && cur_trans->blocked) {
Chris Masonf9295742008-07-17 12:54:14 -0400145 DEFINE_WAIT(wait);
Josef Bacik13c5a932011-04-11 15:45:29 -0400146 atomic_inc(&cur_trans->use_count);
Chris Masond3977122009-01-05 21:25:51 -0500147 while (1) {
Chris Masonf9295742008-07-17 12:54:14 -0400148 prepare_to_wait(&root->fs_info->transaction_wait, &wait,
149 TASK_UNINTERRUPTIBLE);
Zhao Lei471fa172010-03-25 12:35:14 +0000150 if (!cur_trans->blocked)
Chris Masonf9295742008-07-17 12:54:14 -0400151 break;
Zhao Lei471fa172010-03-25 12:35:14 +0000152 mutex_unlock(&root->fs_info->trans_mutex);
153 schedule();
154 mutex_lock(&root->fs_info->trans_mutex);
Chris Masonf9295742008-07-17 12:54:14 -0400155 }
Zhao Lei471fa172010-03-25 12:35:14 +0000156 finish_wait(&root->fs_info->transaction_wait, &wait);
Chris Masonf9295742008-07-17 12:54:14 -0400157 put_transaction(cur_trans);
158 }
Chris Mason37d1aee2008-07-31 10:48:37 -0400159}
160
Josef Bacik249ac1e2009-11-10 21:23:48 -0500161enum btrfs_trans_type {
162 TRANS_START,
163 TRANS_JOIN,
164 TRANS_USERSPACE,
Josef Bacik0af3d002010-06-21 14:48:16 -0400165 TRANS_JOIN_NOLOCK,
Josef Bacik249ac1e2009-11-10 21:23:48 -0500166};
167
Yan, Zhenga22285a2010-05-16 10:48:46 -0400168static int may_wait_transaction(struct btrfs_root *root, int type)
Chris Mason37d1aee2008-07-31 10:48:37 -0400169{
Chris Mason4bef0842008-09-08 11:18:08 -0400170 if (!root->fs_info->log_root_recovering &&
Josef Bacik249ac1e2009-11-10 21:23:48 -0500171 ((type == TRANS_START && !root->fs_info->open_ioctl_trans) ||
172 type == TRANS_USERSPACE))
Yan, Zhenga22285a2010-05-16 10:48:46 -0400173 return 1;
174 return 0;
175}
176
177static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
178 u64 num_items, int type)
179{
180 struct btrfs_trans_handle *h;
181 struct btrfs_transaction *cur_trans;
Josef Bacik06d5a582011-04-05 11:57:27 -0400182 int retries = 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400183 int ret;
liuboacce9522011-01-06 19:30:25 +0800184
185 if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
186 return ERR_PTR(-EROFS);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400187again:
188 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
189 if (!h)
190 return ERR_PTR(-ENOMEM);
191
Josef Bacik0af3d002010-06-21 14:48:16 -0400192 if (type != TRANS_JOIN_NOLOCK)
193 mutex_lock(&root->fs_info->trans_mutex);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400194 if (may_wait_transaction(root, type))
Chris Mason37d1aee2008-07-31 10:48:37 -0400195 wait_current_trans(root);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400196
Chris Mason79154b12007-03-22 15:59:16 -0400197 ret = join_transaction(root);
Tsutomu Itohdb5b4932011-03-23 08:14:16 +0000198 if (ret < 0) {
Yoshinori Sano6e8df2a2011-04-03 12:31:28 +0000199 kmem_cache_free(btrfs_trans_handle_cachep, h);
Tsutomu Itohdb5b4932011-03-23 08:14:16 +0000200 if (type != TRANS_JOIN_NOLOCK)
201 mutex_unlock(&root->fs_info->trans_mutex);
202 return ERR_PTR(ret);
203 }
Chris Mason0f7d52f2007-04-09 10:42:37 -0400204
Yan, Zhenga22285a2010-05-16 10:48:46 -0400205 cur_trans = root->fs_info->running_transaction;
Josef Bacik13c5a932011-04-11 15:45:29 -0400206 atomic_inc(&cur_trans->use_count);
Josef Bacik0af3d002010-06-21 14:48:16 -0400207 if (type != TRANS_JOIN_NOLOCK)
208 mutex_unlock(&root->fs_info->trans_mutex);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400209
210 h->transid = cur_trans->transid;
211 h->transaction = cur_trans;
Chris Mason79154b12007-03-22 15:59:16 -0400212 h->blocks_used = 0;
Yan Zhengd2fb3432008-12-11 16:30:39 -0500213 h->block_group = 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400214 h->bytes_reserved = 0;
Chris Mason56bec292009-03-13 10:10:06 -0400215 h->delayed_ref_updates = 0;
Yan, Zhengf0486c62010-05-16 10:46:25 -0400216 h->block_rsv = NULL;
Chris Masonb7ec40d2009-03-12 20:12:45 -0400217
Yan, Zhenga22285a2010-05-16 10:48:46 -0400218 smp_mb();
219 if (cur_trans->blocked && may_wait_transaction(root, type)) {
220 btrfs_commit_transaction(h, root);
221 goto again;
222 }
Josef Bacik9ed74f22009-09-11 16:12:44 -0400223
Yan, Zhenga22285a2010-05-16 10:48:46 -0400224 if (num_items > 0) {
Josef Bacik8bb8ab22010-10-15 16:52:49 -0400225 ret = btrfs_trans_reserve_metadata(h, root, num_items);
Josef Bacik06d5a582011-04-05 11:57:27 -0400226 if (ret == -EAGAIN && !retries) {
227 retries++;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400228 btrfs_commit_transaction(h, root);
229 goto again;
Josef Bacik06d5a582011-04-05 11:57:27 -0400230 } else if (ret == -EAGAIN) {
231 /*
232 * We have already retried and got EAGAIN, so really we
233 * don't have space, so set ret to -ENOSPC.
234 */
235 ret = -ENOSPC;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400236 }
Josef Bacik06d5a582011-04-05 11:57:27 -0400237
Yan, Zhenga22285a2010-05-16 10:48:46 -0400238 if (ret < 0) {
239 btrfs_end_transaction(h, root);
240 return ERR_PTR(ret);
241 }
242 }
243
Josef Bacik0af3d002010-06-21 14:48:16 -0400244 if (type != TRANS_JOIN_NOLOCK)
245 mutex_lock(&root->fs_info->trans_mutex);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400246 record_root_in_trans(h, root);
Josef Bacik0af3d002010-06-21 14:48:16 -0400247 if (type != TRANS_JOIN_NOLOCK)
248 mutex_unlock(&root->fs_info->trans_mutex);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400249
250 if (!current->journal_info && type != TRANS_USERSPACE)
251 current->journal_info = h;
Chris Mason79154b12007-03-22 15:59:16 -0400252 return h;
253}
254
Chris Masonf9295742008-07-17 12:54:14 -0400255struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
Yan, Zhenga22285a2010-05-16 10:48:46 -0400256 int num_items)
Chris Masonf9295742008-07-17 12:54:14 -0400257{
Yan, Zhenga22285a2010-05-16 10:48:46 -0400258 return start_transaction(root, num_items, TRANS_START);
Chris Masonf9295742008-07-17 12:54:14 -0400259}
260struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
261 int num_blocks)
262{
Yan, Zhenga22285a2010-05-16 10:48:46 -0400263 return start_transaction(root, 0, TRANS_JOIN);
Chris Masonf9295742008-07-17 12:54:14 -0400264}
265
Josef Bacik0af3d002010-06-21 14:48:16 -0400266struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root,
267 int num_blocks)
268{
269 return start_transaction(root, 0, TRANS_JOIN_NOLOCK);
270}
271
Sage Weil9ca9ee02008-08-04 10:41:27 -0400272struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
273 int num_blocks)
274{
Yan, Zhenga22285a2010-05-16 10:48:46 -0400275 return start_transaction(r, 0, TRANS_USERSPACE);
Sage Weil9ca9ee02008-08-04 10:41:27 -0400276}
277
Chris Masond352ac62008-09-29 15:18:18 -0400278/* wait for a transaction commit to be fully complete */
Chris Mason89ce8a62008-06-25 16:01:31 -0400279static noinline int wait_for_commit(struct btrfs_root *root,
280 struct btrfs_transaction *commit)
281{
282 DEFINE_WAIT(wait);
283 mutex_lock(&root->fs_info->trans_mutex);
Chris Masond3977122009-01-05 21:25:51 -0500284 while (!commit->commit_done) {
Chris Mason89ce8a62008-06-25 16:01:31 -0400285 prepare_to_wait(&commit->commit_wait, &wait,
286 TASK_UNINTERRUPTIBLE);
287 if (commit->commit_done)
288 break;
289 mutex_unlock(&root->fs_info->trans_mutex);
290 schedule();
291 mutex_lock(&root->fs_info->trans_mutex);
292 }
293 mutex_unlock(&root->fs_info->trans_mutex);
294 finish_wait(&commit->commit_wait, &wait);
295 return 0;
296}
297
Sage Weil46204592010-10-29 15:41:32 -0400298int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
299{
300 struct btrfs_transaction *cur_trans = NULL, *t;
301 int ret;
302
303 mutex_lock(&root->fs_info->trans_mutex);
304
305 ret = 0;
306 if (transid) {
307 if (transid <= root->fs_info->last_trans_committed)
308 goto out_unlock;
309
310 /* find specified transaction */
311 list_for_each_entry(t, &root->fs_info->trans_list, list) {
312 if (t->transid == transid) {
313 cur_trans = t;
314 break;
315 }
316 if (t->transid > transid)
317 break;
318 }
319 ret = -EINVAL;
320 if (!cur_trans)
321 goto out_unlock; /* bad transid */
322 } else {
323 /* find newest transaction that is committing | committed */
324 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
325 list) {
326 if (t->in_commit) {
327 if (t->commit_done)
328 goto out_unlock;
329 cur_trans = t;
330 break;
331 }
332 }
333 if (!cur_trans)
334 goto out_unlock; /* nothing committing|committed */
335 }
336
Josef Bacik13c5a932011-04-11 15:45:29 -0400337 atomic_inc(&cur_trans->use_count);
Sage Weil46204592010-10-29 15:41:32 -0400338 mutex_unlock(&root->fs_info->trans_mutex);
339
340 wait_for_commit(root, cur_trans);
341
342 mutex_lock(&root->fs_info->trans_mutex);
343 put_transaction(cur_trans);
344 ret = 0;
345out_unlock:
346 mutex_unlock(&root->fs_info->trans_mutex);
347 return ret;
348}
349
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400350#if 0
Chris Masond352ac62008-09-29 15:18:18 -0400351/*
Chris Masond3977122009-01-05 21:25:51 -0500352 * rate limit against the drop_snapshot code. This helps to slow down new
353 * operations if the drop_snapshot code isn't able to keep up.
Chris Masond352ac62008-09-29 15:18:18 -0400354 */
Chris Mason37d1aee2008-07-31 10:48:37 -0400355static void throttle_on_drops(struct btrfs_root *root)
Chris Masonab78c842008-07-29 16:15:18 -0400356{
357 struct btrfs_fs_info *info = root->fs_info;
Chris Mason2dd3e672008-08-04 08:20:15 -0400358 int harder_count = 0;
Chris Masonab78c842008-07-29 16:15:18 -0400359
Chris Mason2dd3e672008-08-04 08:20:15 -0400360harder:
Chris Masonab78c842008-07-29 16:15:18 -0400361 if (atomic_read(&info->throttles)) {
362 DEFINE_WAIT(wait);
363 int thr;
Chris Masonab78c842008-07-29 16:15:18 -0400364 thr = atomic_read(&info->throttle_gen);
365
366 do {
367 prepare_to_wait(&info->transaction_throttle,
368 &wait, TASK_UNINTERRUPTIBLE);
369 if (!atomic_read(&info->throttles)) {
370 finish_wait(&info->transaction_throttle, &wait);
371 break;
372 }
373 schedule();
374 finish_wait(&info->transaction_throttle, &wait);
375 } while (thr == atomic_read(&info->throttle_gen));
Chris Mason2dd3e672008-08-04 08:20:15 -0400376 harder_count++;
377
378 if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 &&
379 harder_count < 2)
380 goto harder;
381
382 if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 &&
383 harder_count < 10)
384 goto harder;
385
386 if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 &&
387 harder_count < 20)
388 goto harder;
Chris Masonab78c842008-07-29 16:15:18 -0400389 }
390}
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400391#endif
Chris Masonab78c842008-07-29 16:15:18 -0400392
Chris Mason37d1aee2008-07-31 10:48:37 -0400393void btrfs_throttle(struct btrfs_root *root)
394{
395 mutex_lock(&root->fs_info->trans_mutex);
Sage Weil9ca9ee02008-08-04 10:41:27 -0400396 if (!root->fs_info->open_ioctl_trans)
397 wait_current_trans(root);
Chris Mason37d1aee2008-07-31 10:48:37 -0400398 mutex_unlock(&root->fs_info->trans_mutex);
Chris Mason37d1aee2008-07-31 10:48:37 -0400399}
400
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400401static int should_end_transaction(struct btrfs_trans_handle *trans,
402 struct btrfs_root *root)
403{
404 int ret;
405 ret = btrfs_block_rsv_check(trans, root,
406 &root->fs_info->global_block_rsv, 0, 5);
407 return ret ? 1 : 0;
408}
409
410int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
411 struct btrfs_root *root)
412{
413 struct btrfs_transaction *cur_trans = trans->transaction;
414 int updates;
415
416 if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
417 return 1;
418
419 updates = trans->delayed_ref_updates;
420 trans->delayed_ref_updates = 0;
421 if (updates)
422 btrfs_run_delayed_refs(trans, root, updates);
423
424 return should_end_transaction(trans, root);
425}
426
Chris Mason89ce8a62008-06-25 16:01:31 -0400427static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
Josef Bacik0af3d002010-06-21 14:48:16 -0400428 struct btrfs_root *root, int throttle, int lock)
Chris Mason79154b12007-03-22 15:59:16 -0400429{
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400430 struct btrfs_transaction *cur_trans = trans->transaction;
Chris Masonab78c842008-07-29 16:15:18 -0400431 struct btrfs_fs_info *info = root->fs_info;
Chris Masonc3e69d52009-03-13 10:17:05 -0400432 int count = 0;
Chris Masond6e4a422007-04-06 15:37:36 -0400433
Chris Masonc3e69d52009-03-13 10:17:05 -0400434 while (count < 4) {
435 unsigned long cur = trans->delayed_ref_updates;
436 trans->delayed_ref_updates = 0;
437 if (cur &&
438 trans->transaction->delayed_refs.num_heads_ready > 64) {
439 trans->delayed_ref_updates = 0;
Chris Masonb7ec40d2009-03-12 20:12:45 -0400440
441 /*
442 * do a full flush if the transaction is trying
443 * to close
444 */
445 if (trans->transaction->delayed_refs.flushing)
446 cur = 0;
Chris Masonc3e69d52009-03-13 10:17:05 -0400447 btrfs_run_delayed_refs(trans, root, cur);
448 } else {
449 break;
450 }
451 count++;
Chris Mason56bec292009-03-13 10:10:06 -0400452 }
453
Yan, Zhenga22285a2010-05-16 10:48:46 -0400454 btrfs_trans_release_metadata(trans, root);
455
Josef Bacik0af3d002010-06-21 14:48:16 -0400456 if (lock && !root->fs_info->open_ioctl_trans &&
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400457 should_end_transaction(trans, root))
458 trans->transaction->blocked = 1;
459
Josef Bacik0af3d002010-06-21 14:48:16 -0400460 if (lock && cur_trans->blocked && !cur_trans->in_commit) {
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400461 if (throttle)
462 return btrfs_commit_transaction(trans, root);
463 else
464 wake_up_process(info->transaction_kthread);
465 }
466
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400467 WARN_ON(cur_trans != info->running_transaction);
Josef Bacik13c5a932011-04-11 15:45:29 -0400468 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
469 atomic_dec(&cur_trans->num_writers);
Chris Mason89ce8a62008-06-25 16:01:31 -0400470
Sage Weil99d16cb2010-10-29 15:37:34 -0400471 smp_mb();
Chris Mason79154b12007-03-22 15:59:16 -0400472 if (waitqueue_active(&cur_trans->writer_wait))
473 wake_up(&cur_trans->writer_wait);
Chris Mason79154b12007-03-22 15:59:16 -0400474 put_transaction(cur_trans);
Josef Bacik9ed74f22009-09-11 16:12:44 -0400475
476 if (current->journal_info == trans)
477 current->journal_info = NULL;
Chris Masond6025572007-03-30 14:27:56 -0400478 memset(trans, 0, sizeof(*trans));
Chris Mason2c90e5d2007-04-02 10:50:19 -0400479 kmem_cache_free(btrfs_trans_handle_cachep, trans);
Chris Masonab78c842008-07-29 16:15:18 -0400480
Yan, Zheng24bbcf02009-11-12 09:36:34 +0000481 if (throttle)
482 btrfs_run_delayed_iputs(root);
483
Chris Mason79154b12007-03-22 15:59:16 -0400484 return 0;
485}
486
Chris Mason89ce8a62008-06-25 16:01:31 -0400487int btrfs_end_transaction(struct btrfs_trans_handle *trans,
488 struct btrfs_root *root)
489{
Miao Xie16cdcec2011-04-22 18:12:22 +0800490 int ret;
491
492 ret = __btrfs_end_transaction(trans, root, 0, 1);
493 if (ret)
494 return ret;
495 return 0;
Chris Mason89ce8a62008-06-25 16:01:31 -0400496}
497
498int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
499 struct btrfs_root *root)
500{
Miao Xie16cdcec2011-04-22 18:12:22 +0800501 int ret;
502
503 ret = __btrfs_end_transaction(trans, root, 1, 1);
504 if (ret)
505 return ret;
506 return 0;
Josef Bacik0af3d002010-06-21 14:48:16 -0400507}
508
509int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans,
510 struct btrfs_root *root)
511{
Miao Xie16cdcec2011-04-22 18:12:22 +0800512 int ret;
513
514 ret = __btrfs_end_transaction(trans, root, 0, 0);
515 if (ret)
516 return ret;
517 return 0;
518}
519
520int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans,
521 struct btrfs_root *root)
522{
523 return __btrfs_end_transaction(trans, root, 1, 1);
Chris Mason89ce8a62008-06-25 16:01:31 -0400524}
525
Chris Masond352ac62008-09-29 15:18:18 -0400526/*
527 * when btree blocks are allocated, they have some corresponding bits set for
528 * them in one of two extent_io trees. This is used to make sure all of
Chris Mason690587d2009-10-13 13:29:19 -0400529 * those extents are sent to disk but does not wait on them
Chris Masond352ac62008-09-29 15:18:18 -0400530 */
Chris Mason690587d2009-10-13 13:29:19 -0400531int btrfs_write_marked_extents(struct btrfs_root *root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000532 struct extent_io_tree *dirty_pages, int mark)
Chris Mason79154b12007-03-22 15:59:16 -0400533{
Chris Mason7c4452b2007-04-28 09:29:35 -0400534 int ret;
Chris Mason777e6bd2008-08-15 15:34:15 -0400535 int err = 0;
Chris Mason7c4452b2007-04-28 09:29:35 -0400536 int werr = 0;
537 struct page *page;
Chris Mason7c4452b2007-04-28 09:29:35 -0400538 struct inode *btree_inode = root->fs_info->btree_inode;
Chris Mason777e6bd2008-08-15 15:34:15 -0400539 u64 start = 0;
Chris Mason5f39d392007-10-15 16:14:19 -0400540 u64 end;
541 unsigned long index;
Chris Mason7c4452b2007-04-28 09:29:35 -0400542
Chris Masond3977122009-01-05 21:25:51 -0500543 while (1) {
Chris Mason777e6bd2008-08-15 15:34:15 -0400544 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000545 mark);
Chris Mason5f39d392007-10-15 16:14:19 -0400546 if (ret)
Chris Mason7c4452b2007-04-28 09:29:35 -0400547 break;
Chris Masond3977122009-01-05 21:25:51 -0500548 while (start <= end) {
Chris Mason777e6bd2008-08-15 15:34:15 -0400549 cond_resched();
550
Chris Mason5f39d392007-10-15 16:14:19 -0400551 index = start >> PAGE_CACHE_SHIFT;
Chris Mason35ebb932007-10-30 16:56:53 -0400552 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
Chris Mason4bef0842008-09-08 11:18:08 -0400553 page = find_get_page(btree_inode->i_mapping, index);
Chris Mason7c4452b2007-04-28 09:29:35 -0400554 if (!page)
555 continue;
Chris Mason4bef0842008-09-08 11:18:08 -0400556
557 btree_lock_page_hook(page);
558 if (!page->mapping) {
559 unlock_page(page);
560 page_cache_release(page);
561 continue;
562 }
563
Chris Mason6702ed42007-08-07 16:15:09 -0400564 if (PageWriteback(page)) {
565 if (PageDirty(page))
566 wait_on_page_writeback(page);
567 else {
568 unlock_page(page);
569 page_cache_release(page);
570 continue;
571 }
572 }
Chris Mason7c4452b2007-04-28 09:29:35 -0400573 err = write_one_page(page, 0);
574 if (err)
575 werr = err;
576 page_cache_release(page);
577 }
578 }
Chris Mason690587d2009-10-13 13:29:19 -0400579 if (err)
580 werr = err;
581 return werr;
582}
583
584/*
585 * when btree blocks are allocated, they have some corresponding bits set for
586 * them in one of two extent_io trees. This is used to make sure all of
587 * those extents are on disk for transaction or log commit. We wait
588 * on all the pages and clear them from the dirty pages state tree
589 */
590int btrfs_wait_marked_extents(struct btrfs_root *root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000591 struct extent_io_tree *dirty_pages, int mark)
Chris Mason690587d2009-10-13 13:29:19 -0400592{
593 int ret;
594 int err = 0;
595 int werr = 0;
596 struct page *page;
597 struct inode *btree_inode = root->fs_info->btree_inode;
598 u64 start = 0;
599 u64 end;
600 unsigned long index;
601
Chris Masond3977122009-01-05 21:25:51 -0500602 while (1) {
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000603 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
604 mark);
Chris Mason777e6bd2008-08-15 15:34:15 -0400605 if (ret)
606 break;
607
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000608 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
Chris Masond3977122009-01-05 21:25:51 -0500609 while (start <= end) {
Chris Mason777e6bd2008-08-15 15:34:15 -0400610 index = start >> PAGE_CACHE_SHIFT;
611 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
612 page = find_get_page(btree_inode->i_mapping, index);
613 if (!page)
614 continue;
615 if (PageDirty(page)) {
Chris Mason4bef0842008-09-08 11:18:08 -0400616 btree_lock_page_hook(page);
617 wait_on_page_writeback(page);
Chris Mason777e6bd2008-08-15 15:34:15 -0400618 err = write_one_page(page, 0);
619 if (err)
620 werr = err;
621 }
Chris Mason105d9312008-11-18 12:13:12 -0500622 wait_on_page_writeback(page);
Chris Mason777e6bd2008-08-15 15:34:15 -0400623 page_cache_release(page);
624 cond_resched();
625 }
626 }
Chris Mason7c4452b2007-04-28 09:29:35 -0400627 if (err)
628 werr = err;
629 return werr;
Chris Mason79154b12007-03-22 15:59:16 -0400630}
631
Chris Mason690587d2009-10-13 13:29:19 -0400632/*
633 * when btree blocks are allocated, they have some corresponding bits set for
634 * them in one of two extent_io trees. This is used to make sure all of
635 * those extents are on disk for transaction or log commit
636 */
637int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000638 struct extent_io_tree *dirty_pages, int mark)
Chris Mason690587d2009-10-13 13:29:19 -0400639{
640 int ret;
641 int ret2;
642
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000643 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
644 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
Chris Mason690587d2009-10-13 13:29:19 -0400645 return ret || ret2;
646}
647
Chris Masond0c803c2008-09-11 16:17:57 -0400648int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
649 struct btrfs_root *root)
650{
651 if (!trans || !trans->transaction) {
652 struct inode *btree_inode;
653 btree_inode = root->fs_info->btree_inode;
654 return filemap_write_and_wait(btree_inode->i_mapping);
655 }
656 return btrfs_write_and_wait_marked_extents(root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000657 &trans->transaction->dirty_pages,
658 EXTENT_DIRTY);
Chris Masond0c803c2008-09-11 16:17:57 -0400659}
660
Chris Masond352ac62008-09-29 15:18:18 -0400661/*
662 * this is used to update the root pointer in the tree of tree roots.
663 *
664 * But, in the case of the extent allocation tree, updating the root
665 * pointer may allocate blocks which may change the root of the extent
666 * allocation tree.
667 *
668 * So, this loops and repeats and makes sure the cowonly root didn't
669 * change while the root pointer was being updated in the metadata.
670 */
Chris Mason0b86a832008-03-24 15:01:56 -0400671static int update_cowonly_root(struct btrfs_trans_handle *trans,
672 struct btrfs_root *root)
673{
674 int ret;
675 u64 old_root_bytenr;
Yan, Zheng86b9f2e2009-11-12 09:36:50 +0000676 u64 old_root_used;
Chris Mason0b86a832008-03-24 15:01:56 -0400677 struct btrfs_root *tree_root = root->fs_info->tree_root;
678
Yan, Zheng86b9f2e2009-11-12 09:36:50 +0000679 old_root_used = btrfs_root_used(&root->root_item);
Chris Mason0b86a832008-03-24 15:01:56 -0400680 btrfs_write_dirty_block_groups(trans, root);
Chris Mason56bec292009-03-13 10:10:06 -0400681
Chris Masond3977122009-01-05 21:25:51 -0500682 while (1) {
Chris Mason0b86a832008-03-24 15:01:56 -0400683 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
Yan, Zheng86b9f2e2009-11-12 09:36:50 +0000684 if (old_root_bytenr == root->node->start &&
685 old_root_used == btrfs_root_used(&root->root_item))
Chris Mason0b86a832008-03-24 15:01:56 -0400686 break;
Chris Mason87ef2bb2008-10-30 11:23:27 -0400687
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400688 btrfs_set_root_node(&root->root_item, root->node);
Chris Mason0b86a832008-03-24 15:01:56 -0400689 ret = btrfs_update_root(trans, tree_root,
690 &root->root_key,
691 &root->root_item);
692 BUG_ON(ret);
Chris Mason56bec292009-03-13 10:10:06 -0400693
Yan, Zheng86b9f2e2009-11-12 09:36:50 +0000694 old_root_used = btrfs_root_used(&root->root_item);
Yan Zheng4a8c9a62009-07-22 10:07:05 -0400695 ret = btrfs_write_dirty_block_groups(trans, root);
Chris Mason56bec292009-03-13 10:10:06 -0400696 BUG_ON(ret);
Chris Mason0b86a832008-03-24 15:01:56 -0400697 }
Yan Zheng276e6802009-07-30 09:40:40 -0400698
699 if (root != root->fs_info->extent_root)
700 switch_commit_root(root);
701
Chris Mason0b86a832008-03-24 15:01:56 -0400702 return 0;
703}
704
Chris Masond352ac62008-09-29 15:18:18 -0400705/*
706 * update all the cowonly tree roots on disk
707 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400708static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
709 struct btrfs_root *root)
Chris Mason79154b12007-03-22 15:59:16 -0400710{
Chris Mason79154b12007-03-22 15:59:16 -0400711 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Mason0b86a832008-03-24 15:01:56 -0400712 struct list_head *next;
Yan Zheng84234f32008-10-29 14:49:05 -0400713 struct extent_buffer *eb;
Chris Mason56bec292009-03-13 10:10:06 -0400714 int ret;
Yan Zheng84234f32008-10-29 14:49:05 -0400715
Chris Mason56bec292009-03-13 10:10:06 -0400716 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
717 BUG_ON(ret);
Chris Mason87ef2bb2008-10-30 11:23:27 -0400718
Yan Zheng84234f32008-10-29 14:49:05 -0400719 eb = btrfs_lock_root_node(fs_info->tree_root);
Chris Mason9fa8cfe2009-03-13 10:24:59 -0400720 btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb);
Yan Zheng84234f32008-10-29 14:49:05 -0400721 btrfs_tree_unlock(eb);
722 free_extent_buffer(eb);
Chris Mason79154b12007-03-22 15:59:16 -0400723
Chris Mason56bec292009-03-13 10:10:06 -0400724 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
725 BUG_ON(ret);
Chris Mason87ef2bb2008-10-30 11:23:27 -0400726
Chris Masond3977122009-01-05 21:25:51 -0500727 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
Chris Mason0b86a832008-03-24 15:01:56 -0400728 next = fs_info->dirty_cowonly_roots.next;
729 list_del_init(next);
730 root = list_entry(next, struct btrfs_root, dirty_list);
Chris Mason87ef2bb2008-10-30 11:23:27 -0400731
Chris Mason0b86a832008-03-24 15:01:56 -0400732 update_cowonly_root(trans, root);
Chris Mason79154b12007-03-22 15:59:16 -0400733 }
Yan Zheng276e6802009-07-30 09:40:40 -0400734
735 down_write(&fs_info->extent_commit_sem);
736 switch_commit_root(fs_info->extent_root);
737 up_write(&fs_info->extent_commit_sem);
738
Chris Mason79154b12007-03-22 15:59:16 -0400739 return 0;
740}
741
Chris Masond352ac62008-09-29 15:18:18 -0400742/*
743 * dead roots are old snapshots that need to be deleted. This allocates
744 * a dirty root struct and adds it into the list of dead roots that need to
745 * be deleted
746 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400747int btrfs_add_dead_root(struct btrfs_root *root)
Chris Mason5eda7b52007-06-22 14:16:25 -0400748{
Yan Zhengb48652c2008-08-04 23:23:47 -0400749 mutex_lock(&root->fs_info->trans_mutex);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400750 list_add(&root->root_list, &root->fs_info->dead_roots);
Yan Zhengb48652c2008-08-04 23:23:47 -0400751 mutex_unlock(&root->fs_info->trans_mutex);
Chris Mason5eda7b52007-06-22 14:16:25 -0400752 return 0;
753}
754
Chris Masond352ac62008-09-29 15:18:18 -0400755/*
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400756 * update all the cowonly tree roots on disk
Chris Masond352ac62008-09-29 15:18:18 -0400757 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400758static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
759 struct btrfs_root *root)
Chris Mason0f7d52f2007-04-09 10:42:37 -0400760{
Chris Mason0f7d52f2007-04-09 10:42:37 -0400761 struct btrfs_root *gang[8];
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400762 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Mason0f7d52f2007-04-09 10:42:37 -0400763 int i;
764 int ret;
Chris Mason54aa1f42007-06-22 14:16:25 -0400765 int err = 0;
766
Chris Masond3977122009-01-05 21:25:51 -0500767 while (1) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400768 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
769 (void **)gang, 0,
Chris Mason0f7d52f2007-04-09 10:42:37 -0400770 ARRAY_SIZE(gang),
771 BTRFS_ROOT_TRANS_TAG);
772 if (ret == 0)
773 break;
774 for (i = 0; i < ret; i++) {
775 root = gang[i];
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400776 radix_tree_tag_clear(&fs_info->fs_roots_radix,
777 (unsigned long)root->root_key.objectid,
778 BTRFS_ROOT_TRANS_TAG);
Yan Zheng31153d82008-07-28 15:32:19 -0400779
Chris Masone02119d2008-09-05 16:13:11 -0400780 btrfs_free_log(trans, root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400781 btrfs_update_reloc_root(trans, root);
Yan, Zhengd68fc572010-05-16 10:49:58 -0400782 btrfs_orphan_commit_root(trans, root);
Chris Masone02119d2008-09-05 16:13:11 -0400783
Yan Zheng978d9102009-06-15 20:01:02 -0400784 if (root->commit_root != root->node) {
Josef Bacik817d52f2009-07-13 21:29:25 -0400785 switch_commit_root(root);
Yan Zheng978d9102009-06-15 20:01:02 -0400786 btrfs_set_root_node(&root->root_item,
787 root->node);
788 }
Chris Mason9f3a7422007-08-07 15:52:19 -0400789
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400790 err = btrfs_update_root(trans, fs_info->tree_root,
Chris Mason0f7d52f2007-04-09 10:42:37 -0400791 &root->root_key,
792 &root->root_item);
Chris Mason54aa1f42007-06-22 14:16:25 -0400793 if (err)
794 break;
Chris Mason0f7d52f2007-04-09 10:42:37 -0400795 }
796 }
Chris Mason54aa1f42007-06-22 14:16:25 -0400797 return err;
Chris Mason0f7d52f2007-04-09 10:42:37 -0400798}
799
Chris Masond352ac62008-09-29 15:18:18 -0400800/*
801 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
802 * otherwise every leaf in the btree is read and defragged.
803 */
Chris Masone9d0b132007-08-10 14:06:19 -0400804int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
805{
806 struct btrfs_fs_info *info = root->fs_info;
Chris Masone9d0b132007-08-10 14:06:19 -0400807 struct btrfs_trans_handle *trans;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400808 int ret;
Chris Masond3c2fdc2007-09-17 10:58:06 -0400809 unsigned long nr;
Chris Masone9d0b132007-08-10 14:06:19 -0400810
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400811 if (xchg(&root->defrag_running, 1))
Chris Masone9d0b132007-08-10 14:06:19 -0400812 return 0;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400813
Chris Mason6b800532007-10-15 16:17:34 -0400814 while (1) {
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400815 trans = btrfs_start_transaction(root, 0);
816 if (IS_ERR(trans))
817 return PTR_ERR(trans);
818
Chris Masone9d0b132007-08-10 14:06:19 -0400819 ret = btrfs_defrag_leaves(trans, root, cacheonly);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400820
Chris Masond3c2fdc2007-09-17 10:58:06 -0400821 nr = trans->blocks_used;
Chris Masone9d0b132007-08-10 14:06:19 -0400822 btrfs_end_transaction(trans, root);
Chris Masond3c2fdc2007-09-17 10:58:06 -0400823 btrfs_btree_balance_dirty(info->tree_root, nr);
Chris Masone9d0b132007-08-10 14:06:19 -0400824 cond_resched();
825
Chris Mason3f157a22008-06-25 16:01:31 -0400826 if (root->fs_info->closing || ret != -EAGAIN)
Chris Masone9d0b132007-08-10 14:06:19 -0400827 break;
828 }
829 root->defrag_running = 0;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400830 return ret;
Chris Masone9d0b132007-08-10 14:06:19 -0400831}
832
Yan Zheng2c47e6052009-06-27 21:07:35 -0400833#if 0
Chris Masond352ac62008-09-29 15:18:18 -0400834/*
Chris Masonb7ec40d2009-03-12 20:12:45 -0400835 * when dropping snapshots, we generate a ton of delayed refs, and it makes
836 * sense not to join the transaction while it is trying to flush the current
837 * queue of delayed refs out.
838 *
839 * This is used by the drop snapshot code only
840 */
841static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info)
842{
843 DEFINE_WAIT(wait);
844
845 mutex_lock(&info->trans_mutex);
846 while (info->running_transaction &&
847 info->running_transaction->delayed_refs.flushing) {
848 prepare_to_wait(&info->transaction_wait, &wait,
849 TASK_UNINTERRUPTIBLE);
850 mutex_unlock(&info->trans_mutex);
Chris Mason59bc5c72009-04-24 14:39:25 -0400851
Chris Masonb7ec40d2009-03-12 20:12:45 -0400852 schedule();
Chris Mason59bc5c72009-04-24 14:39:25 -0400853
Chris Masonb7ec40d2009-03-12 20:12:45 -0400854 mutex_lock(&info->trans_mutex);
855 finish_wait(&info->transaction_wait, &wait);
856 }
857 mutex_unlock(&info->trans_mutex);
858 return 0;
859}
860
861/*
Chris Masond352ac62008-09-29 15:18:18 -0400862 * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
863 * all of them
864 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400865int btrfs_drop_dead_root(struct btrfs_root *root)
Chris Mason0f7d52f2007-04-09 10:42:37 -0400866{
Chris Mason0f7d52f2007-04-09 10:42:37 -0400867 struct btrfs_trans_handle *trans;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400868 struct btrfs_root *tree_root = root->fs_info->tree_root;
Chris Masond3c2fdc2007-09-17 10:58:06 -0400869 unsigned long nr;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400870 int ret;
Chris Mason9f3a7422007-08-07 15:52:19 -0400871
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400872 while (1) {
873 /*
874 * we don't want to jump in and create a bunch of
875 * delayed refs if the transaction is starting to close
876 */
877 wait_transaction_pre_flush(tree_root->fs_info);
878 trans = btrfs_start_transaction(tree_root, 1);
Josef Bacik58176a92007-08-29 15:47:34 -0400879
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400880 /*
881 * we've joined a transaction, make sure it isn't
882 * closing right now
883 */
884 if (trans->transaction->delayed_refs.flushing) {
885 btrfs_end_transaction(trans, tree_root);
886 continue;
Josef Bacik58176a92007-08-29 15:47:34 -0400887 }
Chris Masona2135012008-06-25 16:01:30 -0400888
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400889 ret = btrfs_drop_snapshot(trans, root);
890 if (ret != -EAGAIN)
Chris Mason54aa1f42007-06-22 14:16:25 -0400891 break;
Chris Masona2135012008-06-25 16:01:30 -0400892
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400893 ret = btrfs_update_root(trans, tree_root,
894 &root->root_key,
895 &root->root_item);
896 if (ret)
897 break;
Yanbcc63ab2008-07-30 16:29:20 -0400898
Chris Masond3c2fdc2007-09-17 10:58:06 -0400899 nr = trans->blocks_used;
Chris Mason0f7d52f2007-04-09 10:42:37 -0400900 ret = btrfs_end_transaction(trans, tree_root);
901 BUG_ON(ret);
Chris Mason5eda7b52007-06-22 14:16:25 -0400902
Chris Masond3c2fdc2007-09-17 10:58:06 -0400903 btrfs_btree_balance_dirty(tree_root, nr);
Chris Mason4dc11902007-10-15 16:18:14 -0400904 cond_resched();
Chris Mason0f7d52f2007-04-09 10:42:37 -0400905 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400906 BUG_ON(ret);
907
908 ret = btrfs_del_root(trans, tree_root, &root->root_key);
909 BUG_ON(ret);
910
911 nr = trans->blocks_used;
912 ret = btrfs_end_transaction(trans, tree_root);
913 BUG_ON(ret);
914
915 free_extent_buffer(root->node);
916 free_extent_buffer(root->commit_root);
917 kfree(root);
918
919 btrfs_btree_balance_dirty(tree_root, nr);
Chris Mason54aa1f42007-06-22 14:16:25 -0400920 return ret;
Chris Mason0f7d52f2007-04-09 10:42:37 -0400921}
Yan Zheng2c47e6052009-06-27 21:07:35 -0400922#endif
Chris Mason0f7d52f2007-04-09 10:42:37 -0400923
Chris Masond352ac62008-09-29 15:18:18 -0400924/*
925 * new snapshots need to be created at a very specific time in the
926 * transaction commit. This does the actual creation
927 */
Chris Mason80b67942008-02-01 16:35:04 -0500928static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
Chris Mason3063d292008-01-08 15:46:30 -0500929 struct btrfs_fs_info *fs_info,
930 struct btrfs_pending_snapshot *pending)
931{
932 struct btrfs_key key;
Chris Mason80b67942008-02-01 16:35:04 -0500933 struct btrfs_root_item *new_root_item;
Chris Mason3063d292008-01-08 15:46:30 -0500934 struct btrfs_root *tree_root = fs_info->tree_root;
935 struct btrfs_root *root = pending->root;
Sage Weil6bdb72d2010-03-15 17:27:13 +0000936 struct btrfs_root *parent_root;
937 struct inode *parent_inode;
Josef Bacik6a912212010-11-20 09:48:00 +0000938 struct dentry *parent;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400939 struct dentry *dentry;
Chris Mason3063d292008-01-08 15:46:30 -0500940 struct extent_buffer *tmp;
Chris Mason925baed2008-06-25 16:01:30 -0400941 struct extent_buffer *old;
Chris Mason3063d292008-01-08 15:46:30 -0500942 int ret;
Yan, Zhengd68fc572010-05-16 10:49:58 -0400943 u64 to_reserve = 0;
Sage Weil6bdb72d2010-03-15 17:27:13 +0000944 u64 index = 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400945 u64 objectid;
Li Zefanb83cc962010-12-20 16:04:08 +0800946 u64 root_flags;
Chris Mason3063d292008-01-08 15:46:30 -0500947
Chris Mason80b67942008-02-01 16:35:04 -0500948 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
949 if (!new_root_item) {
Yan, Zhenga22285a2010-05-16 10:48:46 -0400950 pending->error = -ENOMEM;
Chris Mason80b67942008-02-01 16:35:04 -0500951 goto fail;
952 }
Yan, Zhenga22285a2010-05-16 10:48:46 -0400953
Chris Mason3063d292008-01-08 15:46:30 -0500954 ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400955 if (ret) {
956 pending->error = ret;
Chris Mason3063d292008-01-08 15:46:30 -0500957 goto fail;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400958 }
Chris Mason3063d292008-01-08 15:46:30 -0500959
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400960 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
Yan, Zhengd68fc572010-05-16 10:49:58 -0400961 btrfs_orphan_pre_snapshot(trans, pending, &to_reserve);
962
963 if (to_reserve > 0) {
964 ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv,
Josef Bacik8bb8ab22010-10-15 16:52:49 -0400965 to_reserve);
Yan, Zhengd68fc572010-05-16 10:49:58 -0400966 if (ret) {
967 pending->error = ret;
968 goto fail;
969 }
970 }
971
Chris Mason3063d292008-01-08 15:46:30 -0500972 key.objectid = objectid;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400973 key.offset = (u64)-1;
974 key.type = BTRFS_ROOT_ITEM_KEY;
Chris Mason3063d292008-01-08 15:46:30 -0500975
Yan, Zhenga22285a2010-05-16 10:48:46 -0400976 trans->block_rsv = &pending->block_rsv;
Sage Weil6bdb72d2010-03-15 17:27:13 +0000977
Yan, Zhenga22285a2010-05-16 10:48:46 -0400978 dentry = pending->dentry;
Josef Bacik6a912212010-11-20 09:48:00 +0000979 parent = dget_parent(dentry);
980 parent_inode = parent->d_inode;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400981 parent_root = BTRFS_I(parent_inode)->root;
Sage Weil6bdb72d2010-03-15 17:27:13 +0000982 record_root_in_trans(trans, parent_root);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400983
Sage Weil6bdb72d2010-03-15 17:27:13 +0000984 /*
985 * insert the directory item
986 */
Sage Weil6bdb72d2010-03-15 17:27:13 +0000987 ret = btrfs_set_inode_index(parent_inode, &index);
988 BUG_ON(ret);
989 ret = btrfs_insert_dir_item(trans, parent_root,
Yan, Zhenga22285a2010-05-16 10:48:46 -0400990 dentry->d_name.name, dentry->d_name.len,
Miao Xie16cdcec2011-04-22 18:12:22 +0800991 parent_inode, &key,
Yan, Zhenga22285a2010-05-16 10:48:46 -0400992 BTRFS_FT_DIR, index);
Sage Weil6bdb72d2010-03-15 17:27:13 +0000993 BUG_ON(ret);
994
Yan, Zhenga22285a2010-05-16 10:48:46 -0400995 btrfs_i_size_write(parent_inode, parent_inode->i_size +
996 dentry->d_name.len * 2);
Sage Weil6bdb72d2010-03-15 17:27:13 +0000997 ret = btrfs_update_inode(trans, parent_root, parent_inode);
998 BUG_ON(ret);
999
1000 record_root_in_trans(trans, root);
1001 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1002 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
Li Zefan08fe4db2011-03-28 02:01:25 +00001003 btrfs_check_and_init_root_item(new_root_item);
Sage Weil6bdb72d2010-03-15 17:27:13 +00001004
Li Zefanb83cc962010-12-20 16:04:08 +08001005 root_flags = btrfs_root_flags(new_root_item);
1006 if (pending->readonly)
1007 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1008 else
1009 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1010 btrfs_set_root_flags(new_root_item, root_flags);
1011
Chris Mason925baed2008-06-25 16:01:30 -04001012 old = btrfs_lock_root_node(root);
Chris Mason9fa8cfe2009-03-13 10:24:59 -04001013 btrfs_cow_block(trans, root, old, NULL, 0, &old);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001014 btrfs_set_lock_blocking(old);
Chris Mason3063d292008-01-08 15:46:30 -05001015
Chris Mason925baed2008-06-25 16:01:30 -04001016 btrfs_copy_root(trans, root, old, &tmp, objectid);
1017 btrfs_tree_unlock(old);
1018 free_extent_buffer(old);
Chris Mason3063d292008-01-08 15:46:30 -05001019
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001020 btrfs_set_root_node(new_root_item, tmp);
Yan, Zhenga22285a2010-05-16 10:48:46 -04001021 /* record when the snapshot was created in key.offset */
1022 key.offset = trans->transid;
1023 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
Chris Mason925baed2008-06-25 16:01:30 -04001024 btrfs_tree_unlock(tmp);
Chris Mason3063d292008-01-08 15:46:30 -05001025 free_extent_buffer(tmp);
Chris Mason0660b5a2008-11-17 20:37:39 -05001026 BUG_ON(ret);
1027
Yan, Zhenga22285a2010-05-16 10:48:46 -04001028 /*
1029 * insert root back/forward references
1030 */
1031 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1032 parent_root->root_key.objectid,
1033 parent_inode->i_ino, index,
1034 dentry->d_name.name, dentry->d_name.len);
1035 BUG_ON(ret);
Josef Bacik6a912212010-11-20 09:48:00 +00001036 dput(parent);
Yan, Zhenga22285a2010-05-16 10:48:46 -04001037
1038 key.offset = (u64)-1;
1039 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1040 BUG_ON(IS_ERR(pending->snap));
Yan, Zhengd68fc572010-05-16 10:49:58 -04001041
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001042 btrfs_reloc_post_snapshot(trans, pending);
Yan, Zhengd68fc572010-05-16 10:49:58 -04001043 btrfs_orphan_post_snapshot(trans, pending);
Chris Mason3063d292008-01-08 15:46:30 -05001044fail:
Sage Weil6bdb72d2010-03-15 17:27:13 +00001045 kfree(new_root_item);
Yan, Zhenga22285a2010-05-16 10:48:46 -04001046 btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1);
1047 return 0;
Chris Mason3063d292008-01-08 15:46:30 -05001048}
1049
Chris Masond352ac62008-09-29 15:18:18 -04001050/*
1051 * create all the snapshots we've scheduled for creation
1052 */
Chris Mason80b67942008-02-01 16:35:04 -05001053static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1054 struct btrfs_fs_info *fs_info)
Chris Mason3063d292008-01-08 15:46:30 -05001055{
1056 struct btrfs_pending_snapshot *pending;
1057 struct list_head *head = &trans->transaction->pending_snapshots;
Chris Mason3de45862008-11-17 21:02:50 -05001058 int ret;
1059
Qinghuang Fengc6e30872009-01-21 10:59:08 -05001060 list_for_each_entry(pending, head, list) {
Miao Xie16cdcec2011-04-22 18:12:22 +08001061 /*
1062 * We must deal with the delayed items before creating
1063 * snapshots, or we will create a snapthot with inconsistent
1064 * information.
1065 */
1066 ret = btrfs_run_delayed_items(trans, fs_info->fs_root);
1067 BUG_ON(ret);
1068
Chris Mason3de45862008-11-17 21:02:50 -05001069 ret = create_pending_snapshot(trans, fs_info, pending);
1070 BUG_ON(ret);
1071 }
1072 return 0;
1073}
1074
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001075static void update_super_roots(struct btrfs_root *root)
1076{
1077 struct btrfs_root_item *root_item;
1078 struct btrfs_super_block *super;
1079
1080 super = &root->fs_info->super_copy;
1081
1082 root_item = &root->fs_info->chunk_root->root_item;
1083 super->chunk_root = root_item->bytenr;
1084 super->chunk_root_generation = root_item->generation;
1085 super->chunk_root_level = root_item->level;
1086
1087 root_item = &root->fs_info->tree_root->root_item;
1088 super->root = root_item->bytenr;
1089 super->generation = root_item->generation;
1090 super->root_level = root_item->level;
Josef Bacik0af3d002010-06-21 14:48:16 -04001091 if (super->cache_generation != 0 || btrfs_test_opt(root, SPACE_CACHE))
1092 super->cache_generation = root_item->generation;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001093}
1094
Chris Masonf36f3042009-07-30 10:04:48 -04001095int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1096{
1097 int ret = 0;
1098 spin_lock(&info->new_trans_lock);
1099 if (info->running_transaction)
1100 ret = info->running_transaction->in_commit;
1101 spin_unlock(&info->new_trans_lock);
1102 return ret;
1103}
1104
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001105int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1106{
1107 int ret = 0;
1108 spin_lock(&info->new_trans_lock);
1109 if (info->running_transaction)
1110 ret = info->running_transaction->blocked;
1111 spin_unlock(&info->new_trans_lock);
1112 return ret;
1113}
1114
Sage Weilbb9c12c2010-10-29 15:37:34 -04001115/*
1116 * wait for the current transaction commit to start and block subsequent
1117 * transaction joins
1118 */
1119static void wait_current_trans_commit_start(struct btrfs_root *root,
1120 struct btrfs_transaction *trans)
1121{
1122 DEFINE_WAIT(wait);
1123
1124 if (trans->in_commit)
1125 return;
1126
1127 while (1) {
1128 prepare_to_wait(&root->fs_info->transaction_blocked_wait, &wait,
1129 TASK_UNINTERRUPTIBLE);
1130 if (trans->in_commit) {
1131 finish_wait(&root->fs_info->transaction_blocked_wait,
1132 &wait);
1133 break;
1134 }
1135 mutex_unlock(&root->fs_info->trans_mutex);
1136 schedule();
1137 mutex_lock(&root->fs_info->trans_mutex);
1138 finish_wait(&root->fs_info->transaction_blocked_wait, &wait);
1139 }
1140}
1141
1142/*
1143 * wait for the current transaction to start and then become unblocked.
1144 * caller holds ref.
1145 */
1146static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1147 struct btrfs_transaction *trans)
1148{
1149 DEFINE_WAIT(wait);
1150
1151 if (trans->commit_done || (trans->in_commit && !trans->blocked))
1152 return;
1153
1154 while (1) {
1155 prepare_to_wait(&root->fs_info->transaction_wait, &wait,
1156 TASK_UNINTERRUPTIBLE);
1157 if (trans->commit_done ||
1158 (trans->in_commit && !trans->blocked)) {
1159 finish_wait(&root->fs_info->transaction_wait,
1160 &wait);
1161 break;
1162 }
1163 mutex_unlock(&root->fs_info->trans_mutex);
1164 schedule();
1165 mutex_lock(&root->fs_info->trans_mutex);
1166 finish_wait(&root->fs_info->transaction_wait,
1167 &wait);
1168 }
1169}
1170
1171/*
1172 * commit transactions asynchronously. once btrfs_commit_transaction_async
1173 * returns, any subsequent transaction will not be allowed to join.
1174 */
1175struct btrfs_async_commit {
1176 struct btrfs_trans_handle *newtrans;
1177 struct btrfs_root *root;
1178 struct delayed_work work;
1179};
1180
1181static void do_async_commit(struct work_struct *work)
1182{
1183 struct btrfs_async_commit *ac =
1184 container_of(work, struct btrfs_async_commit, work.work);
1185
1186 btrfs_commit_transaction(ac->newtrans, ac->root);
1187 kfree(ac);
1188}
1189
1190int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1191 struct btrfs_root *root,
1192 int wait_for_unblock)
1193{
1194 struct btrfs_async_commit *ac;
1195 struct btrfs_transaction *cur_trans;
1196
1197 ac = kmalloc(sizeof(*ac), GFP_NOFS);
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00001198 if (!ac)
1199 return -ENOMEM;
Sage Weilbb9c12c2010-10-29 15:37:34 -04001200
1201 INIT_DELAYED_WORK(&ac->work, do_async_commit);
1202 ac->root = root;
1203 ac->newtrans = btrfs_join_transaction(root, 0);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00001204 if (IS_ERR(ac->newtrans)) {
1205 int err = PTR_ERR(ac->newtrans);
1206 kfree(ac);
1207 return err;
1208 }
Sage Weilbb9c12c2010-10-29 15:37:34 -04001209
1210 /* take transaction reference */
1211 mutex_lock(&root->fs_info->trans_mutex);
1212 cur_trans = trans->transaction;
Josef Bacik13c5a932011-04-11 15:45:29 -04001213 atomic_inc(&cur_trans->use_count);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001214 mutex_unlock(&root->fs_info->trans_mutex);
1215
1216 btrfs_end_transaction(trans, root);
1217 schedule_delayed_work(&ac->work, 0);
1218
1219 /* wait for transaction to start and unblock */
1220 mutex_lock(&root->fs_info->trans_mutex);
1221 if (wait_for_unblock)
1222 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1223 else
1224 wait_current_trans_commit_start(root, cur_trans);
1225 put_transaction(cur_trans);
1226 mutex_unlock(&root->fs_info->trans_mutex);
1227
1228 return 0;
1229}
1230
1231/*
1232 * btrfs_transaction state sequence:
1233 * in_commit = 0, blocked = 0 (initial)
1234 * in_commit = 1, blocked = 1
1235 * blocked = 0
1236 * commit_done = 1
1237 */
Chris Mason79154b12007-03-22 15:59:16 -04001238int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1239 struct btrfs_root *root)
1240{
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001241 unsigned long joined = 0;
Chris Mason79154b12007-03-22 15:59:16 -04001242 struct btrfs_transaction *cur_trans;
Chris Mason8fd17792007-04-19 21:01:03 -04001243 struct btrfs_transaction *prev_trans = NULL;
Chris Mason79154b12007-03-22 15:59:16 -04001244 DEFINE_WAIT(wait);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001245 int ret;
Chris Mason89573b92009-03-12 20:12:45 -04001246 int should_grow = 0;
1247 unsigned long now = get_seconds();
Sage Weildccae992009-04-02 16:59:01 -04001248 int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
Chris Mason79154b12007-03-22 15:59:16 -04001249
Chris Mason5a3f23d2009-03-31 13:27:11 -04001250 btrfs_run_ordered_operations(root, 0);
1251
Chris Mason56bec292009-03-13 10:10:06 -04001252 /* make a pass through all the delayed refs we have so far
1253 * any runnings procs may add more while we are here
1254 */
1255 ret = btrfs_run_delayed_refs(trans, root, 0);
1256 BUG_ON(ret);
1257
Yan, Zhenga22285a2010-05-16 10:48:46 -04001258 btrfs_trans_release_metadata(trans, root);
1259
Chris Masonb7ec40d2009-03-12 20:12:45 -04001260 cur_trans = trans->transaction;
Chris Mason56bec292009-03-13 10:10:06 -04001261 /*
1262 * set the flushing flag so procs in this transaction have to
1263 * start sending their work down.
1264 */
Chris Masonb7ec40d2009-03-12 20:12:45 -04001265 cur_trans->delayed_refs.flushing = 1;
Chris Mason56bec292009-03-13 10:10:06 -04001266
Chris Masonc3e69d52009-03-13 10:17:05 -04001267 ret = btrfs_run_delayed_refs(trans, root, 0);
Chris Mason56bec292009-03-13 10:10:06 -04001268 BUG_ON(ret);
1269
Chris Mason79154b12007-03-22 15:59:16 -04001270 mutex_lock(&root->fs_info->trans_mutex);
Chris Masonb7ec40d2009-03-12 20:12:45 -04001271 if (cur_trans->in_commit) {
Josef Bacik13c5a932011-04-11 15:45:29 -04001272 atomic_inc(&cur_trans->use_count);
Chris Masonccd467d2007-06-28 15:57:36 -04001273 mutex_unlock(&root->fs_info->trans_mutex);
Chris Mason79154b12007-03-22 15:59:16 -04001274 btrfs_end_transaction(trans, root);
Chris Masonccd467d2007-06-28 15:57:36 -04001275
Chris Mason79154b12007-03-22 15:59:16 -04001276 ret = wait_for_commit(root, cur_trans);
1277 BUG_ON(ret);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001278
1279 mutex_lock(&root->fs_info->trans_mutex);
Chris Mason79154b12007-03-22 15:59:16 -04001280 put_transaction(cur_trans);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001281 mutex_unlock(&root->fs_info->trans_mutex);
1282
Chris Mason79154b12007-03-22 15:59:16 -04001283 return 0;
1284 }
Chris Mason4313b392008-01-03 09:08:48 -05001285
Chris Mason2c90e5d2007-04-02 10:50:19 -04001286 trans->transaction->in_commit = 1;
Chris Masonf9295742008-07-17 12:54:14 -04001287 trans->transaction->blocked = 1;
Sage Weilbb9c12c2010-10-29 15:37:34 -04001288 wake_up(&root->fs_info->transaction_blocked_wait);
1289
Chris Masonccd467d2007-06-28 15:57:36 -04001290 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1291 prev_trans = list_entry(cur_trans->list.prev,
1292 struct btrfs_transaction, list);
1293 if (!prev_trans->commit_done) {
Josef Bacik13c5a932011-04-11 15:45:29 -04001294 atomic_inc(&prev_trans->use_count);
Chris Masonccd467d2007-06-28 15:57:36 -04001295 mutex_unlock(&root->fs_info->trans_mutex);
1296
1297 wait_for_commit(root, prev_trans);
Chris Masonccd467d2007-06-28 15:57:36 -04001298
Chris Masonccd467d2007-06-28 15:57:36 -04001299 mutex_lock(&root->fs_info->trans_mutex);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001300 put_transaction(prev_trans);
Chris Masonccd467d2007-06-28 15:57:36 -04001301 }
1302 }
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001303
Chris Mason89573b92009-03-12 20:12:45 -04001304 if (now < cur_trans->start_time || now - cur_trans->start_time < 1)
1305 should_grow = 1;
1306
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001307 do {
Yan Zheng7ea394f2008-08-05 13:05:02 -04001308 int snap_pending = 0;
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001309 joined = cur_trans->num_joined;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001310 if (!list_empty(&trans->transaction->pending_snapshots))
1311 snap_pending = 1;
1312
Chris Mason2c90e5d2007-04-02 10:50:19 -04001313 WARN_ON(cur_trans != trans->transaction);
Chris Mason79154b12007-03-22 15:59:16 -04001314 mutex_unlock(&root->fs_info->trans_mutex);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001315
Sage Weil0bdb1db2010-02-19 14:13:50 -08001316 if (flush_on_commit || snap_pending) {
Yan, Zheng24bbcf02009-11-12 09:36:34 +00001317 btrfs_start_delalloc_inodes(root, 1);
1318 ret = btrfs_wait_ordered_extents(root, 0, 1);
Sage Weilebecd3d2009-07-24 13:17:44 -04001319 BUG_ON(ret);
Yan Zheng7ea394f2008-08-05 13:05:02 -04001320 }
1321
Miao Xie16cdcec2011-04-22 18:12:22 +08001322 ret = btrfs_run_delayed_items(trans, root);
1323 BUG_ON(ret);
1324
Chris Mason5a3f23d2009-03-31 13:27:11 -04001325 /*
1326 * rename don't use btrfs_join_transaction, so, once we
1327 * set the transaction to blocked above, we aren't going
1328 * to get any new ordered operations. We can safely run
1329 * it here and no for sure that nothing new will be added
1330 * to the list
1331 */
1332 btrfs_run_ordered_operations(root, 1);
1333
Chris Masoned3b3d3142010-05-25 10:12:41 -04001334 prepare_to_wait(&cur_trans->writer_wait, &wait,
1335 TASK_UNINTERRUPTIBLE);
1336
Chris Mason89573b92009-03-12 20:12:45 -04001337 smp_mb();
Josef Bacik13c5a932011-04-11 15:45:29 -04001338 if (atomic_read(&cur_trans->num_writers) > 1)
Sage Weil99d16cb2010-10-29 15:37:34 -04001339 schedule_timeout(MAX_SCHEDULE_TIMEOUT);
1340 else if (should_grow)
1341 schedule_timeout(1);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001342
Chris Mason79154b12007-03-22 15:59:16 -04001343 mutex_lock(&root->fs_info->trans_mutex);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001344 finish_wait(&cur_trans->writer_wait, &wait);
Josef Bacik13c5a932011-04-11 15:45:29 -04001345 } while (atomic_read(&cur_trans->num_writers) > 1 ||
Chris Mason89573b92009-03-12 20:12:45 -04001346 (should_grow && cur_trans->num_joined != joined));
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001347
Chris Mason3063d292008-01-08 15:46:30 -05001348 ret = create_pending_snapshots(trans, root->fs_info);
1349 BUG_ON(ret);
1350
Miao Xie16cdcec2011-04-22 18:12:22 +08001351 ret = btrfs_run_delayed_items(trans, root);
1352 BUG_ON(ret);
1353
Chris Mason56bec292009-03-13 10:10:06 -04001354 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1355 BUG_ON(ret);
1356
Chris Mason2c90e5d2007-04-02 10:50:19 -04001357 WARN_ON(cur_trans != trans->transaction);
Chris Masondc17ff82008-01-08 15:46:30 -05001358
Chris Masone02119d2008-09-05 16:13:11 -04001359 /* btrfs_commit_tree_roots is responsible for getting the
1360 * various roots consistent with each other. Every pointer
1361 * in the tree of tree roots has to point to the most up to date
1362 * root for every subvolume and other tree. So, we have to keep
1363 * the tree logging code from jumping in and changing any
1364 * of the trees.
1365 *
1366 * At this point in the commit, there can't be any tree-log
1367 * writers, but a little lower down we drop the trans mutex
1368 * and let new people in. By holding the tree_log_mutex
1369 * from now until after the super is written, we avoid races
1370 * with the tree-log code.
1371 */
1372 mutex_lock(&root->fs_info->tree_log_mutex);
Zheng Yan1a40e232008-09-26 10:09:34 -04001373
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001374 ret = commit_fs_roots(trans, root);
Chris Mason54aa1f42007-06-22 14:16:25 -04001375 BUG_ON(ret);
1376
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001377 /* commit_fs_roots gets rid of all the tree log roots, it is now
Chris Masone02119d2008-09-05 16:13:11 -04001378 * safe to free the root of tree log roots
1379 */
1380 btrfs_free_log_root_tree(trans, root->fs_info);
1381
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001382 ret = commit_cowonly_roots(trans, root);
Chris Mason79154b12007-03-22 15:59:16 -04001383 BUG_ON(ret);
Chris Mason54aa1f42007-06-22 14:16:25 -04001384
Yan Zheng11833d62009-09-11 16:11:19 -04001385 btrfs_prepare_extent_commit(trans, root);
1386
Chris Mason78fae272007-03-25 11:35:08 -04001387 cur_trans = root->fs_info->running_transaction;
Chris Masoncee36a02008-01-15 08:40:48 -05001388 spin_lock(&root->fs_info->new_trans_lock);
Chris Mason78fae272007-03-25 11:35:08 -04001389 root->fs_info->running_transaction = NULL;
Chris Masoncee36a02008-01-15 08:40:48 -05001390 spin_unlock(&root->fs_info->new_trans_lock);
Chris Mason5f39d392007-10-15 16:14:19 -04001391
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001392 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1393 root->fs_info->tree_root->node);
Josef Bacik817d52f2009-07-13 21:29:25 -04001394 switch_commit_root(root->fs_info->tree_root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001395
1396 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1397 root->fs_info->chunk_root->node);
Josef Bacik817d52f2009-07-13 21:29:25 -04001398 switch_commit_root(root->fs_info->chunk_root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001399
1400 update_super_roots(root);
Chris Masone02119d2008-09-05 16:13:11 -04001401
1402 if (!root->fs_info->log_root_recovering) {
1403 btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
1404 btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0);
1405 }
1406
Chris Masona061fc82008-05-07 11:43:44 -04001407 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
1408 sizeof(root->fs_info->super_copy));
Chris Masonccd467d2007-06-28 15:57:36 -04001409
Chris Masonf9295742008-07-17 12:54:14 -04001410 trans->transaction->blocked = 0;
Chris Masonb7ec40d2009-03-12 20:12:45 -04001411
Chris Masonf9295742008-07-17 12:54:14 -04001412 wake_up(&root->fs_info->transaction_wait);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001413
Chris Mason78fae272007-03-25 11:35:08 -04001414 mutex_unlock(&root->fs_info->trans_mutex);
Chris Mason79154b12007-03-22 15:59:16 -04001415 ret = btrfs_write_and_wait_transaction(trans, root);
1416 BUG_ON(ret);
Yan Zhenga512bbf2008-12-08 16:46:26 -05001417 write_ctree_super(trans, root, 0);
Chris Mason4313b392008-01-03 09:08:48 -05001418
Chris Masone02119d2008-09-05 16:13:11 -04001419 /*
1420 * the super is written, we can safely allow the tree-loggers
1421 * to go about their business
1422 */
1423 mutex_unlock(&root->fs_info->tree_log_mutex);
1424
Yan Zheng11833d62009-09-11 16:11:19 -04001425 btrfs_finish_extent_commit(trans, root);
Chris Mason4313b392008-01-03 09:08:48 -05001426
Zheng Yan1a40e232008-09-26 10:09:34 -04001427 mutex_lock(&root->fs_info->trans_mutex);
1428
Chris Mason2c90e5d2007-04-02 10:50:19 -04001429 cur_trans->commit_done = 1;
Chris Masonb7ec40d2009-03-12 20:12:45 -04001430
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001431 root->fs_info->last_trans_committed = cur_trans->transid;
Josef Bacik817d52f2009-07-13 21:29:25 -04001432
Chris Mason2c90e5d2007-04-02 10:50:19 -04001433 wake_up(&cur_trans->commit_wait);
Chris Mason3de45862008-11-17 21:02:50 -05001434
Josef Bacik13c5a932011-04-11 15:45:29 -04001435 list_del_init(&cur_trans->list);
Chris Mason79154b12007-03-22 15:59:16 -04001436 put_transaction(cur_trans);
Chris Mason78fae272007-03-25 11:35:08 -04001437 put_transaction(cur_trans);
Josef Bacik58176a92007-08-29 15:47:34 -04001438
liubo1abe9b82011-03-24 11:18:59 +00001439 trace_btrfs_transaction_commit(root);
1440
Chris Mason78fae272007-03-25 11:35:08 -04001441 mutex_unlock(&root->fs_info->trans_mutex);
Chris Mason3de45862008-11-17 21:02:50 -05001442
Josef Bacik9ed74f22009-09-11 16:12:44 -04001443 if (current->journal_info == trans)
1444 current->journal_info = NULL;
1445
Chris Mason2c90e5d2007-04-02 10:50:19 -04001446 kmem_cache_free(btrfs_trans_handle_cachep, trans);
Yan, Zheng24bbcf02009-11-12 09:36:34 +00001447
1448 if (current != root->fs_info->transaction_kthread)
1449 btrfs_run_delayed_iputs(root);
1450
Chris Mason79154b12007-03-22 15:59:16 -04001451 return ret;
1452}
1453
Chris Masond352ac62008-09-29 15:18:18 -04001454/*
1455 * interface function to delete all the snapshots we have scheduled for deletion
1456 */
Chris Masone9d0b132007-08-10 14:06:19 -04001457int btrfs_clean_old_snapshots(struct btrfs_root *root)
1458{
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001459 LIST_HEAD(list);
1460 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone9d0b132007-08-10 14:06:19 -04001461
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001462 mutex_lock(&fs_info->trans_mutex);
1463 list_splice_init(&fs_info->dead_roots, &list);
1464 mutex_unlock(&fs_info->trans_mutex);
1465
1466 while (!list_empty(&list)) {
1467 root = list_entry(list.next, struct btrfs_root, root_list);
Yan, Zheng76dda932009-09-21 16:00:26 -04001468 list_del(&root->root_list);
1469
Miao Xie16cdcec2011-04-22 18:12:22 +08001470 btrfs_kill_all_delayed_nodes(root);
1471
Yan, Zheng76dda932009-09-21 16:00:26 -04001472 if (btrfs_header_backref_rev(root->node) <
1473 BTRFS_MIXED_BACKREF_REV)
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001474 btrfs_drop_snapshot(root, NULL, 0);
Yan, Zheng76dda932009-09-21 16:00:26 -04001475 else
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001476 btrfs_drop_snapshot(root, NULL, 1);
Chris Masone9d0b132007-08-10 14:06:19 -04001477 }
1478 return 0;
1479}