blob: 3d73c8d93bbb1c3f20c1d10dfa076e8dd887bd16 [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{
Chris Mason2c90e5d2007-04-02 10:50:19 -040035 WARN_ON(transaction->use_count == 0);
Chris Mason79154b12007-03-22 15:59:16 -040036 transaction->use_count--;
Chris Mason78fae272007-03-25 11:35:08 -040037 if (transaction->use_count == 0) {
Chris Mason8fd17792007-04-19 21:01:03 -040038 list_del_init(&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 */
Chris Mason80b67942008-02-01 16:35:04 -050053static noinline int join_transaction(struct btrfs_root *root)
Chris Mason79154b12007-03-22 15:59:16 -040054{
55 struct btrfs_transaction *cur_trans;
56 cur_trans = root->fs_info->running_transaction;
57 if (!cur_trans) {
Chris Mason2c90e5d2007-04-02 10:50:19 -040058 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
59 GFP_NOFS);
Chris Mason79154b12007-03-22 15:59:16 -040060 BUG_ON(!cur_trans);
Chris Mason0f7d52f2007-04-09 10:42:37 -040061 root->fs_info->generation++;
Josef Bacik15ee9bc2007-08-10 16:22:09 -040062 cur_trans->num_writers = 1;
63 cur_trans->num_joined = 0;
Chris Mason0f7d52f2007-04-09 10:42:37 -040064 cur_trans->transid = root->fs_info->generation;
Chris Mason79154b12007-03-22 15:59:16 -040065 init_waitqueue_head(&cur_trans->writer_wait);
66 init_waitqueue_head(&cur_trans->commit_wait);
67 cur_trans->in_commit = 0;
Chris Masonf9295742008-07-17 12:54:14 -040068 cur_trans->blocked = 0;
Chris Masond5719762007-03-23 10:01:08 -040069 cur_trans->use_count = 1;
Chris Mason79154b12007-03-22 15:59:16 -040070 cur_trans->commit_done = 0;
Chris Mason08607c12007-06-08 15:33:54 -040071 cur_trans->start_time = get_seconds();
Chris Mason56bec292009-03-13 10:10:06 -040072
Eric Paris6bef4d32010-02-23 19:43:04 +000073 cur_trans->delayed_refs.root = RB_ROOT;
Chris Mason56bec292009-03-13 10:10:06 -040074 cur_trans->delayed_refs.num_entries = 0;
Chris Masonc3e69d52009-03-13 10:17:05 -040075 cur_trans->delayed_refs.num_heads_ready = 0;
76 cur_trans->delayed_refs.num_heads = 0;
Chris Mason56bec292009-03-13 10:10:06 -040077 cur_trans->delayed_refs.flushing = 0;
Chris Masonc3e69d52009-03-13 10:17:05 -040078 cur_trans->delayed_refs.run_delayed_start = 0;
Chris Mason56bec292009-03-13 10:10:06 -040079 spin_lock_init(&cur_trans->delayed_refs.lock);
80
Chris Mason3063d292008-01-08 15:46:30 -050081 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
Chris Mason8fd17792007-04-19 21:01:03 -040082 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
Chris Masond1310b22008-01-24 16:13:08 -050083 extent_io_tree_init(&cur_trans->dirty_pages,
Chris Mason5f39d392007-10-15 16:14:19 -040084 root->fs_info->btree_inode->i_mapping,
85 GFP_NOFS);
Chris Mason48ec2cf2008-06-09 09:35:50 -040086 spin_lock(&root->fs_info->new_trans_lock);
87 root->fs_info->running_transaction = cur_trans;
88 spin_unlock(&root->fs_info->new_trans_lock);
Josef Bacik15ee9bc2007-08-10 16:22:09 -040089 } else {
90 cur_trans->num_writers++;
91 cur_trans->num_joined++;
Chris Mason79154b12007-03-22 15:59:16 -040092 }
Josef Bacik15ee9bc2007-08-10 16:22:09 -040093
Chris Mason79154b12007-03-22 15:59:16 -040094 return 0;
95}
96
Chris Masond352ac62008-09-29 15:18:18 -040097/*
Chris Masond3977122009-01-05 21:25:51 -050098 * this does all the record keeping required to make sure that a reference
99 * counted root is properly recorded in a given transaction. This is required
100 * to make sure the old root from before we joined the transaction is deleted
101 * when the transaction commits
Chris Masond352ac62008-09-29 15:18:18 -0400102 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400103static noinline int record_root_in_trans(struct btrfs_trans_handle *trans,
104 struct btrfs_root *root)
Chris Mason6702ed42007-08-07 16:15:09 -0400105{
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400106 if (root->ref_cows && root->last_trans < trans->transid) {
Chris Mason6702ed42007-08-07 16:15:09 -0400107 WARN_ON(root == root->fs_info->extent_root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400108 WARN_ON(root->commit_root != root->node);
Yan Zheng31153d82008-07-28 15:32:19 -0400109
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400110 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
111 (unsigned long)root->root_key.objectid,
112 BTRFS_ROOT_TRANS_TAG);
113 root->last_trans = trans->transid;
114 btrfs_init_reloc_root(trans, root);
Chris Mason6702ed42007-08-07 16:15:09 -0400115 }
116 return 0;
117}
118
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400119int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
120 struct btrfs_root *root)
121{
122 if (!root->ref_cows)
123 return 0;
124
125 mutex_lock(&root->fs_info->trans_mutex);
126 if (root->last_trans == trans->transid) {
127 mutex_unlock(&root->fs_info->trans_mutex);
128 return 0;
129 }
130
131 record_root_in_trans(trans, root);
132 mutex_unlock(&root->fs_info->trans_mutex);
133 return 0;
134}
135
Chris Masond352ac62008-09-29 15:18:18 -0400136/* wait for commit against the current transaction to become unblocked
137 * when this is done, it is safe to start a new transaction, but the current
138 * transaction might not be fully on disk.
139 */
Chris Mason37d1aee2008-07-31 10:48:37 -0400140static void wait_current_trans(struct btrfs_root *root)
Chris Mason79154b12007-03-22 15:59:16 -0400141{
Chris Masonf9295742008-07-17 12:54:14 -0400142 struct btrfs_transaction *cur_trans;
Chris Mason79154b12007-03-22 15:59:16 -0400143
Chris Masonf9295742008-07-17 12:54:14 -0400144 cur_trans = root->fs_info->running_transaction;
Chris Mason37d1aee2008-07-31 10:48:37 -0400145 if (cur_trans && cur_trans->blocked) {
Chris Masonf9295742008-07-17 12:54:14 -0400146 DEFINE_WAIT(wait);
147 cur_trans->use_count++;
Chris Masond3977122009-01-05 21:25:51 -0500148 while (1) {
Chris Masonf9295742008-07-17 12:54:14 -0400149 prepare_to_wait(&root->fs_info->transaction_wait, &wait,
150 TASK_UNINTERRUPTIBLE);
Zhao Lei471fa172010-03-25 12:35:14 +0000151 if (!cur_trans->blocked)
Chris Masonf9295742008-07-17 12:54:14 -0400152 break;
Zhao Lei471fa172010-03-25 12:35:14 +0000153 mutex_unlock(&root->fs_info->trans_mutex);
154 schedule();
155 mutex_lock(&root->fs_info->trans_mutex);
Chris Masonf9295742008-07-17 12:54:14 -0400156 }
Zhao Lei471fa172010-03-25 12:35:14 +0000157 finish_wait(&root->fs_info->transaction_wait, &wait);
Chris Masonf9295742008-07-17 12:54:14 -0400158 put_transaction(cur_trans);
159 }
Chris Mason37d1aee2008-07-31 10:48:37 -0400160}
161
Josef Bacik249ac1e2009-11-10 21:23:48 -0500162enum btrfs_trans_type {
163 TRANS_START,
164 TRANS_JOIN,
165 TRANS_USERSPACE,
Josef Bacik0af3d002010-06-21 14:48:16 -0400166 TRANS_JOIN_NOLOCK,
Josef Bacik249ac1e2009-11-10 21:23:48 -0500167};
168
Yan, Zhenga22285a2010-05-16 10:48:46 -0400169static int may_wait_transaction(struct btrfs_root *root, int type)
Chris Mason37d1aee2008-07-31 10:48:37 -0400170{
Chris Mason4bef0842008-09-08 11:18:08 -0400171 if (!root->fs_info->log_root_recovering &&
Josef Bacik249ac1e2009-11-10 21:23:48 -0500172 ((type == TRANS_START && !root->fs_info->open_ioctl_trans) ||
173 type == TRANS_USERSPACE))
Yan, Zhenga22285a2010-05-16 10:48:46 -0400174 return 1;
175 return 0;
176}
177
178static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
179 u64 num_items, int type)
180{
181 struct btrfs_trans_handle *h;
182 struct btrfs_transaction *cur_trans;
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);
198 BUG_ON(ret);
Chris Mason0f7d52f2007-04-09 10:42:37 -0400199
Yan, Zhenga22285a2010-05-16 10:48:46 -0400200 cur_trans = root->fs_info->running_transaction;
201 cur_trans->use_count++;
Josef Bacik0af3d002010-06-21 14:48:16 -0400202 if (type != TRANS_JOIN_NOLOCK)
203 mutex_unlock(&root->fs_info->trans_mutex);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400204
205 h->transid = cur_trans->transid;
206 h->transaction = cur_trans;
Chris Mason79154b12007-03-22 15:59:16 -0400207 h->blocks_used = 0;
Yan Zhengd2fb3432008-12-11 16:30:39 -0500208 h->block_group = 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400209 h->bytes_reserved = 0;
Chris Mason56bec292009-03-13 10:10:06 -0400210 h->delayed_ref_updates = 0;
Yan, Zhengf0486c62010-05-16 10:46:25 -0400211 h->block_rsv = NULL;
Chris Masonb7ec40d2009-03-12 20:12:45 -0400212
Yan, Zhenga22285a2010-05-16 10:48:46 -0400213 smp_mb();
214 if (cur_trans->blocked && may_wait_transaction(root, type)) {
215 btrfs_commit_transaction(h, root);
216 goto again;
217 }
Josef Bacik9ed74f22009-09-11 16:12:44 -0400218
Yan, Zhenga22285a2010-05-16 10:48:46 -0400219 if (num_items > 0) {
Josef Bacik8bb8ab22010-10-15 16:52:49 -0400220 ret = btrfs_trans_reserve_metadata(h, root, num_items);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400221 if (ret == -EAGAIN) {
222 btrfs_commit_transaction(h, root);
223 goto again;
224 }
225 if (ret < 0) {
226 btrfs_end_transaction(h, root);
227 return ERR_PTR(ret);
228 }
229 }
230
Josef Bacik0af3d002010-06-21 14:48:16 -0400231 if (type != TRANS_JOIN_NOLOCK)
232 mutex_lock(&root->fs_info->trans_mutex);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400233 record_root_in_trans(h, root);
Josef Bacik0af3d002010-06-21 14:48:16 -0400234 if (type != TRANS_JOIN_NOLOCK)
235 mutex_unlock(&root->fs_info->trans_mutex);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400236
237 if (!current->journal_info && type != TRANS_USERSPACE)
238 current->journal_info = h;
Chris Mason79154b12007-03-22 15:59:16 -0400239 return h;
240}
241
Chris Masonf9295742008-07-17 12:54:14 -0400242struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
Yan, Zhenga22285a2010-05-16 10:48:46 -0400243 int num_items)
Chris Masonf9295742008-07-17 12:54:14 -0400244{
Yan, Zhenga22285a2010-05-16 10:48:46 -0400245 return start_transaction(root, num_items, TRANS_START);
Chris Masonf9295742008-07-17 12:54:14 -0400246}
247struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
248 int num_blocks)
249{
Yan, Zhenga22285a2010-05-16 10:48:46 -0400250 return start_transaction(root, 0, TRANS_JOIN);
Chris Masonf9295742008-07-17 12:54:14 -0400251}
252
Josef Bacik0af3d002010-06-21 14:48:16 -0400253struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root,
254 int num_blocks)
255{
256 return start_transaction(root, 0, TRANS_JOIN_NOLOCK);
257}
258
Sage Weil9ca9ee02008-08-04 10:41:27 -0400259struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
260 int num_blocks)
261{
Yan, Zhenga22285a2010-05-16 10:48:46 -0400262 return start_transaction(r, 0, TRANS_USERSPACE);
Sage Weil9ca9ee02008-08-04 10:41:27 -0400263}
264
Chris Masond352ac62008-09-29 15:18:18 -0400265/* wait for a transaction commit to be fully complete */
Chris Mason89ce8a62008-06-25 16:01:31 -0400266static noinline int wait_for_commit(struct btrfs_root *root,
267 struct btrfs_transaction *commit)
268{
269 DEFINE_WAIT(wait);
270 mutex_lock(&root->fs_info->trans_mutex);
Chris Masond3977122009-01-05 21:25:51 -0500271 while (!commit->commit_done) {
Chris Mason89ce8a62008-06-25 16:01:31 -0400272 prepare_to_wait(&commit->commit_wait, &wait,
273 TASK_UNINTERRUPTIBLE);
274 if (commit->commit_done)
275 break;
276 mutex_unlock(&root->fs_info->trans_mutex);
277 schedule();
278 mutex_lock(&root->fs_info->trans_mutex);
279 }
280 mutex_unlock(&root->fs_info->trans_mutex);
281 finish_wait(&commit->commit_wait, &wait);
282 return 0;
283}
284
Sage Weil46204592010-10-29 15:41:32 -0400285int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
286{
287 struct btrfs_transaction *cur_trans = NULL, *t;
288 int ret;
289
290 mutex_lock(&root->fs_info->trans_mutex);
291
292 ret = 0;
293 if (transid) {
294 if (transid <= root->fs_info->last_trans_committed)
295 goto out_unlock;
296
297 /* find specified transaction */
298 list_for_each_entry(t, &root->fs_info->trans_list, list) {
299 if (t->transid == transid) {
300 cur_trans = t;
301 break;
302 }
303 if (t->transid > transid)
304 break;
305 }
306 ret = -EINVAL;
307 if (!cur_trans)
308 goto out_unlock; /* bad transid */
309 } else {
310 /* find newest transaction that is committing | committed */
311 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
312 list) {
313 if (t->in_commit) {
314 if (t->commit_done)
315 goto out_unlock;
316 cur_trans = t;
317 break;
318 }
319 }
320 if (!cur_trans)
321 goto out_unlock; /* nothing committing|committed */
322 }
323
324 cur_trans->use_count++;
325 mutex_unlock(&root->fs_info->trans_mutex);
326
327 wait_for_commit(root, cur_trans);
328
329 mutex_lock(&root->fs_info->trans_mutex);
330 put_transaction(cur_trans);
331 ret = 0;
332out_unlock:
333 mutex_unlock(&root->fs_info->trans_mutex);
334 return ret;
335}
336
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400337#if 0
Chris Masond352ac62008-09-29 15:18:18 -0400338/*
Chris Masond3977122009-01-05 21:25:51 -0500339 * rate limit against the drop_snapshot code. This helps to slow down new
340 * operations if the drop_snapshot code isn't able to keep up.
Chris Masond352ac62008-09-29 15:18:18 -0400341 */
Chris Mason37d1aee2008-07-31 10:48:37 -0400342static void throttle_on_drops(struct btrfs_root *root)
Chris Masonab78c842008-07-29 16:15:18 -0400343{
344 struct btrfs_fs_info *info = root->fs_info;
Chris Mason2dd3e672008-08-04 08:20:15 -0400345 int harder_count = 0;
Chris Masonab78c842008-07-29 16:15:18 -0400346
Chris Mason2dd3e672008-08-04 08:20:15 -0400347harder:
Chris Masonab78c842008-07-29 16:15:18 -0400348 if (atomic_read(&info->throttles)) {
349 DEFINE_WAIT(wait);
350 int thr;
Chris Masonab78c842008-07-29 16:15:18 -0400351 thr = atomic_read(&info->throttle_gen);
352
353 do {
354 prepare_to_wait(&info->transaction_throttle,
355 &wait, TASK_UNINTERRUPTIBLE);
356 if (!atomic_read(&info->throttles)) {
357 finish_wait(&info->transaction_throttle, &wait);
358 break;
359 }
360 schedule();
361 finish_wait(&info->transaction_throttle, &wait);
362 } while (thr == atomic_read(&info->throttle_gen));
Chris Mason2dd3e672008-08-04 08:20:15 -0400363 harder_count++;
364
365 if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 &&
366 harder_count < 2)
367 goto harder;
368
369 if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 &&
370 harder_count < 10)
371 goto harder;
372
373 if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 &&
374 harder_count < 20)
375 goto harder;
Chris Masonab78c842008-07-29 16:15:18 -0400376 }
377}
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400378#endif
Chris Masonab78c842008-07-29 16:15:18 -0400379
Chris Mason37d1aee2008-07-31 10:48:37 -0400380void btrfs_throttle(struct btrfs_root *root)
381{
382 mutex_lock(&root->fs_info->trans_mutex);
Sage Weil9ca9ee02008-08-04 10:41:27 -0400383 if (!root->fs_info->open_ioctl_trans)
384 wait_current_trans(root);
Chris Mason37d1aee2008-07-31 10:48:37 -0400385 mutex_unlock(&root->fs_info->trans_mutex);
Chris Mason37d1aee2008-07-31 10:48:37 -0400386}
387
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400388static int should_end_transaction(struct btrfs_trans_handle *trans,
389 struct btrfs_root *root)
390{
391 int ret;
392 ret = btrfs_block_rsv_check(trans, root,
393 &root->fs_info->global_block_rsv, 0, 5);
394 return ret ? 1 : 0;
395}
396
397int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
398 struct btrfs_root *root)
399{
400 struct btrfs_transaction *cur_trans = trans->transaction;
401 int updates;
402
403 if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
404 return 1;
405
406 updates = trans->delayed_ref_updates;
407 trans->delayed_ref_updates = 0;
408 if (updates)
409 btrfs_run_delayed_refs(trans, root, updates);
410
411 return should_end_transaction(trans, root);
412}
413
Chris Mason89ce8a62008-06-25 16:01:31 -0400414static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
Josef Bacik0af3d002010-06-21 14:48:16 -0400415 struct btrfs_root *root, int throttle, int lock)
Chris Mason79154b12007-03-22 15:59:16 -0400416{
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400417 struct btrfs_transaction *cur_trans = trans->transaction;
Chris Masonab78c842008-07-29 16:15:18 -0400418 struct btrfs_fs_info *info = root->fs_info;
Chris Masonc3e69d52009-03-13 10:17:05 -0400419 int count = 0;
Chris Masond6e4a422007-04-06 15:37:36 -0400420
Chris Masonc3e69d52009-03-13 10:17:05 -0400421 while (count < 4) {
422 unsigned long cur = trans->delayed_ref_updates;
423 trans->delayed_ref_updates = 0;
424 if (cur &&
425 trans->transaction->delayed_refs.num_heads_ready > 64) {
426 trans->delayed_ref_updates = 0;
Chris Masonb7ec40d2009-03-12 20:12:45 -0400427
428 /*
429 * do a full flush if the transaction is trying
430 * to close
431 */
432 if (trans->transaction->delayed_refs.flushing)
433 cur = 0;
Chris Masonc3e69d52009-03-13 10:17:05 -0400434 btrfs_run_delayed_refs(trans, root, cur);
435 } else {
436 break;
437 }
438 count++;
Chris Mason56bec292009-03-13 10:10:06 -0400439 }
440
Yan, Zhenga22285a2010-05-16 10:48:46 -0400441 btrfs_trans_release_metadata(trans, root);
442
Josef Bacik0af3d002010-06-21 14:48:16 -0400443 if (lock && !root->fs_info->open_ioctl_trans &&
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400444 should_end_transaction(trans, root))
445 trans->transaction->blocked = 1;
446
Josef Bacik0af3d002010-06-21 14:48:16 -0400447 if (lock && cur_trans->blocked && !cur_trans->in_commit) {
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400448 if (throttle)
449 return btrfs_commit_transaction(trans, root);
450 else
451 wake_up_process(info->transaction_kthread);
452 }
453
Josef Bacik0af3d002010-06-21 14:48:16 -0400454 if (lock)
455 mutex_lock(&info->trans_mutex);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400456 WARN_ON(cur_trans != info->running_transaction);
Chris Masond5719762007-03-23 10:01:08 -0400457 WARN_ON(cur_trans->num_writers < 1);
Chris Masonccd467d2007-06-28 15:57:36 -0400458 cur_trans->num_writers--;
Chris Mason89ce8a62008-06-25 16:01:31 -0400459
Sage Weil99d16cb2010-10-29 15:37:34 -0400460 smp_mb();
Chris Mason79154b12007-03-22 15:59:16 -0400461 if (waitqueue_active(&cur_trans->writer_wait))
462 wake_up(&cur_trans->writer_wait);
Chris Mason79154b12007-03-22 15:59:16 -0400463 put_transaction(cur_trans);
Josef Bacik0af3d002010-06-21 14:48:16 -0400464 if (lock)
465 mutex_unlock(&info->trans_mutex);
Josef Bacik9ed74f22009-09-11 16:12:44 -0400466
467 if (current->journal_info == trans)
468 current->journal_info = NULL;
Chris Masond6025572007-03-30 14:27:56 -0400469 memset(trans, 0, sizeof(*trans));
Chris Mason2c90e5d2007-04-02 10:50:19 -0400470 kmem_cache_free(btrfs_trans_handle_cachep, trans);
Chris Masonab78c842008-07-29 16:15:18 -0400471
Yan, Zheng24bbcf02009-11-12 09:36:34 +0000472 if (throttle)
473 btrfs_run_delayed_iputs(root);
474
Chris Mason79154b12007-03-22 15:59:16 -0400475 return 0;
476}
477
Chris Mason89ce8a62008-06-25 16:01:31 -0400478int btrfs_end_transaction(struct btrfs_trans_handle *trans,
479 struct btrfs_root *root)
480{
Josef Bacik0af3d002010-06-21 14:48:16 -0400481 return __btrfs_end_transaction(trans, root, 0, 1);
Chris Mason89ce8a62008-06-25 16:01:31 -0400482}
483
484int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
485 struct btrfs_root *root)
486{
Josef Bacik0af3d002010-06-21 14:48:16 -0400487 return __btrfs_end_transaction(trans, root, 1, 1);
488}
489
490int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans,
491 struct btrfs_root *root)
492{
493 return __btrfs_end_transaction(trans, root, 0, 0);
Chris Mason89ce8a62008-06-25 16:01:31 -0400494}
495
Chris Masond352ac62008-09-29 15:18:18 -0400496/*
497 * when btree blocks are allocated, they have some corresponding bits set for
498 * them in one of two extent_io trees. This is used to make sure all of
Chris Mason690587d2009-10-13 13:29:19 -0400499 * those extents are sent to disk but does not wait on them
Chris Masond352ac62008-09-29 15:18:18 -0400500 */
Chris Mason690587d2009-10-13 13:29:19 -0400501int btrfs_write_marked_extents(struct btrfs_root *root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000502 struct extent_io_tree *dirty_pages, int mark)
Chris Mason79154b12007-03-22 15:59:16 -0400503{
Chris Mason7c4452b2007-04-28 09:29:35 -0400504 int ret;
Chris Mason777e6bd2008-08-15 15:34:15 -0400505 int err = 0;
Chris Mason7c4452b2007-04-28 09:29:35 -0400506 int werr = 0;
507 struct page *page;
Chris Mason7c4452b2007-04-28 09:29:35 -0400508 struct inode *btree_inode = root->fs_info->btree_inode;
Chris Mason777e6bd2008-08-15 15:34:15 -0400509 u64 start = 0;
Chris Mason5f39d392007-10-15 16:14:19 -0400510 u64 end;
511 unsigned long index;
Chris Mason7c4452b2007-04-28 09:29:35 -0400512
Chris Masond3977122009-01-05 21:25:51 -0500513 while (1) {
Chris Mason777e6bd2008-08-15 15:34:15 -0400514 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000515 mark);
Chris Mason5f39d392007-10-15 16:14:19 -0400516 if (ret)
Chris Mason7c4452b2007-04-28 09:29:35 -0400517 break;
Chris Masond3977122009-01-05 21:25:51 -0500518 while (start <= end) {
Chris Mason777e6bd2008-08-15 15:34:15 -0400519 cond_resched();
520
Chris Mason5f39d392007-10-15 16:14:19 -0400521 index = start >> PAGE_CACHE_SHIFT;
Chris Mason35ebb932007-10-30 16:56:53 -0400522 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
Chris Mason4bef0842008-09-08 11:18:08 -0400523 page = find_get_page(btree_inode->i_mapping, index);
Chris Mason7c4452b2007-04-28 09:29:35 -0400524 if (!page)
525 continue;
Chris Mason4bef0842008-09-08 11:18:08 -0400526
527 btree_lock_page_hook(page);
528 if (!page->mapping) {
529 unlock_page(page);
530 page_cache_release(page);
531 continue;
532 }
533
Chris Mason6702ed42007-08-07 16:15:09 -0400534 if (PageWriteback(page)) {
535 if (PageDirty(page))
536 wait_on_page_writeback(page);
537 else {
538 unlock_page(page);
539 page_cache_release(page);
540 continue;
541 }
542 }
Chris Mason7c4452b2007-04-28 09:29:35 -0400543 err = write_one_page(page, 0);
544 if (err)
545 werr = err;
546 page_cache_release(page);
547 }
548 }
Chris Mason690587d2009-10-13 13:29:19 -0400549 if (err)
550 werr = err;
551 return werr;
552}
553
554/*
555 * when btree blocks are allocated, they have some corresponding bits set for
556 * them in one of two extent_io trees. This is used to make sure all of
557 * those extents are on disk for transaction or log commit. We wait
558 * on all the pages and clear them from the dirty pages state tree
559 */
560int btrfs_wait_marked_extents(struct btrfs_root *root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000561 struct extent_io_tree *dirty_pages, int mark)
Chris Mason690587d2009-10-13 13:29:19 -0400562{
563 int ret;
564 int err = 0;
565 int werr = 0;
566 struct page *page;
567 struct inode *btree_inode = root->fs_info->btree_inode;
568 u64 start = 0;
569 u64 end;
570 unsigned long index;
571
Chris Masond3977122009-01-05 21:25:51 -0500572 while (1) {
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000573 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
574 mark);
Chris Mason777e6bd2008-08-15 15:34:15 -0400575 if (ret)
576 break;
577
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000578 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
Chris Masond3977122009-01-05 21:25:51 -0500579 while (start <= end) {
Chris Mason777e6bd2008-08-15 15:34:15 -0400580 index = start >> PAGE_CACHE_SHIFT;
581 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
582 page = find_get_page(btree_inode->i_mapping, index);
583 if (!page)
584 continue;
585 if (PageDirty(page)) {
Chris Mason4bef0842008-09-08 11:18:08 -0400586 btree_lock_page_hook(page);
587 wait_on_page_writeback(page);
Chris Mason777e6bd2008-08-15 15:34:15 -0400588 err = write_one_page(page, 0);
589 if (err)
590 werr = err;
591 }
Chris Mason105d9312008-11-18 12:13:12 -0500592 wait_on_page_writeback(page);
Chris Mason777e6bd2008-08-15 15:34:15 -0400593 page_cache_release(page);
594 cond_resched();
595 }
596 }
Chris Mason7c4452b2007-04-28 09:29:35 -0400597 if (err)
598 werr = err;
599 return werr;
Chris Mason79154b12007-03-22 15:59:16 -0400600}
601
Chris Mason690587d2009-10-13 13:29:19 -0400602/*
603 * when btree blocks are allocated, they have some corresponding bits set for
604 * them in one of two extent_io trees. This is used to make sure all of
605 * those extents are on disk for transaction or log commit
606 */
607int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000608 struct extent_io_tree *dirty_pages, int mark)
Chris Mason690587d2009-10-13 13:29:19 -0400609{
610 int ret;
611 int ret2;
612
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000613 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
614 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
Chris Mason690587d2009-10-13 13:29:19 -0400615 return ret || ret2;
616}
617
Chris Masond0c803c2008-09-11 16:17:57 -0400618int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
619 struct btrfs_root *root)
620{
621 if (!trans || !trans->transaction) {
622 struct inode *btree_inode;
623 btree_inode = root->fs_info->btree_inode;
624 return filemap_write_and_wait(btree_inode->i_mapping);
625 }
626 return btrfs_write_and_wait_marked_extents(root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000627 &trans->transaction->dirty_pages,
628 EXTENT_DIRTY);
Chris Masond0c803c2008-09-11 16:17:57 -0400629}
630
Chris Masond352ac62008-09-29 15:18:18 -0400631/*
632 * this is used to update the root pointer in the tree of tree roots.
633 *
634 * But, in the case of the extent allocation tree, updating the root
635 * pointer may allocate blocks which may change the root of the extent
636 * allocation tree.
637 *
638 * So, this loops and repeats and makes sure the cowonly root didn't
639 * change while the root pointer was being updated in the metadata.
640 */
Chris Mason0b86a832008-03-24 15:01:56 -0400641static int update_cowonly_root(struct btrfs_trans_handle *trans,
642 struct btrfs_root *root)
643{
644 int ret;
645 u64 old_root_bytenr;
Yan, Zheng86b9f2e2009-11-12 09:36:50 +0000646 u64 old_root_used;
Chris Mason0b86a832008-03-24 15:01:56 -0400647 struct btrfs_root *tree_root = root->fs_info->tree_root;
648
Yan, Zheng86b9f2e2009-11-12 09:36:50 +0000649 old_root_used = btrfs_root_used(&root->root_item);
Chris Mason0b86a832008-03-24 15:01:56 -0400650 btrfs_write_dirty_block_groups(trans, root);
Chris Mason56bec292009-03-13 10:10:06 -0400651
Chris Masond3977122009-01-05 21:25:51 -0500652 while (1) {
Chris Mason0b86a832008-03-24 15:01:56 -0400653 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
Yan, Zheng86b9f2e2009-11-12 09:36:50 +0000654 if (old_root_bytenr == root->node->start &&
655 old_root_used == btrfs_root_used(&root->root_item))
Chris Mason0b86a832008-03-24 15:01:56 -0400656 break;
Chris Mason87ef2bb2008-10-30 11:23:27 -0400657
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400658 btrfs_set_root_node(&root->root_item, root->node);
Chris Mason0b86a832008-03-24 15:01:56 -0400659 ret = btrfs_update_root(trans, tree_root,
660 &root->root_key,
661 &root->root_item);
662 BUG_ON(ret);
Chris Mason56bec292009-03-13 10:10:06 -0400663
Yan, Zheng86b9f2e2009-11-12 09:36:50 +0000664 old_root_used = btrfs_root_used(&root->root_item);
Yan Zheng4a8c9a62009-07-22 10:07:05 -0400665 ret = btrfs_write_dirty_block_groups(trans, root);
Chris Mason56bec292009-03-13 10:10:06 -0400666 BUG_ON(ret);
Chris Mason0b86a832008-03-24 15:01:56 -0400667 }
Yan Zheng276e6802009-07-30 09:40:40 -0400668
669 if (root != root->fs_info->extent_root)
670 switch_commit_root(root);
671
Chris Mason0b86a832008-03-24 15:01:56 -0400672 return 0;
673}
674
Chris Masond352ac62008-09-29 15:18:18 -0400675/*
676 * update all the cowonly tree roots on disk
677 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400678static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
679 struct btrfs_root *root)
Chris Mason79154b12007-03-22 15:59:16 -0400680{
Chris Mason79154b12007-03-22 15:59:16 -0400681 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Mason0b86a832008-03-24 15:01:56 -0400682 struct list_head *next;
Yan Zheng84234f32008-10-29 14:49:05 -0400683 struct extent_buffer *eb;
Chris Mason56bec292009-03-13 10:10:06 -0400684 int ret;
Yan Zheng84234f32008-10-29 14:49:05 -0400685
Chris Mason56bec292009-03-13 10:10:06 -0400686 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
687 BUG_ON(ret);
Chris Mason87ef2bb2008-10-30 11:23:27 -0400688
Yan Zheng84234f32008-10-29 14:49:05 -0400689 eb = btrfs_lock_root_node(fs_info->tree_root);
Chris Mason9fa8cfe2009-03-13 10:24:59 -0400690 btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb);
Yan Zheng84234f32008-10-29 14:49:05 -0400691 btrfs_tree_unlock(eb);
692 free_extent_buffer(eb);
Chris Mason79154b12007-03-22 15:59:16 -0400693
Chris Mason56bec292009-03-13 10:10:06 -0400694 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
695 BUG_ON(ret);
Chris Mason87ef2bb2008-10-30 11:23:27 -0400696
Chris Masond3977122009-01-05 21:25:51 -0500697 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
Chris Mason0b86a832008-03-24 15:01:56 -0400698 next = fs_info->dirty_cowonly_roots.next;
699 list_del_init(next);
700 root = list_entry(next, struct btrfs_root, dirty_list);
Chris Mason87ef2bb2008-10-30 11:23:27 -0400701
Chris Mason0b86a832008-03-24 15:01:56 -0400702 update_cowonly_root(trans, root);
Chris Mason79154b12007-03-22 15:59:16 -0400703 }
Yan Zheng276e6802009-07-30 09:40:40 -0400704
705 down_write(&fs_info->extent_commit_sem);
706 switch_commit_root(fs_info->extent_root);
707 up_write(&fs_info->extent_commit_sem);
708
Chris Mason79154b12007-03-22 15:59:16 -0400709 return 0;
710}
711
Chris Masond352ac62008-09-29 15:18:18 -0400712/*
713 * dead roots are old snapshots that need to be deleted. This allocates
714 * a dirty root struct and adds it into the list of dead roots that need to
715 * be deleted
716 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400717int btrfs_add_dead_root(struct btrfs_root *root)
Chris Mason5eda7b52007-06-22 14:16:25 -0400718{
Yan Zhengb48652c2008-08-04 23:23:47 -0400719 mutex_lock(&root->fs_info->trans_mutex);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400720 list_add(&root->root_list, &root->fs_info->dead_roots);
Yan Zhengb48652c2008-08-04 23:23:47 -0400721 mutex_unlock(&root->fs_info->trans_mutex);
Chris Mason5eda7b52007-06-22 14:16:25 -0400722 return 0;
723}
724
Chris Masond352ac62008-09-29 15:18:18 -0400725/*
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400726 * update all the cowonly tree roots on disk
Chris Masond352ac62008-09-29 15:18:18 -0400727 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400728static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
729 struct btrfs_root *root)
Chris Mason0f7d52f2007-04-09 10:42:37 -0400730{
Chris Mason0f7d52f2007-04-09 10:42:37 -0400731 struct btrfs_root *gang[8];
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400732 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Mason0f7d52f2007-04-09 10:42:37 -0400733 int i;
734 int ret;
Chris Mason54aa1f42007-06-22 14:16:25 -0400735 int err = 0;
736
Chris Masond3977122009-01-05 21:25:51 -0500737 while (1) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400738 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
739 (void **)gang, 0,
Chris Mason0f7d52f2007-04-09 10:42:37 -0400740 ARRAY_SIZE(gang),
741 BTRFS_ROOT_TRANS_TAG);
742 if (ret == 0)
743 break;
744 for (i = 0; i < ret; i++) {
745 root = gang[i];
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400746 radix_tree_tag_clear(&fs_info->fs_roots_radix,
747 (unsigned long)root->root_key.objectid,
748 BTRFS_ROOT_TRANS_TAG);
Yan Zheng31153d82008-07-28 15:32:19 -0400749
Chris Masone02119d2008-09-05 16:13:11 -0400750 btrfs_free_log(trans, root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400751 btrfs_update_reloc_root(trans, root);
Yan, Zhengd68fc572010-05-16 10:49:58 -0400752 btrfs_orphan_commit_root(trans, root);
Chris Masone02119d2008-09-05 16:13:11 -0400753
Yan Zheng978d9102009-06-15 20:01:02 -0400754 if (root->commit_root != root->node) {
Josef Bacik817d52f2009-07-13 21:29:25 -0400755 switch_commit_root(root);
Yan Zheng978d9102009-06-15 20:01:02 -0400756 btrfs_set_root_node(&root->root_item,
757 root->node);
758 }
Chris Mason9f3a7422007-08-07 15:52:19 -0400759
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400760 err = btrfs_update_root(trans, fs_info->tree_root,
Chris Mason0f7d52f2007-04-09 10:42:37 -0400761 &root->root_key,
762 &root->root_item);
Chris Mason54aa1f42007-06-22 14:16:25 -0400763 if (err)
764 break;
Chris Mason0f7d52f2007-04-09 10:42:37 -0400765 }
766 }
Chris Mason54aa1f42007-06-22 14:16:25 -0400767 return err;
Chris Mason0f7d52f2007-04-09 10:42:37 -0400768}
769
Chris Masond352ac62008-09-29 15:18:18 -0400770/*
771 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
772 * otherwise every leaf in the btree is read and defragged.
773 */
Chris Masone9d0b132007-08-10 14:06:19 -0400774int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
775{
776 struct btrfs_fs_info *info = root->fs_info;
Chris Masone9d0b132007-08-10 14:06:19 -0400777 struct btrfs_trans_handle *trans;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400778 int ret;
Chris Masond3c2fdc2007-09-17 10:58:06 -0400779 unsigned long nr;
Chris Masone9d0b132007-08-10 14:06:19 -0400780
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400781 if (xchg(&root->defrag_running, 1))
Chris Masone9d0b132007-08-10 14:06:19 -0400782 return 0;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400783
Chris Mason6b800532007-10-15 16:17:34 -0400784 while (1) {
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400785 trans = btrfs_start_transaction(root, 0);
786 if (IS_ERR(trans))
787 return PTR_ERR(trans);
788
Chris Masone9d0b132007-08-10 14:06:19 -0400789 ret = btrfs_defrag_leaves(trans, root, cacheonly);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400790
Chris Masond3c2fdc2007-09-17 10:58:06 -0400791 nr = trans->blocks_used;
Chris Masone9d0b132007-08-10 14:06:19 -0400792 btrfs_end_transaction(trans, root);
Chris Masond3c2fdc2007-09-17 10:58:06 -0400793 btrfs_btree_balance_dirty(info->tree_root, nr);
Chris Masone9d0b132007-08-10 14:06:19 -0400794 cond_resched();
795
Chris Mason3f157a22008-06-25 16:01:31 -0400796 if (root->fs_info->closing || ret != -EAGAIN)
Chris Masone9d0b132007-08-10 14:06:19 -0400797 break;
798 }
799 root->defrag_running = 0;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400800 return ret;
Chris Masone9d0b132007-08-10 14:06:19 -0400801}
802
Yan Zheng2c47e6052009-06-27 21:07:35 -0400803#if 0
Chris Masond352ac62008-09-29 15:18:18 -0400804/*
Chris Masonb7ec40d2009-03-12 20:12:45 -0400805 * when dropping snapshots, we generate a ton of delayed refs, and it makes
806 * sense not to join the transaction while it is trying to flush the current
807 * queue of delayed refs out.
808 *
809 * This is used by the drop snapshot code only
810 */
811static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info)
812{
813 DEFINE_WAIT(wait);
814
815 mutex_lock(&info->trans_mutex);
816 while (info->running_transaction &&
817 info->running_transaction->delayed_refs.flushing) {
818 prepare_to_wait(&info->transaction_wait, &wait,
819 TASK_UNINTERRUPTIBLE);
820 mutex_unlock(&info->trans_mutex);
Chris Mason59bc5c72009-04-24 14:39:25 -0400821
Chris Masonb7ec40d2009-03-12 20:12:45 -0400822 schedule();
Chris Mason59bc5c72009-04-24 14:39:25 -0400823
Chris Masonb7ec40d2009-03-12 20:12:45 -0400824 mutex_lock(&info->trans_mutex);
825 finish_wait(&info->transaction_wait, &wait);
826 }
827 mutex_unlock(&info->trans_mutex);
828 return 0;
829}
830
831/*
Chris Masond352ac62008-09-29 15:18:18 -0400832 * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
833 * all of them
834 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400835int btrfs_drop_dead_root(struct btrfs_root *root)
Chris Mason0f7d52f2007-04-09 10:42:37 -0400836{
Chris Mason0f7d52f2007-04-09 10:42:37 -0400837 struct btrfs_trans_handle *trans;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400838 struct btrfs_root *tree_root = root->fs_info->tree_root;
Chris Masond3c2fdc2007-09-17 10:58:06 -0400839 unsigned long nr;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400840 int ret;
Chris Mason9f3a7422007-08-07 15:52:19 -0400841
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400842 while (1) {
843 /*
844 * we don't want to jump in and create a bunch of
845 * delayed refs if the transaction is starting to close
846 */
847 wait_transaction_pre_flush(tree_root->fs_info);
848 trans = btrfs_start_transaction(tree_root, 1);
Josef Bacik58176a92007-08-29 15:47:34 -0400849
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400850 /*
851 * we've joined a transaction, make sure it isn't
852 * closing right now
853 */
854 if (trans->transaction->delayed_refs.flushing) {
855 btrfs_end_transaction(trans, tree_root);
856 continue;
Josef Bacik58176a92007-08-29 15:47:34 -0400857 }
Chris Masona2135012008-06-25 16:01:30 -0400858
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400859 ret = btrfs_drop_snapshot(trans, root);
860 if (ret != -EAGAIN)
Chris Mason54aa1f42007-06-22 14:16:25 -0400861 break;
Chris Masona2135012008-06-25 16:01:30 -0400862
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400863 ret = btrfs_update_root(trans, tree_root,
864 &root->root_key,
865 &root->root_item);
866 if (ret)
867 break;
Yanbcc63ab2008-07-30 16:29:20 -0400868
Chris Masond3c2fdc2007-09-17 10:58:06 -0400869 nr = trans->blocks_used;
Chris Mason0f7d52f2007-04-09 10:42:37 -0400870 ret = btrfs_end_transaction(trans, tree_root);
871 BUG_ON(ret);
Chris Mason5eda7b52007-06-22 14:16:25 -0400872
Chris Masond3c2fdc2007-09-17 10:58:06 -0400873 btrfs_btree_balance_dirty(tree_root, nr);
Chris Mason4dc11902007-10-15 16:18:14 -0400874 cond_resched();
Chris Mason0f7d52f2007-04-09 10:42:37 -0400875 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400876 BUG_ON(ret);
877
878 ret = btrfs_del_root(trans, tree_root, &root->root_key);
879 BUG_ON(ret);
880
881 nr = trans->blocks_used;
882 ret = btrfs_end_transaction(trans, tree_root);
883 BUG_ON(ret);
884
885 free_extent_buffer(root->node);
886 free_extent_buffer(root->commit_root);
887 kfree(root);
888
889 btrfs_btree_balance_dirty(tree_root, nr);
Chris Mason54aa1f42007-06-22 14:16:25 -0400890 return ret;
Chris Mason0f7d52f2007-04-09 10:42:37 -0400891}
Yan Zheng2c47e6052009-06-27 21:07:35 -0400892#endif
Chris Mason0f7d52f2007-04-09 10:42:37 -0400893
Chris Masond352ac62008-09-29 15:18:18 -0400894/*
895 * new snapshots need to be created at a very specific time in the
896 * transaction commit. This does the actual creation
897 */
Chris Mason80b67942008-02-01 16:35:04 -0500898static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
Chris Mason3063d292008-01-08 15:46:30 -0500899 struct btrfs_fs_info *fs_info,
900 struct btrfs_pending_snapshot *pending)
901{
902 struct btrfs_key key;
Chris Mason80b67942008-02-01 16:35:04 -0500903 struct btrfs_root_item *new_root_item;
Chris Mason3063d292008-01-08 15:46:30 -0500904 struct btrfs_root *tree_root = fs_info->tree_root;
905 struct btrfs_root *root = pending->root;
Sage Weil6bdb72d2010-03-15 17:27:13 +0000906 struct btrfs_root *parent_root;
907 struct inode *parent_inode;
Josef Bacik6a912212010-11-20 09:48:00 +0000908 struct dentry *parent;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400909 struct dentry *dentry;
Chris Mason3063d292008-01-08 15:46:30 -0500910 struct extent_buffer *tmp;
Chris Mason925baed2008-06-25 16:01:30 -0400911 struct extent_buffer *old;
Chris Mason3063d292008-01-08 15:46:30 -0500912 int ret;
Yan, Zhengd68fc572010-05-16 10:49:58 -0400913 u64 to_reserve = 0;
Sage Weil6bdb72d2010-03-15 17:27:13 +0000914 u64 index = 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400915 u64 objectid;
Li Zefanb83cc962010-12-20 16:04:08 +0800916 u64 root_flags;
Chris Mason3063d292008-01-08 15:46:30 -0500917
Chris Mason80b67942008-02-01 16:35:04 -0500918 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
919 if (!new_root_item) {
Yan, Zhenga22285a2010-05-16 10:48:46 -0400920 pending->error = -ENOMEM;
Chris Mason80b67942008-02-01 16:35:04 -0500921 goto fail;
922 }
Yan, Zhenga22285a2010-05-16 10:48:46 -0400923
Chris Mason3063d292008-01-08 15:46:30 -0500924 ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400925 if (ret) {
926 pending->error = ret;
Chris Mason3063d292008-01-08 15:46:30 -0500927 goto fail;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400928 }
Chris Mason3063d292008-01-08 15:46:30 -0500929
Yan, Zheng3fd0a552010-05-16 10:49:59 -0400930 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
Yan, Zhengd68fc572010-05-16 10:49:58 -0400931 btrfs_orphan_pre_snapshot(trans, pending, &to_reserve);
932
933 if (to_reserve > 0) {
934 ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv,
Josef Bacik8bb8ab22010-10-15 16:52:49 -0400935 to_reserve);
Yan, Zhengd68fc572010-05-16 10:49:58 -0400936 if (ret) {
937 pending->error = ret;
938 goto fail;
939 }
940 }
941
Chris Mason3063d292008-01-08 15:46:30 -0500942 key.objectid = objectid;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400943 key.offset = (u64)-1;
944 key.type = BTRFS_ROOT_ITEM_KEY;
Chris Mason3063d292008-01-08 15:46:30 -0500945
Yan, Zhenga22285a2010-05-16 10:48:46 -0400946 trans->block_rsv = &pending->block_rsv;
Sage Weil6bdb72d2010-03-15 17:27:13 +0000947
Yan, Zhenga22285a2010-05-16 10:48:46 -0400948 dentry = pending->dentry;
Josef Bacik6a912212010-11-20 09:48:00 +0000949 parent = dget_parent(dentry);
950 parent_inode = parent->d_inode;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400951 parent_root = BTRFS_I(parent_inode)->root;
Sage Weil6bdb72d2010-03-15 17:27:13 +0000952 record_root_in_trans(trans, parent_root);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400953
Sage Weil6bdb72d2010-03-15 17:27:13 +0000954 /*
955 * insert the directory item
956 */
Sage Weil6bdb72d2010-03-15 17:27:13 +0000957 ret = btrfs_set_inode_index(parent_inode, &index);
958 BUG_ON(ret);
959 ret = btrfs_insert_dir_item(trans, parent_root,
Yan, Zhenga22285a2010-05-16 10:48:46 -0400960 dentry->d_name.name, dentry->d_name.len,
961 parent_inode->i_ino, &key,
962 BTRFS_FT_DIR, index);
Sage Weil6bdb72d2010-03-15 17:27:13 +0000963 BUG_ON(ret);
964
Yan, Zhenga22285a2010-05-16 10:48:46 -0400965 btrfs_i_size_write(parent_inode, parent_inode->i_size +
966 dentry->d_name.len * 2);
Sage Weil6bdb72d2010-03-15 17:27:13 +0000967 ret = btrfs_update_inode(trans, parent_root, parent_inode);
968 BUG_ON(ret);
969
970 record_root_in_trans(trans, root);
971 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
972 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
973
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,
1002 parent_inode->i_ino, index,
1003 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;
1059 spin_lock(&info->new_trans_lock);
1060 if (info->running_transaction)
1061 ret = info->running_transaction->in_commit;
1062 spin_unlock(&info->new_trans_lock);
1063 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;
1069 spin_lock(&info->new_trans_lock);
1070 if (info->running_transaction)
1071 ret = info->running_transaction->blocked;
1072 spin_unlock(&info->new_trans_lock);
1073 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 }
1096 mutex_unlock(&root->fs_info->trans_mutex);
1097 schedule();
1098 mutex_lock(&root->fs_info->trans_mutex);
1099 finish_wait(&root->fs_info->transaction_blocked_wait, &wait);
1100 }
1101}
1102
1103/*
1104 * wait for the current transaction to start and then become unblocked.
1105 * caller holds ref.
1106 */
1107static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1108 struct btrfs_transaction *trans)
1109{
1110 DEFINE_WAIT(wait);
1111
1112 if (trans->commit_done || (trans->in_commit && !trans->blocked))
1113 return;
1114
1115 while (1) {
1116 prepare_to_wait(&root->fs_info->transaction_wait, &wait,
1117 TASK_UNINTERRUPTIBLE);
1118 if (trans->commit_done ||
1119 (trans->in_commit && !trans->blocked)) {
1120 finish_wait(&root->fs_info->transaction_wait,
1121 &wait);
1122 break;
1123 }
1124 mutex_unlock(&root->fs_info->trans_mutex);
1125 schedule();
1126 mutex_lock(&root->fs_info->trans_mutex);
1127 finish_wait(&root->fs_info->transaction_wait,
1128 &wait);
1129 }
1130}
1131
1132/*
1133 * commit transactions asynchronously. once btrfs_commit_transaction_async
1134 * returns, any subsequent transaction will not be allowed to join.
1135 */
1136struct btrfs_async_commit {
1137 struct btrfs_trans_handle *newtrans;
1138 struct btrfs_root *root;
1139 struct delayed_work work;
1140};
1141
1142static void do_async_commit(struct work_struct *work)
1143{
1144 struct btrfs_async_commit *ac =
1145 container_of(work, struct btrfs_async_commit, work.work);
1146
1147 btrfs_commit_transaction(ac->newtrans, ac->root);
1148 kfree(ac);
1149}
1150
1151int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1152 struct btrfs_root *root,
1153 int wait_for_unblock)
1154{
1155 struct btrfs_async_commit *ac;
1156 struct btrfs_transaction *cur_trans;
1157
1158 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1159 BUG_ON(!ac);
1160
1161 INIT_DELAYED_WORK(&ac->work, do_async_commit);
1162 ac->root = root;
1163 ac->newtrans = btrfs_join_transaction(root, 0);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00001164 if (IS_ERR(ac->newtrans)) {
1165 int err = PTR_ERR(ac->newtrans);
1166 kfree(ac);
1167 return err;
1168 }
Sage Weilbb9c12c2010-10-29 15:37:34 -04001169
1170 /* take transaction reference */
1171 mutex_lock(&root->fs_info->trans_mutex);
1172 cur_trans = trans->transaction;
1173 cur_trans->use_count++;
1174 mutex_unlock(&root->fs_info->trans_mutex);
1175
1176 btrfs_end_transaction(trans, root);
1177 schedule_delayed_work(&ac->work, 0);
1178
1179 /* wait for transaction to start and unblock */
1180 mutex_lock(&root->fs_info->trans_mutex);
1181 if (wait_for_unblock)
1182 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1183 else
1184 wait_current_trans_commit_start(root, cur_trans);
1185 put_transaction(cur_trans);
1186 mutex_unlock(&root->fs_info->trans_mutex);
1187
1188 return 0;
1189}
1190
1191/*
1192 * btrfs_transaction state sequence:
1193 * in_commit = 0, blocked = 0 (initial)
1194 * in_commit = 1, blocked = 1
1195 * blocked = 0
1196 * commit_done = 1
1197 */
Chris Mason79154b12007-03-22 15:59:16 -04001198int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1199 struct btrfs_root *root)
1200{
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001201 unsigned long joined = 0;
Chris Mason79154b12007-03-22 15:59:16 -04001202 struct btrfs_transaction *cur_trans;
Chris Mason8fd17792007-04-19 21:01:03 -04001203 struct btrfs_transaction *prev_trans = NULL;
Chris Mason79154b12007-03-22 15:59:16 -04001204 DEFINE_WAIT(wait);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001205 int ret;
Chris Mason89573b92009-03-12 20:12:45 -04001206 int should_grow = 0;
1207 unsigned long now = get_seconds();
Sage Weildccae992009-04-02 16:59:01 -04001208 int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
Chris Mason79154b12007-03-22 15:59:16 -04001209
Chris Mason5a3f23d2009-03-31 13:27:11 -04001210 btrfs_run_ordered_operations(root, 0);
1211
Chris Mason56bec292009-03-13 10:10:06 -04001212 /* make a pass through all the delayed refs we have so far
1213 * any runnings procs may add more while we are here
1214 */
1215 ret = btrfs_run_delayed_refs(trans, root, 0);
1216 BUG_ON(ret);
1217
Yan, Zhenga22285a2010-05-16 10:48:46 -04001218 btrfs_trans_release_metadata(trans, root);
1219
Chris Masonb7ec40d2009-03-12 20:12:45 -04001220 cur_trans = trans->transaction;
Chris Mason56bec292009-03-13 10:10:06 -04001221 /*
1222 * set the flushing flag so procs in this transaction have to
1223 * start sending their work down.
1224 */
Chris Masonb7ec40d2009-03-12 20:12:45 -04001225 cur_trans->delayed_refs.flushing = 1;
Chris Mason56bec292009-03-13 10:10:06 -04001226
Chris Masonc3e69d52009-03-13 10:17:05 -04001227 ret = btrfs_run_delayed_refs(trans, root, 0);
Chris Mason56bec292009-03-13 10:10:06 -04001228 BUG_ON(ret);
1229
Chris Mason79154b12007-03-22 15:59:16 -04001230 mutex_lock(&root->fs_info->trans_mutex);
Chris Masonb7ec40d2009-03-12 20:12:45 -04001231 if (cur_trans->in_commit) {
1232 cur_trans->use_count++;
Chris Masonccd467d2007-06-28 15:57:36 -04001233 mutex_unlock(&root->fs_info->trans_mutex);
Chris Mason79154b12007-03-22 15:59:16 -04001234 btrfs_end_transaction(trans, root);
Chris Masonccd467d2007-06-28 15:57:36 -04001235
Chris Mason79154b12007-03-22 15:59:16 -04001236 ret = wait_for_commit(root, cur_trans);
1237 BUG_ON(ret);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001238
1239 mutex_lock(&root->fs_info->trans_mutex);
Chris Mason79154b12007-03-22 15:59:16 -04001240 put_transaction(cur_trans);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001241 mutex_unlock(&root->fs_info->trans_mutex);
1242
Chris Mason79154b12007-03-22 15:59:16 -04001243 return 0;
1244 }
Chris Mason4313b392008-01-03 09:08:48 -05001245
Chris Mason2c90e5d2007-04-02 10:50:19 -04001246 trans->transaction->in_commit = 1;
Chris Masonf9295742008-07-17 12:54:14 -04001247 trans->transaction->blocked = 1;
Sage Weilbb9c12c2010-10-29 15:37:34 -04001248 wake_up(&root->fs_info->transaction_blocked_wait);
1249
Chris Masonccd467d2007-06-28 15:57:36 -04001250 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1251 prev_trans = list_entry(cur_trans->list.prev,
1252 struct btrfs_transaction, list);
1253 if (!prev_trans->commit_done) {
1254 prev_trans->use_count++;
Chris Masonccd467d2007-06-28 15:57:36 -04001255 mutex_unlock(&root->fs_info->trans_mutex);
1256
1257 wait_for_commit(root, prev_trans);
Chris Masonccd467d2007-06-28 15:57:36 -04001258
Chris Masonccd467d2007-06-28 15:57:36 -04001259 mutex_lock(&root->fs_info->trans_mutex);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001260 put_transaction(prev_trans);
Chris Masonccd467d2007-06-28 15:57:36 -04001261 }
1262 }
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001263
Chris Mason89573b92009-03-12 20:12:45 -04001264 if (now < cur_trans->start_time || now - cur_trans->start_time < 1)
1265 should_grow = 1;
1266
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001267 do {
Yan Zheng7ea394f2008-08-05 13:05:02 -04001268 int snap_pending = 0;
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);
Chris Mason79154b12007-03-22 15:59:16 -04001274 mutex_unlock(&root->fs_info->trans_mutex);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001275
Sage Weil0bdb1db2010-02-19 14:13:50 -08001276 if (flush_on_commit || snap_pending) {
Yan, Zheng24bbcf02009-11-12 09:36:34 +00001277 btrfs_start_delalloc_inodes(root, 1);
1278 ret = btrfs_wait_ordered_extents(root, 0, 1);
Sage Weilebecd3d2009-07-24 13:17:44 -04001279 BUG_ON(ret);
Yan Zheng7ea394f2008-08-05 13:05:02 -04001280 }
1281
Chris Mason5a3f23d2009-03-31 13:27:11 -04001282 /*
1283 * rename don't use btrfs_join_transaction, so, once we
1284 * set the transaction to blocked above, we aren't going
1285 * to get any new ordered operations. We can safely run
1286 * it here and no for sure that nothing new will be added
1287 * to the list
1288 */
1289 btrfs_run_ordered_operations(root, 1);
1290
Chris Masoned3b3d3142010-05-25 10:12:41 -04001291 prepare_to_wait(&cur_trans->writer_wait, &wait,
1292 TASK_UNINTERRUPTIBLE);
1293
Chris Mason89573b92009-03-12 20:12:45 -04001294 smp_mb();
Sage Weil99d16cb2010-10-29 15:37:34 -04001295 if (cur_trans->num_writers > 1)
1296 schedule_timeout(MAX_SCHEDULE_TIMEOUT);
1297 else if (should_grow)
1298 schedule_timeout(1);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001299
Chris Mason79154b12007-03-22 15:59:16 -04001300 mutex_lock(&root->fs_info->trans_mutex);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001301 finish_wait(&cur_trans->writer_wait, &wait);
1302 } while (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 Mason3063d292008-01-08 15:46:30 -05001305 ret = create_pending_snapshots(trans, root->fs_info);
1306 BUG_ON(ret);
1307
Chris Mason56bec292009-03-13 10:10:06 -04001308 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1309 BUG_ON(ret);
1310
Chris Mason2c90e5d2007-04-02 10:50:19 -04001311 WARN_ON(cur_trans != trans->transaction);
Chris Masondc17ff82008-01-08 15:46:30 -05001312
Chris Masone02119d2008-09-05 16:13:11 -04001313 /* btrfs_commit_tree_roots is responsible for getting the
1314 * various roots consistent with each other. Every pointer
1315 * in the tree of tree roots has to point to the most up to date
1316 * root for every subvolume and other tree. So, we have to keep
1317 * the tree logging code from jumping in and changing any
1318 * of the trees.
1319 *
1320 * At this point in the commit, there can't be any tree-log
1321 * writers, but a little lower down we drop the trans mutex
1322 * and let new people in. By holding the tree_log_mutex
1323 * from now until after the super is written, we avoid races
1324 * with the tree-log code.
1325 */
1326 mutex_lock(&root->fs_info->tree_log_mutex);
Zheng Yan1a40e232008-09-26 10:09:34 -04001327
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001328 ret = commit_fs_roots(trans, root);
Chris Mason54aa1f42007-06-22 14:16:25 -04001329 BUG_ON(ret);
1330
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001331 /* commit_fs_roots gets rid of all the tree log roots, it is now
Chris Masone02119d2008-09-05 16:13:11 -04001332 * safe to free the root of tree log roots
1333 */
1334 btrfs_free_log_root_tree(trans, root->fs_info);
1335
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001336 ret = commit_cowonly_roots(trans, root);
Chris Mason79154b12007-03-22 15:59:16 -04001337 BUG_ON(ret);
Chris Mason54aa1f42007-06-22 14:16:25 -04001338
Yan Zheng11833d62009-09-11 16:11:19 -04001339 btrfs_prepare_extent_commit(trans, root);
1340
Chris Mason78fae272007-03-25 11:35:08 -04001341 cur_trans = root->fs_info->running_transaction;
Chris Masoncee36a02008-01-15 08:40:48 -05001342 spin_lock(&root->fs_info->new_trans_lock);
Chris Mason78fae272007-03-25 11:35:08 -04001343 root->fs_info->running_transaction = NULL;
Chris Masoncee36a02008-01-15 08:40:48 -05001344 spin_unlock(&root->fs_info->new_trans_lock);
Chris Mason5f39d392007-10-15 16:14:19 -04001345
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001346 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1347 root->fs_info->tree_root->node);
Josef Bacik817d52f2009-07-13 21:29:25 -04001348 switch_commit_root(root->fs_info->tree_root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001349
1350 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1351 root->fs_info->chunk_root->node);
Josef Bacik817d52f2009-07-13 21:29:25 -04001352 switch_commit_root(root->fs_info->chunk_root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001353
1354 update_super_roots(root);
Chris Masone02119d2008-09-05 16:13:11 -04001355
1356 if (!root->fs_info->log_root_recovering) {
1357 btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
1358 btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0);
1359 }
1360
Chris Masona061fc82008-05-07 11:43:44 -04001361 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
1362 sizeof(root->fs_info->super_copy));
Chris Masonccd467d2007-06-28 15:57:36 -04001363
Chris Masonf9295742008-07-17 12:54:14 -04001364 trans->transaction->blocked = 0;
Chris Masonb7ec40d2009-03-12 20:12:45 -04001365
Chris Masonf9295742008-07-17 12:54:14 -04001366 wake_up(&root->fs_info->transaction_wait);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001367
Chris Mason78fae272007-03-25 11:35:08 -04001368 mutex_unlock(&root->fs_info->trans_mutex);
Chris Mason79154b12007-03-22 15:59:16 -04001369 ret = btrfs_write_and_wait_transaction(trans, root);
1370 BUG_ON(ret);
Yan Zhenga512bbf2008-12-08 16:46:26 -05001371 write_ctree_super(trans, root, 0);
Chris Mason4313b392008-01-03 09:08:48 -05001372
Chris Masone02119d2008-09-05 16:13:11 -04001373 /*
1374 * the super is written, we can safely allow the tree-loggers
1375 * to go about their business
1376 */
1377 mutex_unlock(&root->fs_info->tree_log_mutex);
1378
Yan Zheng11833d62009-09-11 16:11:19 -04001379 btrfs_finish_extent_commit(trans, root);
Chris Mason4313b392008-01-03 09:08:48 -05001380
Zheng Yan1a40e232008-09-26 10:09:34 -04001381 mutex_lock(&root->fs_info->trans_mutex);
1382
Chris Mason2c90e5d2007-04-02 10:50:19 -04001383 cur_trans->commit_done = 1;
Chris Masonb7ec40d2009-03-12 20:12:45 -04001384
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001385 root->fs_info->last_trans_committed = cur_trans->transid;
Josef Bacik817d52f2009-07-13 21:29:25 -04001386
Chris Mason2c90e5d2007-04-02 10:50:19 -04001387 wake_up(&cur_trans->commit_wait);
Chris Mason3de45862008-11-17 21:02:50 -05001388
Chris Mason79154b12007-03-22 15:59:16 -04001389 put_transaction(cur_trans);
Chris Mason78fae272007-03-25 11:35:08 -04001390 put_transaction(cur_trans);
Josef Bacik58176a92007-08-29 15:47:34 -04001391
Chris Mason78fae272007-03-25 11:35:08 -04001392 mutex_unlock(&root->fs_info->trans_mutex);
Chris Mason3de45862008-11-17 21:02:50 -05001393
Josef Bacik9ed74f22009-09-11 16:12:44 -04001394 if (current->journal_info == trans)
1395 current->journal_info = NULL;
1396
Chris Mason2c90e5d2007-04-02 10:50:19 -04001397 kmem_cache_free(btrfs_trans_handle_cachep, trans);
Yan, Zheng24bbcf02009-11-12 09:36:34 +00001398
1399 if (current != root->fs_info->transaction_kthread)
1400 btrfs_run_delayed_iputs(root);
1401
Chris Mason79154b12007-03-22 15:59:16 -04001402 return ret;
1403}
1404
Chris Masond352ac62008-09-29 15:18:18 -04001405/*
1406 * interface function to delete all the snapshots we have scheduled for deletion
1407 */
Chris Masone9d0b132007-08-10 14:06:19 -04001408int btrfs_clean_old_snapshots(struct btrfs_root *root)
1409{
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001410 LIST_HEAD(list);
1411 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone9d0b132007-08-10 14:06:19 -04001412
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001413 mutex_lock(&fs_info->trans_mutex);
1414 list_splice_init(&fs_info->dead_roots, &list);
1415 mutex_unlock(&fs_info->trans_mutex);
1416
1417 while (!list_empty(&list)) {
1418 root = list_entry(list.next, struct btrfs_root, root_list);
Yan, Zheng76dda932009-09-21 16:00:26 -04001419 list_del(&root->root_list);
1420
1421 if (btrfs_header_backref_rev(root->node) <
1422 BTRFS_MIXED_BACKREF_REV)
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001423 btrfs_drop_snapshot(root, NULL, 0);
Yan, Zheng76dda932009-09-21 16:00:26 -04001424 else
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001425 btrfs_drop_snapshot(root, NULL, 1);
Chris Masone9d0b132007-08-10 14:06:19 -04001426 }
1427 return 0;
1428}