2 * Copyright (C) 2007 Oracle. All rights reserved.
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.
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.
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.
20 #include <linux/sched.h>
23 #include "transaction.h"
25 static int total_trans = 0;
26 extern struct kmem_cache *btrfs_trans_handle_cachep;
27 extern struct kmem_cache *btrfs_transaction_cachep;
29 static struct workqueue_struct *trans_wq;
31 #define BTRFS_ROOT_TRANS_TAG 0
33 static void put_transaction(struct btrfs_transaction *transaction)
35 WARN_ON(transaction->use_count == 0);
36 transaction->use_count--;
37 if (transaction->use_count == 0) {
38 WARN_ON(total_trans == 0);
40 list_del_init(&transaction->list);
41 memset(transaction, 0, sizeof(*transaction));
42 kmem_cache_free(btrfs_transaction_cachep, transaction);
46 static int join_transaction(struct btrfs_root *root)
48 struct btrfs_transaction *cur_trans;
49 cur_trans = root->fs_info->running_transaction;
51 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
55 root->fs_info->generation++;
56 root->fs_info->running_transaction = cur_trans;
57 cur_trans->num_writers = 0;
58 cur_trans->transid = root->fs_info->generation;
59 init_waitqueue_head(&cur_trans->writer_wait);
60 init_waitqueue_head(&cur_trans->commit_wait);
61 cur_trans->in_commit = 0;
62 cur_trans->use_count = 1;
63 cur_trans->commit_done = 0;
64 cur_trans->start_time = get_seconds();
65 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
66 init_bit_radix(&cur_trans->dirty_pages);
68 cur_trans->num_writers++;
72 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
75 struct btrfs_trans_handle *h =
76 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
80 mutex_lock(&root->fs_info->trans_mutex);
81 ret = join_transaction(root);
83 running_trans_id = root->fs_info->running_transaction->transid;
85 if (root != root->fs_info->tree_root && root->last_trans <
87 WARN_ON(root == root->fs_info->extent_root);
88 WARN_ON(root->ref_cows != 1);
89 if (root->root_item.refs != 0) {
90 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
91 (unsigned long)root->root_key.objectid,
92 BTRFS_ROOT_TRANS_TAG);
93 root->commit_root = root->node;
99 root->last_trans = running_trans_id;
100 h->transid = running_trans_id;
101 h->transaction = root->fs_info->running_transaction;
102 h->blocks_reserved = num_blocks;
104 h->block_group = NULL;
105 root->fs_info->running_transaction->use_count++;
106 mutex_unlock(&root->fs_info->trans_mutex);
110 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
111 struct btrfs_root *root)
113 struct btrfs_transaction *cur_trans;
115 mutex_lock(&root->fs_info->trans_mutex);
116 cur_trans = root->fs_info->running_transaction;
117 WARN_ON(cur_trans != trans->transaction);
118 WARN_ON(cur_trans->num_writers < 1);
119 cur_trans->num_writers--;
120 if (waitqueue_active(&cur_trans->writer_wait))
121 wake_up(&cur_trans->writer_wait);
122 put_transaction(cur_trans);
123 mutex_unlock(&root->fs_info->trans_mutex);
124 memset(trans, 0, sizeof(*trans));
125 kmem_cache_free(btrfs_trans_handle_cachep, trans);
130 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
131 struct btrfs_root *root)
133 unsigned long gang[16];
139 struct radix_tree_root *dirty_pages;
140 struct inode *btree_inode = root->fs_info->btree_inode;
142 if (!trans || !trans->transaction) {
143 return filemap_write_and_wait(btree_inode->i_mapping);
145 dirty_pages = &trans->transaction->dirty_pages;
147 ret = find_first_radix_bit(dirty_pages, gang,
148 0, ARRAY_SIZE(gang));
151 for (i = 0; i < ret; i++) {
153 clear_radix_bit(dirty_pages, gang[i]);
154 page = find_lock_page(btree_inode->i_mapping,
158 err = write_one_page(page, 0);
161 page_cache_release(page);
164 err = filemap_fdatawait(btree_inode->i_mapping);
170 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
171 struct btrfs_root *root)
174 u64 old_extent_block;
175 struct btrfs_fs_info *fs_info = root->fs_info;
176 struct btrfs_root *tree_root = fs_info->tree_root;
177 struct btrfs_root *extent_root = fs_info->extent_root;
179 btrfs_write_dirty_block_groups(trans, extent_root);
181 old_extent_block = btrfs_root_blocknr(&extent_root->root_item);
182 if (old_extent_block == bh_blocknr(extent_root->node))
184 btrfs_set_root_blocknr(&extent_root->root_item,
185 bh_blocknr(extent_root->node));
186 ret = btrfs_update_root(trans, tree_root,
187 &extent_root->root_key,
188 &extent_root->root_item);
190 btrfs_write_dirty_block_groups(trans, extent_root);
195 static int wait_for_commit(struct btrfs_root *root,
196 struct btrfs_transaction *commit)
199 mutex_lock(&root->fs_info->trans_mutex);
200 while(!commit->commit_done) {
201 prepare_to_wait(&commit->commit_wait, &wait,
202 TASK_UNINTERRUPTIBLE);
203 if (commit->commit_done)
205 mutex_unlock(&root->fs_info->trans_mutex);
207 mutex_lock(&root->fs_info->trans_mutex);
209 mutex_unlock(&root->fs_info->trans_mutex);
210 finish_wait(&commit->commit_wait, &wait);
215 struct list_head list;
216 struct btrfs_key snap_key;
217 struct buffer_head *commit_root;
218 struct btrfs_root *root;
222 int btrfs_add_dead_root(struct btrfs_root *root, struct list_head *dead_list)
224 struct dirty_root *dirty;
226 dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
229 memcpy(&dirty->snap_key, &root->root_key, sizeof(root->root_key));
230 dirty->commit_root = root->node;
232 dirty->free_on_drop = 1;
233 list_add(&dirty->list, dead_list);
237 static int add_dirty_roots(struct btrfs_trans_handle *trans,
238 struct radix_tree_root *radix,
239 struct list_head *list)
241 struct dirty_root *dirty;
242 struct btrfs_root *gang[8];
243 struct btrfs_root *root;
244 struct btrfs_root_item tmp_item;
251 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
253 BTRFS_ROOT_TRANS_TAG);
256 for (i = 0; i < ret; i++) {
258 radix_tree_tag_clear(radix,
259 (unsigned long)root->root_key.objectid,
260 BTRFS_ROOT_TRANS_TAG);
261 if (root->commit_root == root->node) {
262 WARN_ON(bh_blocknr(root->node) !=
263 btrfs_root_blocknr(&root->root_item));
264 brelse(root->commit_root);
265 root->commit_root = NULL;
268 dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
270 memcpy(&dirty->snap_key, &root->root_key,
271 sizeof(root->root_key));
272 dirty->commit_root = root->commit_root;
273 root->commit_root = NULL;
275 dirty->free_on_drop = 0;
276 memcpy(&tmp_item, &root->root_item, sizeof(tmp_item));
278 root->root_key.offset = root->fs_info->generation;
279 btrfs_set_root_blocknr(&root->root_item,
280 bh_blocknr(root->node));
281 err = btrfs_insert_root(trans, root->fs_info->tree_root,
286 refs = btrfs_root_refs(&tmp_item);
287 btrfs_set_root_refs(&tmp_item, refs - 1);
288 err = btrfs_update_root(trans, root->fs_info->tree_root,
294 list_add(&dirty->list, list);
302 static int drop_dirty_roots(struct btrfs_root *tree_root,
303 struct list_head *list)
305 struct dirty_root *dirty;
306 struct btrfs_trans_handle *trans;
308 while(!list_empty(list)) {
309 mutex_lock(&tree_root->fs_info->fs_mutex);
310 dirty = list_entry(list->next, struct dirty_root, list);
311 list_del_init(&dirty->list);
313 trans = btrfs_start_transaction(tree_root, 1);
314 ret = btrfs_drop_snapshot(trans, dirty->root,
317 ret = btrfs_del_root(trans, tree_root, &dirty->snap_key);
320 ret = btrfs_end_transaction(trans, tree_root);
323 if (dirty->free_on_drop)
326 mutex_unlock(&tree_root->fs_info->fs_mutex);
327 btrfs_btree_balance_dirty(tree_root);
332 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
333 struct btrfs_root *root)
336 struct btrfs_transaction *cur_trans;
337 struct btrfs_transaction *prev_trans = NULL;
338 struct list_head dirty_fs_roots;
339 struct radix_tree_root pinned_copy;
342 init_bit_radix(&pinned_copy);
343 INIT_LIST_HEAD(&dirty_fs_roots);
345 mutex_lock(&root->fs_info->trans_mutex);
346 if (trans->transaction->in_commit) {
347 cur_trans = trans->transaction;
348 trans->transaction->use_count++;
349 mutex_unlock(&root->fs_info->trans_mutex);
350 btrfs_end_transaction(trans, root);
352 mutex_unlock(&root->fs_info->fs_mutex);
353 ret = wait_for_commit(root, cur_trans);
355 put_transaction(cur_trans);
356 mutex_lock(&root->fs_info->fs_mutex);
359 trans->transaction->in_commit = 1;
360 cur_trans = trans->transaction;
361 if (cur_trans->list.prev != &root->fs_info->trans_list) {
362 prev_trans = list_entry(cur_trans->list.prev,
363 struct btrfs_transaction, list);
364 if (!prev_trans->commit_done) {
365 prev_trans->use_count++;
366 mutex_unlock(&root->fs_info->fs_mutex);
367 mutex_unlock(&root->fs_info->trans_mutex);
369 wait_for_commit(root, prev_trans);
370 put_transaction(prev_trans);
372 mutex_lock(&root->fs_info->fs_mutex);
373 mutex_lock(&root->fs_info->trans_mutex);
376 while (trans->transaction->num_writers > 1) {
377 WARN_ON(cur_trans != trans->transaction);
378 prepare_to_wait(&trans->transaction->writer_wait, &wait,
379 TASK_UNINTERRUPTIBLE);
380 if (trans->transaction->num_writers <= 1)
382 mutex_unlock(&root->fs_info->fs_mutex);
383 mutex_unlock(&root->fs_info->trans_mutex);
385 mutex_lock(&root->fs_info->fs_mutex);
386 mutex_lock(&root->fs_info->trans_mutex);
387 finish_wait(&trans->transaction->writer_wait, &wait);
389 finish_wait(&trans->transaction->writer_wait, &wait);
390 WARN_ON(cur_trans != trans->transaction);
391 ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
395 ret = btrfs_commit_tree_roots(trans, root);
398 cur_trans = root->fs_info->running_transaction;
399 root->fs_info->running_transaction = NULL;
400 btrfs_set_super_generation(&root->fs_info->super_copy,
402 btrfs_set_super_root(&root->fs_info->super_copy,
403 bh_blocknr(root->fs_info->tree_root->node));
404 memcpy(root->fs_info->disk_super, &root->fs_info->super_copy,
405 sizeof(root->fs_info->super_copy));
407 btrfs_copy_pinned(root, &pinned_copy);
409 mutex_unlock(&root->fs_info->trans_mutex);
410 mutex_unlock(&root->fs_info->fs_mutex);
411 ret = btrfs_write_and_wait_transaction(trans, root);
413 write_ctree_super(trans, root);
414 mutex_lock(&root->fs_info->fs_mutex);
415 btrfs_finish_extent_commit(trans, root, &pinned_copy);
416 mutex_lock(&root->fs_info->trans_mutex);
417 cur_trans->commit_done = 1;
418 wake_up(&cur_trans->commit_wait);
419 put_transaction(cur_trans);
420 put_transaction(cur_trans);
421 if (root->fs_info->closing)
422 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
424 list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
425 mutex_unlock(&root->fs_info->trans_mutex);
426 kmem_cache_free(btrfs_trans_handle_cachep, trans);
428 if (root->fs_info->closing) {
429 mutex_unlock(&root->fs_info->fs_mutex);
430 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
431 mutex_lock(&root->fs_info->fs_mutex);
436 void btrfs_transaction_cleaner(struct work_struct *work)
438 struct btrfs_fs_info *fs_info = container_of(work,
439 struct btrfs_fs_info,
442 struct btrfs_root *root = fs_info->tree_root;
443 struct btrfs_transaction *cur;
444 struct btrfs_trans_handle *trans;
445 struct list_head dirty_roots;
447 unsigned long delay = HZ * 30;
450 INIT_LIST_HEAD(&dirty_roots);
451 mutex_lock(&root->fs_info->fs_mutex);
452 mutex_lock(&root->fs_info->trans_mutex);
453 cur = root->fs_info->running_transaction;
455 mutex_unlock(&root->fs_info->trans_mutex);
459 if (now < cur->start_time || now - cur->start_time < 30) {
460 mutex_unlock(&root->fs_info->trans_mutex);
464 mutex_unlock(&root->fs_info->trans_mutex);
465 trans = btrfs_start_transaction(root, 1);
466 ret = btrfs_commit_transaction(trans, root);
468 mutex_unlock(&root->fs_info->fs_mutex);
470 mutex_lock(&root->fs_info->trans_mutex);
471 list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
472 mutex_unlock(&root->fs_info->trans_mutex);
474 if (!list_empty(&dirty_roots)) {
475 drop_dirty_roots(root, &dirty_roots);
477 btrfs_transaction_queue_work(root, delay);
480 void btrfs_transaction_queue_work(struct btrfs_root *root, int delay)
482 queue_delayed_work(trans_wq, &root->fs_info->trans_work, delay);
485 void btrfs_transaction_flush_work(struct btrfs_root *root)
487 cancel_rearming_delayed_workqueue(trans_wq, &root->fs_info->trans_work);
488 flush_workqueue(trans_wq);
491 void __init btrfs_init_transaction_sys(void)
493 trans_wq = create_workqueue("btrfs");
496 void __exit btrfs_exit_transaction_sys(void)
498 destroy_workqueue(trans_wq);