Btrfs: Run igrab on data=ordered inodes to prevent deadlocks during writeout
[linux-2.6] / fs / btrfs / transaction.c
1 /*
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
19 #include <linux/fs.h>
20 #include <linux/sched.h>
21 #include <linux/writeback.h>
22 #include <linux/pagemap.h>
23 #include "ctree.h"
24 #include "disk-io.h"
25 #include "transaction.h"
26
27 static int total_trans = 0;
28 extern struct kmem_cache *btrfs_trans_handle_cachep;
29 extern struct kmem_cache *btrfs_transaction_cachep;
30
31 static struct workqueue_struct *trans_wq;
32
33 #define BTRFS_ROOT_TRANS_TAG 0
34 #define BTRFS_ROOT_DEFRAG_TAG 1
35
36 static void put_transaction(struct btrfs_transaction *transaction)
37 {
38         WARN_ON(transaction->use_count == 0);
39         transaction->use_count--;
40         if (transaction->use_count == 0) {
41                 WARN_ON(total_trans == 0);
42                 total_trans--;
43                 list_del_init(&transaction->list);
44                 memset(transaction, 0, sizeof(*transaction));
45                 kmem_cache_free(btrfs_transaction_cachep, transaction);
46         }
47 }
48
49 static int join_transaction(struct btrfs_root *root)
50 {
51         struct btrfs_transaction *cur_trans;
52         cur_trans = root->fs_info->running_transaction;
53         if (!cur_trans) {
54                 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
55                                              GFP_NOFS);
56                 total_trans++;
57                 BUG_ON(!cur_trans);
58                 root->fs_info->generation++;
59                 root->fs_info->running_transaction = cur_trans;
60                 cur_trans->num_writers = 1;
61                 cur_trans->num_joined = 0;
62                 cur_trans->transid = root->fs_info->generation;
63                 init_waitqueue_head(&cur_trans->writer_wait);
64                 init_waitqueue_head(&cur_trans->commit_wait);
65                 cur_trans->in_commit = 0;
66                 cur_trans->use_count = 1;
67                 cur_trans->commit_done = 0;
68                 cur_trans->start_time = get_seconds();
69                 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
70                 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
71                 btrfs_ordered_inode_tree_init(&cur_trans->ordered_inode_tree);
72                 extent_map_tree_init(&cur_trans->dirty_pages,
73                                      root->fs_info->btree_inode->i_mapping,
74                                      GFP_NOFS);
75         } else {
76                 cur_trans->num_writers++;
77                 cur_trans->num_joined++;
78         }
79
80         return 0;
81 }
82
83 static int record_root_in_trans(struct btrfs_root *root)
84 {
85         u64 running_trans_id = root->fs_info->running_transaction->transid;
86         if (root->ref_cows && root->last_trans < running_trans_id) {
87                 WARN_ON(root == root->fs_info->extent_root);
88                 if (root->root_item.refs != 0) {
89                         radix_tree_tag_set(&root->fs_info->fs_roots_radix,
90                                    (unsigned long)root->root_key.objectid,
91                                    BTRFS_ROOT_TRANS_TAG);
92                         radix_tree_tag_set(&root->fs_info->fs_roots_radix,
93                                    (unsigned long)root->root_key.objectid,
94                                    BTRFS_ROOT_DEFRAG_TAG);
95                         root->commit_root = root->node;
96                         extent_buffer_get(root->node);
97                 } else {
98                         WARN_ON(1);
99                 }
100                 root->last_trans = running_trans_id;
101         }
102         return 0;
103 }
104
105 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
106                                                    int num_blocks)
107 {
108         struct btrfs_trans_handle *h =
109                 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
110         int ret;
111
112         mutex_lock(&root->fs_info->trans_mutex);
113         ret = join_transaction(root);
114         BUG_ON(ret);
115
116         record_root_in_trans(root);
117         h->transid = root->fs_info->running_transaction->transid;
118         h->transaction = root->fs_info->running_transaction;
119         h->blocks_reserved = num_blocks;
120         h->blocks_used = 0;
121         h->block_group = NULL;
122         h->alloc_exclude_nr = 0;
123         h->alloc_exclude_start = 0;
124         root->fs_info->running_transaction->use_count++;
125         mutex_unlock(&root->fs_info->trans_mutex);
126         return h;
127 }
128
129 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
130                           struct btrfs_root *root)
131 {
132         struct btrfs_transaction *cur_trans;
133
134         mutex_lock(&root->fs_info->trans_mutex);
135         cur_trans = root->fs_info->running_transaction;
136         WARN_ON(cur_trans != trans->transaction);
137         WARN_ON(cur_trans->num_writers < 1);
138         cur_trans->num_writers--;
139         if (waitqueue_active(&cur_trans->writer_wait))
140                 wake_up(&cur_trans->writer_wait);
141         put_transaction(cur_trans);
142         mutex_unlock(&root->fs_info->trans_mutex);
143         memset(trans, 0, sizeof(*trans));
144         kmem_cache_free(btrfs_trans_handle_cachep, trans);
145         return 0;
146 }
147
148
149 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
150                                      struct btrfs_root *root)
151 {
152         int ret;
153         int err;
154         int werr = 0;
155         struct extent_map_tree *dirty_pages;
156         struct page *page;
157         struct inode *btree_inode = root->fs_info->btree_inode;
158         u64 start;
159         u64 end;
160         unsigned long index;
161
162         if (!trans || !trans->transaction) {
163                 return filemap_write_and_wait(btree_inode->i_mapping);
164         }
165         dirty_pages = &trans->transaction->dirty_pages;
166         while(1) {
167                 ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
168                                             EXTENT_DIRTY);
169                 if (ret)
170                         break;
171                 clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
172                 while(start <= end) {
173                         index = start >> PAGE_CACHE_SHIFT;
174                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
175                         page = find_lock_page(btree_inode->i_mapping, index);
176                         if (!page)
177                                 continue;
178                         if (PageWriteback(page)) {
179                                 if (PageDirty(page))
180                                         wait_on_page_writeback(page);
181                                 else {
182                                         unlock_page(page);
183                                         page_cache_release(page);
184                                         continue;
185                                 }
186                         }
187                         err = write_one_page(page, 0);
188                         if (err)
189                                 werr = err;
190                         page_cache_release(page);
191                 }
192         }
193         err = filemap_fdatawait(btree_inode->i_mapping);
194         if (err)
195                 werr = err;
196         return werr;
197 }
198
199 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
200                             struct btrfs_root *root)
201 {
202         int ret;
203         u64 old_extent_block;
204         struct btrfs_fs_info *fs_info = root->fs_info;
205         struct btrfs_root *tree_root = fs_info->tree_root;
206         struct btrfs_root *extent_root = fs_info->extent_root;
207
208         btrfs_write_dirty_block_groups(trans, extent_root);
209         while(1) {
210                 old_extent_block = btrfs_root_bytenr(&extent_root->root_item);
211                 if (old_extent_block == extent_root->node->start)
212                         break;
213                 btrfs_set_root_bytenr(&extent_root->root_item,
214                                       extent_root->node->start);
215                 btrfs_set_root_level(&extent_root->root_item,
216                                      btrfs_header_level(extent_root->node));
217                 ret = btrfs_update_root(trans, tree_root,
218                                         &extent_root->root_key,
219                                         &extent_root->root_item);
220                 BUG_ON(ret);
221                 btrfs_write_dirty_block_groups(trans, extent_root);
222         }
223         return 0;
224 }
225
226 static int wait_for_commit(struct btrfs_root *root,
227                            struct btrfs_transaction *commit)
228 {
229         DEFINE_WAIT(wait);
230         mutex_lock(&root->fs_info->trans_mutex);
231         while(!commit->commit_done) {
232                 prepare_to_wait(&commit->commit_wait, &wait,
233                                 TASK_UNINTERRUPTIBLE);
234                 if (commit->commit_done)
235                         break;
236                 mutex_unlock(&root->fs_info->trans_mutex);
237                 schedule();
238                 mutex_lock(&root->fs_info->trans_mutex);
239         }
240         mutex_unlock(&root->fs_info->trans_mutex);
241         finish_wait(&commit->commit_wait, &wait);
242         return 0;
243 }
244
245 struct dirty_root {
246         struct list_head list;
247         struct btrfs_root *root;
248         struct btrfs_root *latest_root;
249 };
250
251 int btrfs_add_dead_root(struct btrfs_root *root,
252                         struct btrfs_root *latest,
253                         struct list_head *dead_list)
254 {
255         struct dirty_root *dirty;
256
257         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
258         if (!dirty)
259                 return -ENOMEM;
260         dirty->root = root;
261         dirty->latest_root = latest;
262         list_add(&dirty->list, dead_list);
263         return 0;
264 }
265
266 static int add_dirty_roots(struct btrfs_trans_handle *trans,
267                            struct radix_tree_root *radix,
268                            struct list_head *list)
269 {
270         struct dirty_root *dirty;
271         struct btrfs_root *gang[8];
272         struct btrfs_root *root;
273         int i;
274         int ret;
275         int err = 0;
276         u32 refs;
277
278         while(1) {
279                 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
280                                                  ARRAY_SIZE(gang),
281                                                  BTRFS_ROOT_TRANS_TAG);
282                 if (ret == 0)
283                         break;
284                 for (i = 0; i < ret; i++) {
285                         root = gang[i];
286                         radix_tree_tag_clear(radix,
287                                      (unsigned long)root->root_key.objectid,
288                                      BTRFS_ROOT_TRANS_TAG);
289                         if (root->commit_root == root->node) {
290                                 WARN_ON(root->node->start !=
291                                         btrfs_root_bytenr(&root->root_item));
292                                 free_extent_buffer(root->commit_root);
293                                 root->commit_root = NULL;
294
295                                 /* make sure to update the root on disk
296                                  * so we get any updates to the block used
297                                  * counts
298                                  */
299                                 err = btrfs_update_root(trans,
300                                                 root->fs_info->tree_root,
301                                                 &root->root_key,
302                                                 &root->root_item);
303                                 continue;
304                         }
305                         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
306                         BUG_ON(!dirty);
307                         dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
308                         BUG_ON(!dirty->root);
309
310                         memset(&root->root_item.drop_progress, 0,
311                                sizeof(struct btrfs_disk_key));
312                         root->root_item.drop_level = 0;
313
314                         memcpy(dirty->root, root, sizeof(*root));
315                         dirty->root->node = root->commit_root;
316                         dirty->latest_root = root;
317                         root->commit_root = NULL;
318
319                         root->root_key.offset = root->fs_info->generation;
320                         btrfs_set_root_bytenr(&root->root_item,
321                                               root->node->start);
322                         btrfs_set_root_level(&root->root_item,
323                                              btrfs_header_level(root->node));
324                         err = btrfs_insert_root(trans, root->fs_info->tree_root,
325                                                 &root->root_key,
326                                                 &root->root_item);
327                         if (err)
328                                 break;
329
330                         refs = btrfs_root_refs(&dirty->root->root_item);
331                         btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
332                         err = btrfs_update_root(trans, root->fs_info->tree_root,
333                                                 &dirty->root->root_key,
334                                                 &dirty->root->root_item);
335
336                         BUG_ON(err);
337                         if (refs == 1) {
338                                 list_add(&dirty->list, list);
339                         } else {
340                                 WARN_ON(1);
341                                 kfree(dirty->root);
342                                 kfree(dirty);
343                         }
344                 }
345         }
346         return err;
347 }
348
349 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
350 {
351         struct btrfs_fs_info *info = root->fs_info;
352         int ret;
353         struct btrfs_trans_handle *trans;
354         unsigned long nr;
355
356         if (root->defrag_running)
357                 return 0;
358         trans = btrfs_start_transaction(root, 1);
359         while (1) {
360                 root->defrag_running = 1;
361                 ret = btrfs_defrag_leaves(trans, root, cacheonly);
362                 nr = trans->blocks_used;
363                 btrfs_end_transaction(trans, root);
364                 mutex_unlock(&info->fs_mutex);
365                 btrfs_btree_balance_dirty(info->tree_root, nr);
366                 cond_resched();
367
368                 mutex_lock(&info->fs_mutex);
369                 trans = btrfs_start_transaction(root, 1);
370                 if (ret != -EAGAIN)
371                         break;
372         }
373         root->defrag_running = 0;
374         radix_tree_tag_clear(&info->fs_roots_radix,
375                      (unsigned long)root->root_key.objectid,
376                      BTRFS_ROOT_DEFRAG_TAG);
377         btrfs_end_transaction(trans, root);
378         return 0;
379 }
380
381 int btrfs_defrag_dirty_roots(struct btrfs_fs_info *info)
382 {
383         struct btrfs_root *gang[1];
384         struct btrfs_root *root;
385         int i;
386         int ret;
387         int err = 0;
388         u64 last = 0;
389
390         while(1) {
391                 ret = radix_tree_gang_lookup_tag(&info->fs_roots_radix,
392                                                  (void **)gang, last,
393                                                  ARRAY_SIZE(gang),
394                                                  BTRFS_ROOT_DEFRAG_TAG);
395                 if (ret == 0)
396                         break;
397                 for (i = 0; i < ret; i++) {
398                         root = gang[i];
399                         last = root->root_key.objectid + 1;
400                         btrfs_defrag_root(root, 1);
401                 }
402         }
403         btrfs_defrag_root(info->extent_root, 1);
404         return err;
405 }
406
407 static int drop_dirty_roots(struct btrfs_root *tree_root,
408                             struct list_head *list)
409 {
410         struct dirty_root *dirty;
411         struct btrfs_trans_handle *trans;
412         unsigned long nr;
413         u64 num_bytes;
414         u64 bytes_used;
415         int ret = 0;
416         int err;
417
418         while(!list_empty(list)) {
419                 struct btrfs_root *root;
420
421                 mutex_lock(&tree_root->fs_info->fs_mutex);
422                 dirty = list_entry(list->next, struct dirty_root, list);
423                 list_del_init(&dirty->list);
424
425                 num_bytes = btrfs_root_used(&dirty->root->root_item);
426                 root = dirty->latest_root;
427                 root->fs_info->throttles++;
428
429                 while(1) {
430                         trans = btrfs_start_transaction(tree_root, 1);
431                         ret = btrfs_drop_snapshot(trans, dirty->root);
432                         if (ret != -EAGAIN) {
433                                 break;
434                         }
435
436                         err = btrfs_update_root(trans,
437                                         tree_root,
438                                         &dirty->root->root_key,
439                                         &dirty->root->root_item);
440                         if (err)
441                                 ret = err;
442                         nr = trans->blocks_used;
443                         ret = btrfs_end_transaction(trans, tree_root);
444                         BUG_ON(ret);
445                         mutex_unlock(&tree_root->fs_info->fs_mutex);
446                         btrfs_btree_balance_dirty(tree_root, nr);
447                         cond_resched();
448                         mutex_lock(&tree_root->fs_info->fs_mutex);
449                 }
450                 BUG_ON(ret);
451                 root->fs_info->throttles--;
452
453                 num_bytes -= btrfs_root_used(&dirty->root->root_item);
454                 bytes_used = btrfs_root_used(&root->root_item);
455                 if (num_bytes) {
456                         record_root_in_trans(root);
457                         btrfs_set_root_used(&root->root_item,
458                                             bytes_used - num_bytes);
459                 }
460                 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
461                 if (ret) {
462                         BUG();
463                         break;
464                 }
465                 nr = trans->blocks_used;
466                 ret = btrfs_end_transaction(trans, tree_root);
467                 BUG_ON(ret);
468
469                 free_extent_buffer(dirty->root->node);
470                 kfree(dirty->root);
471                 kfree(dirty);
472                 mutex_unlock(&tree_root->fs_info->fs_mutex);
473
474                 btrfs_btree_balance_dirty(tree_root, nr);
475                 cond_resched();
476         }
477         return ret;
478 }
479
480 int btrfs_write_ordered_inodes(struct btrfs_trans_handle *trans,
481                                 struct btrfs_root *root)
482 {
483         struct btrfs_transaction *cur_trans = trans->transaction;
484         struct inode *inode;
485         u64 root_objectid = 0;
486         u64 objectid = 0;
487         int ret;
488
489         root->fs_info->throttles++;
490         while(1) {
491                 ret = btrfs_find_first_ordered_inode(
492                                 &cur_trans->ordered_inode_tree,
493                                 &root_objectid, &objectid);
494                 if (!ret)
495                         break;
496
497                 mutex_unlock(&root->fs_info->trans_mutex);
498                 mutex_unlock(&root->fs_info->fs_mutex);
499                 inode = btrfs_ilookup(root->fs_info->sb, objectid,
500                                       root_objectid);
501                 if (inode) {
502                         if (S_ISREG(inode->i_mode))
503                                 filemap_fdatawrite(inode->i_mapping);
504                         iput(inode);
505                 }
506                 mutex_lock(&root->fs_info->fs_mutex);
507                 mutex_lock(&root->fs_info->trans_mutex);
508         }
509         while(1) {
510                 root_objectid = 0;
511                 objectid = 0;
512                 ret = btrfs_find_del_first_ordered_inode(
513                                 &cur_trans->ordered_inode_tree,
514                                 &root_objectid, &objectid);
515                 if (!ret)
516                         break;
517                 mutex_unlock(&root->fs_info->trans_mutex);
518                 mutex_unlock(&root->fs_info->fs_mutex);
519                 inode = btrfs_ilookup(root->fs_info->sb, objectid,
520                                       root_objectid);
521                 if (inode) {
522                         if (S_ISREG(inode->i_mode))
523                                 filemap_write_and_wait(inode->i_mapping);
524                         atomic_dec(&inode->i_count);
525                         iput(inode);
526                 }
527                 mutex_lock(&root->fs_info->fs_mutex);
528                 mutex_lock(&root->fs_info->trans_mutex);
529         }
530         root->fs_info->throttles--;
531         return 0;
532 }
533
534 static int create_pending_snapshot(struct btrfs_trans_handle *trans,
535                                    struct btrfs_fs_info *fs_info,
536                                    struct btrfs_pending_snapshot *pending)
537 {
538         struct btrfs_key key;
539         struct btrfs_root_item new_root_item;
540         struct btrfs_root *tree_root = fs_info->tree_root;
541         struct btrfs_root *root = pending->root;
542         struct extent_buffer *tmp;
543         int ret;
544         u64 objectid;
545
546         ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
547         if (ret)
548                 goto fail;
549
550         memcpy(&new_root_item, &root->root_item, sizeof(new_root_item));
551
552         key.objectid = objectid;
553         key.offset = 1;
554         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
555
556         extent_buffer_get(root->node);
557         btrfs_cow_block(trans, root, root->node, NULL, 0, &tmp);
558         free_extent_buffer(tmp);
559
560         btrfs_copy_root(trans, root, root->node, &tmp, objectid);
561
562         btrfs_set_root_bytenr(&new_root_item, tmp->start);
563         btrfs_set_root_level(&new_root_item, btrfs_header_level(tmp));
564         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
565                                 &new_root_item);
566         free_extent_buffer(tmp);
567         if (ret)
568                 goto fail;
569
570         /*
571          * insert the directory item
572          */
573         key.offset = (u64)-1;
574         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
575                                     pending->name, strlen(pending->name),
576                                     root->fs_info->sb->s_root->d_inode->i_ino,
577                                     &key, BTRFS_FT_DIR);
578
579         if (ret)
580                 goto fail;
581
582         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
583                              pending->name, strlen(pending->name), objectid,
584                              root->fs_info->sb->s_root->d_inode->i_ino);
585 fail:
586         return ret;
587 }
588
589 static int create_pending_snapshots(struct btrfs_trans_handle *trans,
590                                    struct btrfs_fs_info *fs_info)
591 {
592         struct btrfs_pending_snapshot *pending;
593         struct list_head *head = &trans->transaction->pending_snapshots;
594         int ret;
595
596         while(!list_empty(head)) {
597                 pending = list_entry(head->next,
598                                      struct btrfs_pending_snapshot, list);
599                 ret = create_pending_snapshot(trans, fs_info, pending);
600                 BUG_ON(ret);
601                 list_del(&pending->list);
602                 kfree(pending->name);
603                 kfree(pending);
604         }
605         return 0;
606 }
607
608 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
609                              struct btrfs_root *root)
610 {
611         unsigned long joined = 0;
612         unsigned long timeout = 1;
613         struct btrfs_transaction *cur_trans;
614         struct btrfs_transaction *prev_trans = NULL;
615         struct list_head dirty_fs_roots;
616         struct extent_map_tree *pinned_copy;
617         DEFINE_WAIT(wait);
618         int ret;
619
620         INIT_LIST_HEAD(&dirty_fs_roots);
621
622         mutex_lock(&root->fs_info->trans_mutex);
623         if (trans->transaction->in_commit) {
624                 cur_trans = trans->transaction;
625                 trans->transaction->use_count++;
626                 mutex_unlock(&root->fs_info->trans_mutex);
627                 btrfs_end_transaction(trans, root);
628
629                 mutex_unlock(&root->fs_info->fs_mutex);
630                 ret = wait_for_commit(root, cur_trans);
631                 BUG_ON(ret);
632
633                 mutex_lock(&root->fs_info->trans_mutex);
634                 put_transaction(cur_trans);
635                 mutex_unlock(&root->fs_info->trans_mutex);
636
637                 mutex_lock(&root->fs_info->fs_mutex);
638                 return 0;
639         }
640
641         pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
642         if (!pinned_copy)
643                 return -ENOMEM;
644
645         extent_map_tree_init(pinned_copy,
646                              root->fs_info->btree_inode->i_mapping, GFP_NOFS);
647
648         trans->transaction->in_commit = 1;
649         cur_trans = trans->transaction;
650         if (cur_trans->list.prev != &root->fs_info->trans_list) {
651                 prev_trans = list_entry(cur_trans->list.prev,
652                                         struct btrfs_transaction, list);
653                 if (!prev_trans->commit_done) {
654                         prev_trans->use_count++;
655                         mutex_unlock(&root->fs_info->fs_mutex);
656                         mutex_unlock(&root->fs_info->trans_mutex);
657
658                         wait_for_commit(root, prev_trans);
659
660                         mutex_lock(&root->fs_info->fs_mutex);
661                         mutex_lock(&root->fs_info->trans_mutex);
662                         put_transaction(prev_trans);
663                 }
664         }
665
666         do {
667                 joined = cur_trans->num_joined;
668                 WARN_ON(cur_trans != trans->transaction);
669                 prepare_to_wait(&cur_trans->writer_wait, &wait,
670                                 TASK_UNINTERRUPTIBLE);
671
672                 if (cur_trans->num_writers > 1)
673                         timeout = MAX_SCHEDULE_TIMEOUT;
674                 else
675                         timeout = 1;
676
677                 mutex_unlock(&root->fs_info->fs_mutex);
678                 mutex_unlock(&root->fs_info->trans_mutex);
679
680                 schedule_timeout(timeout);
681
682                 mutex_lock(&root->fs_info->fs_mutex);
683                 mutex_lock(&root->fs_info->trans_mutex);
684                 finish_wait(&cur_trans->writer_wait, &wait);
685                 ret = btrfs_write_ordered_inodes(trans, root);
686
687         } while (cur_trans->num_writers > 1 ||
688                  (cur_trans->num_joined != joined));
689
690         ret = create_pending_snapshots(trans, root->fs_info);
691         BUG_ON(ret);
692
693         WARN_ON(cur_trans != trans->transaction);
694
695         ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
696                               &dirty_fs_roots);
697         BUG_ON(ret);
698
699         ret = btrfs_commit_tree_roots(trans, root);
700         BUG_ON(ret);
701
702         cur_trans = root->fs_info->running_transaction;
703         spin_lock(&root->fs_info->new_trans_lock);
704         root->fs_info->running_transaction = NULL;
705         spin_unlock(&root->fs_info->new_trans_lock);
706         btrfs_set_super_generation(&root->fs_info->super_copy,
707                                    cur_trans->transid);
708         btrfs_set_super_root(&root->fs_info->super_copy,
709                              root->fs_info->tree_root->node->start);
710         btrfs_set_super_root_level(&root->fs_info->super_copy,
711                            btrfs_header_level(root->fs_info->tree_root->node));
712
713         write_extent_buffer(root->fs_info->sb_buffer,
714                             &root->fs_info->super_copy, 0,
715                             sizeof(root->fs_info->super_copy));
716
717         btrfs_copy_pinned(root, pinned_copy);
718
719         mutex_unlock(&root->fs_info->trans_mutex);
720         mutex_unlock(&root->fs_info->fs_mutex);
721         ret = btrfs_write_and_wait_transaction(trans, root);
722         BUG_ON(ret);
723         write_ctree_super(trans, root);
724
725         mutex_lock(&root->fs_info->fs_mutex);
726         btrfs_finish_extent_commit(trans, root, pinned_copy);
727         mutex_lock(&root->fs_info->trans_mutex);
728
729         kfree(pinned_copy);
730
731         cur_trans->commit_done = 1;
732         root->fs_info->last_trans_committed = cur_trans->transid;
733         wake_up(&cur_trans->commit_wait);
734         put_transaction(cur_trans);
735         put_transaction(cur_trans);
736
737         if (root->fs_info->closing)
738                 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
739         else
740                 list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
741
742         mutex_unlock(&root->fs_info->trans_mutex);
743         kmem_cache_free(btrfs_trans_handle_cachep, trans);
744
745         if (root->fs_info->closing) {
746                 mutex_unlock(&root->fs_info->fs_mutex);
747                 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
748                 mutex_lock(&root->fs_info->fs_mutex);
749         }
750         return ret;
751 }
752
753 int btrfs_clean_old_snapshots(struct btrfs_root *root)
754 {
755         struct list_head dirty_roots;
756         INIT_LIST_HEAD(&dirty_roots);
757
758         mutex_lock(&root->fs_info->trans_mutex);
759         list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
760         mutex_unlock(&root->fs_info->trans_mutex);
761
762         if (!list_empty(&dirty_roots)) {
763                 drop_dirty_roots(root, &dirty_roots);
764         }
765         return 0;
766 }
767 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
768 void btrfs_transaction_cleaner(void *p)
769 #else
770 void btrfs_transaction_cleaner(struct work_struct *work)
771 #endif
772 {
773 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
774         struct btrfs_fs_info *fs_info = p;
775 #else
776         struct btrfs_fs_info *fs_info = container_of(work,
777                                                      struct btrfs_fs_info,
778                                                      trans_work.work);
779
780 #endif
781         struct btrfs_root *root = fs_info->tree_root;
782         struct btrfs_transaction *cur;
783         struct btrfs_trans_handle *trans;
784         unsigned long now;
785         unsigned long delay = HZ * 30;
786         int ret;
787
788         mutex_lock(&root->fs_info->fs_mutex);
789         mutex_lock(&root->fs_info->trans_mutex);
790         cur = root->fs_info->running_transaction;
791         if (!cur) {
792                 mutex_unlock(&root->fs_info->trans_mutex);
793                 goto out;
794         }
795         now = get_seconds();
796         if (now < cur->start_time || now - cur->start_time < 30) {
797                 mutex_unlock(&root->fs_info->trans_mutex);
798                 delay = HZ * 5;
799                 goto out;
800         }
801         mutex_unlock(&root->fs_info->trans_mutex);
802         btrfs_defrag_dirty_roots(root->fs_info);
803         trans = btrfs_start_transaction(root, 1);
804         ret = btrfs_commit_transaction(trans, root);
805 out:
806         mutex_unlock(&root->fs_info->fs_mutex);
807         btrfs_clean_old_snapshots(root);
808         btrfs_transaction_queue_work(root, delay);
809 }
810
811 void btrfs_transaction_queue_work(struct btrfs_root *root, int delay)
812 {
813         queue_delayed_work(trans_wq, &root->fs_info->trans_work, delay);
814 }
815
816 void btrfs_transaction_flush_work(struct btrfs_root *root)
817 {
818         cancel_rearming_delayed_workqueue(trans_wq, &root->fs_info->trans_work);
819         flush_workqueue(trans_wq);
820 }
821
822 void __init btrfs_init_transaction_sys(void)
823 {
824         trans_wq = create_workqueue("btrfs");
825 }
826
827 void btrfs_exit_transaction_sys(void)
828 {
829         destroy_workqueue(trans_wq);
830 }
831