Btrfs: Start btree concurrency work.
[linux-2.6-block.git] / 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 #include "locking.h"
27
28 static int total_trans = 0;
29 extern struct kmem_cache *btrfs_trans_handle_cachep;
30 extern struct kmem_cache *btrfs_transaction_cachep;
31
32 static struct workqueue_struct *trans_wq;
33
34 #define BTRFS_ROOT_TRANS_TAG 0
35 #define BTRFS_ROOT_DEFRAG_TAG 1
36
37 static noinline void put_transaction(struct btrfs_transaction *transaction)
38 {
39         WARN_ON(transaction->use_count == 0);
40         transaction->use_count--;
41         if (transaction->use_count == 0) {
42                 WARN_ON(total_trans == 0);
43                 total_trans--;
44                 list_del_init(&transaction->list);
45                 memset(transaction, 0, sizeof(*transaction));
46                 kmem_cache_free(btrfs_transaction_cachep, transaction);
47         }
48 }
49
50 static noinline int join_transaction(struct btrfs_root *root)
51 {
52         struct btrfs_transaction *cur_trans;
53         cur_trans = root->fs_info->running_transaction;
54         if (!cur_trans) {
55                 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
56                                              GFP_NOFS);
57                 total_trans++;
58                 BUG_ON(!cur_trans);
59                 root->fs_info->generation++;
60                 root->fs_info->last_alloc = 0;
61                 root->fs_info->last_data_alloc = 0;
62                 cur_trans->num_writers = 1;
63                 cur_trans->num_joined = 0;
64                 cur_trans->transid = root->fs_info->generation;
65                 init_waitqueue_head(&cur_trans->writer_wait);
66                 init_waitqueue_head(&cur_trans->commit_wait);
67                 cur_trans->in_commit = 0;
68                 cur_trans->use_count = 1;
69                 cur_trans->commit_done = 0;
70                 cur_trans->start_time = get_seconds();
71                 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
72                 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
73                 btrfs_ordered_inode_tree_init(&cur_trans->ordered_inode_tree);
74                 extent_io_tree_init(&cur_trans->dirty_pages,
75                                      root->fs_info->btree_inode->i_mapping,
76                                      GFP_NOFS);
77                 spin_lock(&root->fs_info->new_trans_lock);
78                 root->fs_info->running_transaction = cur_trans;
79                 spin_unlock(&root->fs_info->new_trans_lock);
80         } else {
81                 cur_trans->num_writers++;
82                 cur_trans->num_joined++;
83         }
84
85         return 0;
86 }
87
88 static noinline int record_root_in_trans(struct btrfs_root *root)
89 {
90         u64 running_trans_id = root->fs_info->running_transaction->transid;
91         if (root->ref_cows && root->last_trans < running_trans_id) {
92                 WARN_ON(root == root->fs_info->extent_root);
93                 if (root->root_item.refs != 0) {
94                         radix_tree_tag_set(&root->fs_info->fs_roots_radix,
95                                    (unsigned long)root->root_key.objectid,
96                                    BTRFS_ROOT_TRANS_TAG);
97                         radix_tree_tag_set(&root->fs_info->fs_roots_radix,
98                                    (unsigned long)root->root_key.objectid,
99                                    BTRFS_ROOT_DEFRAG_TAG);
100                         root->commit_root = btrfs_root_node(root);
101                 } else {
102                         WARN_ON(1);
103                 }
104                 root->last_trans = running_trans_id;
105         }
106         return 0;
107 }
108
109 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
110                                                    int num_blocks)
111 {
112         struct btrfs_trans_handle *h =
113                 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
114         int ret;
115
116         mutex_lock(&root->fs_info->trans_mutex);
117         ret = join_transaction(root);
118         BUG_ON(ret);
119
120         record_root_in_trans(root);
121         h->transid = root->fs_info->running_transaction->transid;
122         h->transaction = root->fs_info->running_transaction;
123         h->blocks_reserved = num_blocks;
124         h->blocks_used = 0;
125         h->block_group = NULL;
126         h->alloc_exclude_nr = 0;
127         h->alloc_exclude_start = 0;
128         root->fs_info->running_transaction->use_count++;
129         mutex_unlock(&root->fs_info->trans_mutex);
130         return h;
131 }
132
133 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
134                           struct btrfs_root *root)
135 {
136         struct btrfs_transaction *cur_trans;
137
138         mutex_lock(&root->fs_info->trans_mutex);
139         cur_trans = root->fs_info->running_transaction;
140         WARN_ON(cur_trans != trans->transaction);
141         WARN_ON(cur_trans->num_writers < 1);
142         cur_trans->num_writers--;
143         if (waitqueue_active(&cur_trans->writer_wait))
144                 wake_up(&cur_trans->writer_wait);
145         put_transaction(cur_trans);
146         mutex_unlock(&root->fs_info->trans_mutex);
147         memset(trans, 0, sizeof(*trans));
148         kmem_cache_free(btrfs_trans_handle_cachep, trans);
149         return 0;
150 }
151
152
153 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
154                                      struct btrfs_root *root)
155 {
156         int ret;
157         int err;
158         int werr = 0;
159         struct extent_io_tree *dirty_pages;
160         struct page *page;
161         struct inode *btree_inode = root->fs_info->btree_inode;
162         u64 start;
163         u64 end;
164         unsigned long index;
165
166         if (!trans || !trans->transaction) {
167                 return filemap_write_and_wait(btree_inode->i_mapping);
168         }
169         dirty_pages = &trans->transaction->dirty_pages;
170         while(1) {
171                 ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
172                                             EXTENT_DIRTY);
173                 if (ret)
174                         break;
175                 clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
176                 while(start <= end) {
177                         index = start >> PAGE_CACHE_SHIFT;
178                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
179                         page = find_lock_page(btree_inode->i_mapping, index);
180                         if (!page)
181                                 continue;
182                         if (PageWriteback(page)) {
183                                 if (PageDirty(page))
184                                         wait_on_page_writeback(page);
185                                 else {
186                                         unlock_page(page);
187                                         page_cache_release(page);
188                                         continue;
189                                 }
190                         }
191                         err = write_one_page(page, 0);
192                         if (err)
193                                 werr = err;
194                         page_cache_release(page);
195                 }
196         }
197         err = filemap_fdatawait(btree_inode->i_mapping);
198         if (err)
199                 werr = err;
200         return werr;
201 }
202
203 static int update_cowonly_root(struct btrfs_trans_handle *trans,
204                                struct btrfs_root *root)
205 {
206         int ret;
207         u64 old_root_bytenr;
208         struct btrfs_root *tree_root = root->fs_info->tree_root;
209
210         btrfs_write_dirty_block_groups(trans, root);
211         while(1) {
212                 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
213                 if (old_root_bytenr == root->node->start)
214                         break;
215                 btrfs_set_root_bytenr(&root->root_item,
216                                        root->node->start);
217                 btrfs_set_root_level(&root->root_item,
218                                      btrfs_header_level(root->node));
219                 ret = btrfs_update_root(trans, tree_root,
220                                         &root->root_key,
221                                         &root->root_item);
222                 BUG_ON(ret);
223                 btrfs_write_dirty_block_groups(trans, root);
224         }
225         return 0;
226 }
227
228 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
229                             struct btrfs_root *root)
230 {
231         struct btrfs_fs_info *fs_info = root->fs_info;
232         struct list_head *next;
233
234         while(!list_empty(&fs_info->dirty_cowonly_roots)) {
235                 next = fs_info->dirty_cowonly_roots.next;
236                 list_del_init(next);
237                 root = list_entry(next, struct btrfs_root, dirty_list);
238                 update_cowonly_root(trans, root);
239         }
240         return 0;
241 }
242
243 static noinline int wait_for_commit(struct btrfs_root *root,
244                                     struct btrfs_transaction *commit)
245 {
246         DEFINE_WAIT(wait);
247         mutex_lock(&root->fs_info->trans_mutex);
248         while(!commit->commit_done) {
249                 prepare_to_wait(&commit->commit_wait, &wait,
250                                 TASK_UNINTERRUPTIBLE);
251                 if (commit->commit_done)
252                         break;
253                 mutex_unlock(&root->fs_info->trans_mutex);
254                 schedule();
255                 mutex_lock(&root->fs_info->trans_mutex);
256         }
257         mutex_unlock(&root->fs_info->trans_mutex);
258         finish_wait(&commit->commit_wait, &wait);
259         return 0;
260 }
261
262 struct dirty_root {
263         struct list_head list;
264         struct btrfs_root *root;
265         struct btrfs_root *latest_root;
266 };
267
268 int btrfs_add_dead_root(struct btrfs_root *root,
269                         struct btrfs_root *latest,
270                         struct list_head *dead_list)
271 {
272         struct dirty_root *dirty;
273
274         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
275         if (!dirty)
276                 return -ENOMEM;
277         dirty->root = root;
278         dirty->latest_root = latest;
279         list_add(&dirty->list, dead_list);
280         return 0;
281 }
282
283 static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
284                                     struct radix_tree_root *radix,
285                                     struct list_head *list)
286 {
287         struct dirty_root *dirty;
288         struct btrfs_root *gang[8];
289         struct btrfs_root *root;
290         int i;
291         int ret;
292         int err = 0;
293         u32 refs;
294
295         while(1) {
296                 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
297                                                  ARRAY_SIZE(gang),
298                                                  BTRFS_ROOT_TRANS_TAG);
299                 if (ret == 0)
300                         break;
301                 for (i = 0; i < ret; i++) {
302                         root = gang[i];
303                         radix_tree_tag_clear(radix,
304                                      (unsigned long)root->root_key.objectid,
305                                      BTRFS_ROOT_TRANS_TAG);
306                         if (root->commit_root == root->node) {
307                                 WARN_ON(root->node->start !=
308                                         btrfs_root_bytenr(&root->root_item));
309                                 free_extent_buffer(root->commit_root);
310                                 root->commit_root = NULL;
311
312                                 /* make sure to update the root on disk
313                                  * so we get any updates to the block used
314                                  * counts
315                                  */
316                                 err = btrfs_update_root(trans,
317                                                 root->fs_info->tree_root,
318                                                 &root->root_key,
319                                                 &root->root_item);
320                                 continue;
321                         }
322                         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
323                         BUG_ON(!dirty);
324                         dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
325                         BUG_ON(!dirty->root);
326
327                         memset(&root->root_item.drop_progress, 0,
328                                sizeof(struct btrfs_disk_key));
329                         root->root_item.drop_level = 0;
330
331                         memcpy(dirty->root, root, sizeof(*root));
332                         dirty->root->node = root->commit_root;
333                         dirty->latest_root = root;
334                         root->commit_root = NULL;
335
336                         root->root_key.offset = root->fs_info->generation;
337                         btrfs_set_root_bytenr(&root->root_item,
338                                               root->node->start);
339                         btrfs_set_root_level(&root->root_item,
340                                              btrfs_header_level(root->node));
341                         err = btrfs_insert_root(trans, root->fs_info->tree_root,
342                                                 &root->root_key,
343                                                 &root->root_item);
344                         if (err)
345                                 break;
346
347                         refs = btrfs_root_refs(&dirty->root->root_item);
348                         btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
349                         err = btrfs_update_root(trans, root->fs_info->tree_root,
350                                                 &dirty->root->root_key,
351                                                 &dirty->root->root_item);
352
353                         BUG_ON(err);
354                         if (refs == 1) {
355                                 list_add(&dirty->list, list);
356                         } else {
357                                 WARN_ON(1);
358                                 kfree(dirty->root);
359                                 kfree(dirty);
360                         }
361                 }
362         }
363         return err;
364 }
365
366 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
367 {
368         struct btrfs_fs_info *info = root->fs_info;
369         int ret;
370         struct btrfs_trans_handle *trans;
371         unsigned long nr;
372
373         if (root->defrag_running)
374                 return 0;
375         trans = btrfs_start_transaction(root, 1);
376         while (1) {
377                 root->defrag_running = 1;
378                 ret = btrfs_defrag_leaves(trans, root, cacheonly);
379                 nr = trans->blocks_used;
380                 btrfs_end_transaction(trans, root);
381                 mutex_unlock(&info->fs_mutex);
382                 btrfs_btree_balance_dirty(info->tree_root, nr);
383                 cond_resched();
384
385                 mutex_lock(&info->fs_mutex);
386                 trans = btrfs_start_transaction(root, 1);
387                 if (ret != -EAGAIN)
388                         break;
389         }
390         root->defrag_running = 0;
391         radix_tree_tag_clear(&info->fs_roots_radix,
392                      (unsigned long)root->root_key.objectid,
393                      BTRFS_ROOT_DEFRAG_TAG);
394         btrfs_end_transaction(trans, root);
395         return 0;
396 }
397
398 int btrfs_defrag_dirty_roots(struct btrfs_fs_info *info)
399 {
400         struct btrfs_root *gang[1];
401         struct btrfs_root *root;
402         int i;
403         int ret;
404         int err = 0;
405         u64 last = 0;
406
407         while(1) {
408                 ret = radix_tree_gang_lookup_tag(&info->fs_roots_radix,
409                                                  (void **)gang, last,
410                                                  ARRAY_SIZE(gang),
411                                                  BTRFS_ROOT_DEFRAG_TAG);
412                 if (ret == 0)
413                         break;
414                 for (i = 0; i < ret; i++) {
415                         root = gang[i];
416                         last = root->root_key.objectid + 1;
417                         btrfs_defrag_root(root, 1);
418                 }
419         }
420         btrfs_defrag_root(info->extent_root, 1);
421         return err;
422 }
423
424 static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
425                                      struct list_head *list)
426 {
427         struct dirty_root *dirty;
428         struct btrfs_trans_handle *trans;
429         unsigned long nr;
430         u64 num_bytes;
431         u64 bytes_used;
432         int ret = 0;
433         int err;
434
435         while(!list_empty(list)) {
436                 struct btrfs_root *root;
437
438                 mutex_lock(&tree_root->fs_info->fs_mutex);
439                 dirty = list_entry(list->next, struct dirty_root, list);
440                 list_del_init(&dirty->list);
441
442                 num_bytes = btrfs_root_used(&dirty->root->root_item);
443                 root = dirty->latest_root;
444                 root->fs_info->throttles++;
445
446                 while(1) {
447                         trans = btrfs_start_transaction(tree_root, 1);
448                         ret = btrfs_drop_snapshot(trans, dirty->root);
449                         if (ret != -EAGAIN) {
450                                 break;
451                         }
452
453                         err = btrfs_update_root(trans,
454                                         tree_root,
455                                         &dirty->root->root_key,
456                                         &dirty->root->root_item);
457                         if (err)
458                                 ret = err;
459                         nr = trans->blocks_used;
460                         ret = btrfs_end_transaction(trans, tree_root);
461                         BUG_ON(ret);
462                         mutex_unlock(&tree_root->fs_info->fs_mutex);
463                         btrfs_btree_balance_dirty(tree_root, nr);
464                         cond_resched();
465                         mutex_lock(&tree_root->fs_info->fs_mutex);
466                 }
467                 BUG_ON(ret);
468                 root->fs_info->throttles--;
469
470                 num_bytes -= btrfs_root_used(&dirty->root->root_item);
471                 bytes_used = btrfs_root_used(&root->root_item);
472                 if (num_bytes) {
473                         record_root_in_trans(root);
474                         btrfs_set_root_used(&root->root_item,
475                                             bytes_used - num_bytes);
476                 }
477                 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
478                 if (ret) {
479                         BUG();
480                         break;
481                 }
482                 nr = trans->blocks_used;
483                 ret = btrfs_end_transaction(trans, tree_root);
484                 BUG_ON(ret);
485
486                 free_extent_buffer(dirty->root->node);
487                 kfree(dirty->root);
488                 kfree(dirty);
489                 mutex_unlock(&tree_root->fs_info->fs_mutex);
490
491                 btrfs_btree_balance_dirty(tree_root, nr);
492                 cond_resched();
493         }
494         return ret;
495 }
496
497 int btrfs_write_ordered_inodes(struct btrfs_trans_handle *trans,
498                                 struct btrfs_root *root)
499 {
500         struct btrfs_transaction *cur_trans = trans->transaction;
501         struct inode *inode;
502         u64 root_objectid = 0;
503         u64 objectid = 0;
504         int ret;
505
506         root->fs_info->throttles++;
507         while(1) {
508                 ret = btrfs_find_first_ordered_inode(
509                                 &cur_trans->ordered_inode_tree,
510                                 &root_objectid, &objectid, &inode);
511                 if (!ret)
512                         break;
513
514                 mutex_unlock(&root->fs_info->trans_mutex);
515                 mutex_unlock(&root->fs_info->fs_mutex);
516
517                 if (S_ISREG(inode->i_mode)) {
518                         atomic_inc(&BTRFS_I(inode)->ordered_writeback);
519                         filemap_fdatawrite(inode->i_mapping);
520                         atomic_dec(&BTRFS_I(inode)->ordered_writeback);
521                 }
522                 iput(inode);
523
524                 mutex_lock(&root->fs_info->fs_mutex);
525                 mutex_lock(&root->fs_info->trans_mutex);
526         }
527         while(1) {
528                 root_objectid = 0;
529                 objectid = 0;
530                 ret = btrfs_find_del_first_ordered_inode(
531                                 &cur_trans->ordered_inode_tree,
532                                 &root_objectid, &objectid, &inode);
533                 if (!ret)
534                         break;
535                 mutex_unlock(&root->fs_info->trans_mutex);
536                 mutex_unlock(&root->fs_info->fs_mutex);
537
538                 if (S_ISREG(inode->i_mode)) {
539                         atomic_inc(&BTRFS_I(inode)->ordered_writeback);
540                         filemap_write_and_wait(inode->i_mapping);
541                         atomic_dec(&BTRFS_I(inode)->ordered_writeback);
542                 }
543                 atomic_dec(&inode->i_count);
544                 iput(inode);
545
546                 mutex_lock(&root->fs_info->fs_mutex);
547                 mutex_lock(&root->fs_info->trans_mutex);
548         }
549         root->fs_info->throttles--;
550         return 0;
551 }
552
553 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
554                                    struct btrfs_fs_info *fs_info,
555                                    struct btrfs_pending_snapshot *pending)
556 {
557         struct btrfs_key key;
558         struct btrfs_root_item *new_root_item;
559         struct btrfs_root *tree_root = fs_info->tree_root;
560         struct btrfs_root *root = pending->root;
561         struct extent_buffer *tmp;
562         struct extent_buffer *old;
563         int ret;
564         int namelen;
565         u64 objectid;
566
567         new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
568         if (!new_root_item) {
569                 ret = -ENOMEM;
570                 goto fail;
571         }
572         ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
573         if (ret)
574                 goto fail;
575
576         memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
577
578         key.objectid = objectid;
579         key.offset = 1;
580         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
581
582         old = btrfs_lock_root_node(root);
583         btrfs_cow_block(trans, root, old, NULL, 0, &old);
584
585         btrfs_copy_root(trans, root, old, &tmp, objectid);
586         btrfs_tree_unlock(old);
587         free_extent_buffer(old);
588
589         btrfs_set_root_bytenr(new_root_item, tmp->start);
590         btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
591         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
592                                 new_root_item);
593         btrfs_tree_unlock(tmp);
594         free_extent_buffer(tmp);
595         if (ret)
596                 goto fail;
597
598         /*
599          * insert the directory item
600          */
601         key.offset = (u64)-1;
602         namelen = strlen(pending->name);
603         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
604                                     pending->name, namelen,
605                                     root->fs_info->sb->s_root->d_inode->i_ino,
606                                     &key, BTRFS_FT_DIR);
607
608         if (ret)
609                 goto fail;
610
611         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
612                              pending->name, strlen(pending->name), objectid,
613                              root->fs_info->sb->s_root->d_inode->i_ino);
614
615         /* Invalidate existing dcache entry for new snapshot. */
616         btrfs_invalidate_dcache_root(root, pending->name, namelen);
617
618 fail:
619         kfree(new_root_item);
620         return ret;
621 }
622
623 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
624                                              struct btrfs_fs_info *fs_info)
625 {
626         struct btrfs_pending_snapshot *pending;
627         struct list_head *head = &trans->transaction->pending_snapshots;
628         int ret;
629
630         while(!list_empty(head)) {
631                 pending = list_entry(head->next,
632                                      struct btrfs_pending_snapshot, list);
633                 ret = create_pending_snapshot(trans, fs_info, pending);
634                 BUG_ON(ret);
635                 list_del(&pending->list);
636                 kfree(pending->name);
637                 kfree(pending);
638         }
639         return 0;
640 }
641
642 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
643                              struct btrfs_root *root)
644 {
645         unsigned long joined = 0;
646         unsigned long timeout = 1;
647         struct btrfs_transaction *cur_trans;
648         struct btrfs_transaction *prev_trans = NULL;
649         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
650         struct list_head dirty_fs_roots;
651         struct extent_io_tree *pinned_copy;
652         DEFINE_WAIT(wait);
653         int ret;
654
655         INIT_LIST_HEAD(&dirty_fs_roots);
656
657         mutex_lock(&root->fs_info->trans_mutex);
658         if (trans->transaction->in_commit) {
659                 cur_trans = trans->transaction;
660                 trans->transaction->use_count++;
661                 mutex_unlock(&root->fs_info->trans_mutex);
662                 btrfs_end_transaction(trans, root);
663
664                 mutex_unlock(&root->fs_info->fs_mutex);
665                 ret = wait_for_commit(root, cur_trans);
666                 BUG_ON(ret);
667
668                 mutex_lock(&root->fs_info->trans_mutex);
669                 put_transaction(cur_trans);
670                 mutex_unlock(&root->fs_info->trans_mutex);
671
672                 mutex_lock(&root->fs_info->fs_mutex);
673                 return 0;
674         }
675
676         pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
677         if (!pinned_copy)
678                 return -ENOMEM;
679
680         extent_io_tree_init(pinned_copy,
681                              root->fs_info->btree_inode->i_mapping, GFP_NOFS);
682
683         trans->transaction->in_commit = 1;
684         cur_trans = trans->transaction;
685         if (cur_trans->list.prev != &root->fs_info->trans_list) {
686                 prev_trans = list_entry(cur_trans->list.prev,
687                                         struct btrfs_transaction, list);
688                 if (!prev_trans->commit_done) {
689                         prev_trans->use_count++;
690                         mutex_unlock(&root->fs_info->fs_mutex);
691                         mutex_unlock(&root->fs_info->trans_mutex);
692
693                         wait_for_commit(root, prev_trans);
694
695                         mutex_lock(&root->fs_info->fs_mutex);
696                         mutex_lock(&root->fs_info->trans_mutex);
697                         put_transaction(prev_trans);
698                 }
699         }
700
701         do {
702                 joined = cur_trans->num_joined;
703                 WARN_ON(cur_trans != trans->transaction);
704                 prepare_to_wait(&cur_trans->writer_wait, &wait,
705                                 TASK_UNINTERRUPTIBLE);
706
707                 if (cur_trans->num_writers > 1)
708                         timeout = MAX_SCHEDULE_TIMEOUT;
709                 else
710                         timeout = 1;
711
712                 mutex_unlock(&root->fs_info->fs_mutex);
713                 mutex_unlock(&root->fs_info->trans_mutex);
714
715                 schedule_timeout(timeout);
716
717                 mutex_lock(&root->fs_info->fs_mutex);
718                 mutex_lock(&root->fs_info->trans_mutex);
719                 finish_wait(&cur_trans->writer_wait, &wait);
720                 ret = btrfs_write_ordered_inodes(trans, root);
721
722         } while (cur_trans->num_writers > 1 ||
723                  (cur_trans->num_joined != joined));
724
725         ret = create_pending_snapshots(trans, root->fs_info);
726         BUG_ON(ret);
727
728         WARN_ON(cur_trans != trans->transaction);
729
730         ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
731                               &dirty_fs_roots);
732         BUG_ON(ret);
733
734         ret = btrfs_commit_tree_roots(trans, root);
735         BUG_ON(ret);
736
737         cur_trans = root->fs_info->running_transaction;
738         spin_lock(&root->fs_info->new_trans_lock);
739         root->fs_info->running_transaction = NULL;
740         spin_unlock(&root->fs_info->new_trans_lock);
741         btrfs_set_super_generation(&root->fs_info->super_copy,
742                                    cur_trans->transid);
743         btrfs_set_super_root(&root->fs_info->super_copy,
744                              root->fs_info->tree_root->node->start);
745         btrfs_set_super_root_level(&root->fs_info->super_copy,
746                            btrfs_header_level(root->fs_info->tree_root->node));
747
748         btrfs_set_super_chunk_root(&root->fs_info->super_copy,
749                                    chunk_root->node->start);
750         btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
751                                          btrfs_header_level(chunk_root->node));
752         memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
753                sizeof(root->fs_info->super_copy));
754
755         btrfs_copy_pinned(root, pinned_copy);
756
757         mutex_unlock(&root->fs_info->trans_mutex);
758         mutex_unlock(&root->fs_info->fs_mutex);
759         ret = btrfs_write_and_wait_transaction(trans, root);
760         BUG_ON(ret);
761         write_ctree_super(trans, root);
762
763         mutex_lock(&root->fs_info->fs_mutex);
764         btrfs_finish_extent_commit(trans, root, pinned_copy);
765         mutex_lock(&root->fs_info->trans_mutex);
766
767         kfree(pinned_copy);
768
769         cur_trans->commit_done = 1;
770         root->fs_info->last_trans_committed = cur_trans->transid;
771         wake_up(&cur_trans->commit_wait);
772         put_transaction(cur_trans);
773         put_transaction(cur_trans);
774
775         if (root->fs_info->closing)
776                 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
777         else
778                 list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
779
780         mutex_unlock(&root->fs_info->trans_mutex);
781         kmem_cache_free(btrfs_trans_handle_cachep, trans);
782
783         if (root->fs_info->closing) {
784                 mutex_unlock(&root->fs_info->fs_mutex);
785                 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
786                 mutex_lock(&root->fs_info->fs_mutex);
787         }
788         return ret;
789 }
790
791 int btrfs_clean_old_snapshots(struct btrfs_root *root)
792 {
793         struct list_head dirty_roots;
794         INIT_LIST_HEAD(&dirty_roots);
795
796         mutex_lock(&root->fs_info->trans_mutex);
797         list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
798         mutex_unlock(&root->fs_info->trans_mutex);
799
800         if (!list_empty(&dirty_roots)) {
801                 drop_dirty_roots(root, &dirty_roots);
802         }
803         return 0;
804 }
805 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
806 void btrfs_transaction_cleaner(void *p)
807 #else
808 void btrfs_transaction_cleaner(struct work_struct *work)
809 #endif
810 {
811 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
812         struct btrfs_fs_info *fs_info = p;
813 #else
814         struct btrfs_fs_info *fs_info = container_of(work,
815                                                      struct btrfs_fs_info,
816                                                      trans_work.work);
817
818 #endif
819         struct btrfs_root *root = fs_info->tree_root;
820         struct btrfs_transaction *cur;
821         struct btrfs_trans_handle *trans;
822         unsigned long now;
823         unsigned long delay = HZ * 30;
824         int ret;
825
826         mutex_lock(&root->fs_info->fs_mutex);
827         if (root->fs_info->closing)
828                 goto out;
829
830         mutex_lock(&root->fs_info->trans_mutex);
831         cur = root->fs_info->running_transaction;
832         if (!cur) {
833                 mutex_unlock(&root->fs_info->trans_mutex);
834                 goto out;
835         }
836         now = get_seconds();
837         if (now < cur->start_time || now - cur->start_time < 30) {
838                 mutex_unlock(&root->fs_info->trans_mutex);
839                 delay = HZ * 5;
840                 goto out;
841         }
842         mutex_unlock(&root->fs_info->trans_mutex);
843         btrfs_defrag_dirty_roots(root->fs_info);
844         trans = btrfs_start_transaction(root, 1);
845         ret = btrfs_commit_transaction(trans, root);
846 out:
847         mutex_unlock(&root->fs_info->fs_mutex);
848         btrfs_clean_old_snapshots(root);
849         btrfs_transaction_queue_work(root, delay);
850 }
851
852 void btrfs_transaction_queue_work(struct btrfs_root *root, int delay)
853 {
854         if (!root->fs_info->closing)
855                 queue_delayed_work(trans_wq, &root->fs_info->trans_work, delay);
856 }
857
858 void btrfs_transaction_flush_work(struct btrfs_root *root)
859 {
860         cancel_delayed_work(&root->fs_info->trans_work);
861         flush_workqueue(trans_wq);
862 }
863
864 void __init btrfs_init_transaction_sys(void)
865 {
866         trans_wq = create_workqueue("btrfs-transaction");
867 }
868
869 void btrfs_exit_transaction_sys(void)
870 {
871         destroy_workqueue(trans_wq);
872 }
873