Btrfs: Find and remove dead roots the first time a root is loaded.
[safe/jmp/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 "ctree.h"
22 #include "disk-io.h"
23 #include "transaction.h"
24
25 static int total_trans = 0;
26 extern struct kmem_cache *btrfs_trans_handle_cachep;
27 extern struct kmem_cache *btrfs_transaction_cachep;
28
29 static struct workqueue_struct *trans_wq;
30
31 #define BTRFS_ROOT_TRANS_TAG 0
32 #define BTRFS_ROOT_DEFRAG_TAG 1
33
34 static void put_transaction(struct btrfs_transaction *transaction)
35 {
36         WARN_ON(transaction->use_count == 0);
37         transaction->use_count--;
38         if (transaction->use_count == 0) {
39                 WARN_ON(total_trans == 0);
40                 total_trans--;
41                 list_del_init(&transaction->list);
42                 memset(transaction, 0, sizeof(*transaction));
43                 kmem_cache_free(btrfs_transaction_cachep, transaction);
44         }
45 }
46
47 static int join_transaction(struct btrfs_root *root)
48 {
49         struct btrfs_transaction *cur_trans;
50         cur_trans = root->fs_info->running_transaction;
51         if (!cur_trans) {
52                 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
53                                              GFP_NOFS);
54                 total_trans++;
55                 BUG_ON(!cur_trans);
56                 root->fs_info->generation++;
57                 root->fs_info->running_transaction = cur_trans;
58                 cur_trans->num_writers = 1;
59                 cur_trans->num_joined = 0;
60                 cur_trans->transid = root->fs_info->generation;
61                 init_waitqueue_head(&cur_trans->writer_wait);
62                 init_waitqueue_head(&cur_trans->commit_wait);
63                 cur_trans->in_commit = 0;
64                 cur_trans->use_count = 1;
65                 cur_trans->commit_done = 0;
66                 cur_trans->start_time = get_seconds();
67                 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
68                 init_bit_radix(&cur_trans->dirty_pages);
69         } else {
70                 cur_trans->num_writers++;
71                 cur_trans->num_joined++;
72         }
73
74         return 0;
75 }
76
77 static int record_root_in_trans(struct btrfs_root *root)
78 {
79         u64 running_trans_id = root->fs_info->running_transaction->transid;
80         if (root->ref_cows && root->last_trans < running_trans_id) {
81                 WARN_ON(root == root->fs_info->extent_root);
82                 if (root->root_item.refs != 0) {
83                         radix_tree_tag_set(&root->fs_info->fs_roots_radix,
84                                    (unsigned long)root->root_key.objectid,
85                                    BTRFS_ROOT_TRANS_TAG);
86                         radix_tree_tag_set(&root->fs_info->fs_roots_radix,
87                                    (unsigned long)root->root_key.objectid,
88                                    BTRFS_ROOT_DEFRAG_TAG);
89                         root->commit_root = root->node;
90                         get_bh(root->node);
91                 } else {
92                         WARN_ON(1);
93                 }
94                 root->last_trans = running_trans_id;
95         }
96         return 0;
97 }
98
99 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
100                                                    int num_blocks)
101 {
102         struct btrfs_trans_handle *h =
103                 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
104         int ret;
105
106         mutex_lock(&root->fs_info->trans_mutex);
107         ret = join_transaction(root);
108         BUG_ON(ret);
109
110         record_root_in_trans(root);
111         h->transid = root->fs_info->running_transaction->transid;
112         h->transaction = root->fs_info->running_transaction;
113         h->blocks_reserved = num_blocks;
114         h->blocks_used = 0;
115         h->block_group = NULL;
116         h->alloc_exclude_nr = 0;
117         h->alloc_exclude_start = 0;
118         root->fs_info->running_transaction->use_count++;
119         mutex_unlock(&root->fs_info->trans_mutex);
120         return h;
121 }
122
123 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
124                           struct btrfs_root *root)
125 {
126         struct btrfs_transaction *cur_trans;
127
128         mutex_lock(&root->fs_info->trans_mutex);
129         cur_trans = root->fs_info->running_transaction;
130         WARN_ON(cur_trans != trans->transaction);
131         WARN_ON(cur_trans->num_writers < 1);
132         cur_trans->num_writers--;
133         if (waitqueue_active(&cur_trans->writer_wait))
134                 wake_up(&cur_trans->writer_wait);
135         put_transaction(cur_trans);
136         mutex_unlock(&root->fs_info->trans_mutex);
137         memset(trans, 0, sizeof(*trans));
138         kmem_cache_free(btrfs_trans_handle_cachep, trans);
139         return 0;
140 }
141
142
143 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
144                                      struct btrfs_root *root)
145 {
146         unsigned long gang[16];
147         int ret;
148         int i;
149         int err;
150         int werr = 0;
151         struct page *page;
152         struct radix_tree_root *dirty_pages;
153         struct inode *btree_inode = root->fs_info->btree_inode;
154
155         if (!trans || !trans->transaction) {
156                 return filemap_write_and_wait(btree_inode->i_mapping);
157         }
158         dirty_pages = &trans->transaction->dirty_pages;
159         while(1) {
160                 ret = find_first_radix_bit(dirty_pages, gang,
161                                            0, ARRAY_SIZE(gang));
162                 if (!ret)
163                         break;
164                 for (i = 0; i < ret; i++) {
165                         /* FIXME EIO */
166                         clear_radix_bit(dirty_pages, gang[i]);
167                         page = find_lock_page(btree_inode->i_mapping,
168                                               gang[i]);
169                         if (!page)
170                                 continue;
171                         if (PageWriteback(page)) {
172                                 if (PageDirty(page))
173                                         wait_on_page_writeback(page);
174                                 else {
175                                         unlock_page(page);
176                                         page_cache_release(page);
177                                         continue;
178                                 }
179                         }
180                         err = write_one_page(page, 0);
181                         if (err)
182                                 werr = err;
183                         page_cache_release(page);
184                 }
185         }
186         err = filemap_fdatawait(btree_inode->i_mapping);
187         if (err)
188                 werr = err;
189         return werr;
190 }
191
192 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
193                             struct btrfs_root *root)
194 {
195         int ret;
196         u64 old_extent_block;
197         struct btrfs_fs_info *fs_info = root->fs_info;
198         struct btrfs_root *tree_root = fs_info->tree_root;
199         struct btrfs_root *extent_root = fs_info->extent_root;
200
201         btrfs_write_dirty_block_groups(trans, extent_root);
202         while(1) {
203                 old_extent_block = btrfs_root_blocknr(&extent_root->root_item);
204                 if (old_extent_block == bh_blocknr(extent_root->node))
205                         break;
206                 btrfs_set_root_blocknr(&extent_root->root_item,
207                                        bh_blocknr(extent_root->node));
208                 ret = btrfs_update_root(trans, tree_root,
209                                         &extent_root->root_key,
210                                         &extent_root->root_item);
211                 BUG_ON(ret);
212                 btrfs_write_dirty_block_groups(trans, extent_root);
213         }
214         return 0;
215 }
216
217 static int wait_for_commit(struct btrfs_root *root,
218                            struct btrfs_transaction *commit)
219 {
220         DEFINE_WAIT(wait);
221         mutex_lock(&root->fs_info->trans_mutex);
222         while(!commit->commit_done) {
223                 prepare_to_wait(&commit->commit_wait, &wait,
224                                 TASK_UNINTERRUPTIBLE);
225                 if (commit->commit_done)
226                         break;
227                 mutex_unlock(&root->fs_info->trans_mutex);
228                 schedule();
229                 mutex_lock(&root->fs_info->trans_mutex);
230         }
231         mutex_unlock(&root->fs_info->trans_mutex);
232         finish_wait(&commit->commit_wait, &wait);
233         return 0;
234 }
235
236 struct dirty_root {
237         struct list_head list;
238         struct btrfs_root *root;
239         struct btrfs_root *latest_root;
240 };
241
242 int btrfs_add_dead_root(struct btrfs_root *root,
243                         struct btrfs_root *latest,
244                         struct list_head *dead_list)
245 {
246         struct dirty_root *dirty;
247
248         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
249         if (!dirty)
250                 return -ENOMEM;
251         dirty->root = root;
252         dirty->latest_root = latest;
253         list_add(&dirty->list, dead_list);
254         return 0;
255 }
256
257 static int add_dirty_roots(struct btrfs_trans_handle *trans,
258                            struct radix_tree_root *radix,
259                            struct list_head *list)
260 {
261         struct dirty_root *dirty;
262         struct btrfs_root *gang[8];
263         struct btrfs_root *root;
264         int i;
265         int ret;
266         int err = 0;
267         u32 refs;
268
269         while(1) {
270                 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
271                                                  ARRAY_SIZE(gang),
272                                                  BTRFS_ROOT_TRANS_TAG);
273                 if (ret == 0)
274                         break;
275                 for (i = 0; i < ret; i++) {
276                         root = gang[i];
277                         radix_tree_tag_clear(radix,
278                                      (unsigned long)root->root_key.objectid,
279                                      BTRFS_ROOT_TRANS_TAG);
280                         if (root->commit_root == root->node) {
281                                 WARN_ON(bh_blocknr(root->node) !=
282                                         btrfs_root_blocknr(&root->root_item));
283                                 brelse(root->commit_root);
284                                 root->commit_root = NULL;
285
286                                 /* make sure to update the root on disk
287                                  * so we get any updates to the block used
288                                  * counts
289                                  */
290                                 err = btrfs_update_root(trans,
291                                                 root->fs_info->tree_root,
292                                                 &root->root_key,
293                                                 &root->root_item);
294                                 continue;
295                         }
296                         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
297                         BUG_ON(!dirty);
298                         dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
299                         BUG_ON(!dirty->root);
300
301                         memset(&root->root_item.drop_progress, 0,
302                                sizeof(struct btrfs_disk_key));
303                         root->root_item.drop_level = 0;
304
305                         memcpy(dirty->root, root, sizeof(*root));
306                         dirty->root->node = root->commit_root;
307                         dirty->latest_root = root;
308                         root->commit_root = NULL;
309
310                         root->root_key.offset = root->fs_info->generation;
311                         btrfs_set_root_blocknr(&root->root_item,
312                                                bh_blocknr(root->node));
313                         err = btrfs_insert_root(trans, root->fs_info->tree_root,
314                                                 &root->root_key,
315                                                 &root->root_item);
316                         if (err)
317                                 break;
318
319                         refs = btrfs_root_refs(&dirty->root->root_item);
320                         btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
321                         err = btrfs_update_root(trans, root->fs_info->tree_root,
322                                                 &dirty->root->root_key,
323                                                 &dirty->root->root_item);
324
325                         BUG_ON(err);
326                         if (refs == 1) {
327                                 list_add(&dirty->list, list);
328                         } else {
329                                 WARN_ON(1);
330                                 kfree(dirty->root);
331                                 kfree(dirty);
332                         }
333                 }
334         }
335         return err;
336 }
337
338 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
339 {
340         struct btrfs_fs_info *info = root->fs_info;
341         int ret;
342         struct btrfs_trans_handle *trans;
343
344         if (root->defrag_running)
345                 return 0;
346
347         trans = btrfs_start_transaction(root, 1);
348         while (1) {
349                 root->defrag_running = 1;
350                 ret = btrfs_defrag_leaves(trans, root, cacheonly);
351                 btrfs_end_transaction(trans, root);
352                 mutex_unlock(&info->fs_mutex);
353
354                 btrfs_btree_balance_dirty(root);
355                 cond_resched();
356
357                 mutex_lock(&info->fs_mutex);
358                 trans = btrfs_start_transaction(root, 1);
359                 if (ret != -EAGAIN)
360                         break;
361         }
362         root->defrag_running = 0;
363         radix_tree_tag_clear(&info->fs_roots_radix,
364                      (unsigned long)root->root_key.objectid,
365                      BTRFS_ROOT_DEFRAG_TAG);
366         btrfs_end_transaction(trans, root);
367         return 0;
368 }
369
370 int btrfs_defrag_dirty_roots(struct btrfs_fs_info *info)
371 {
372         struct btrfs_root *gang[1];
373         struct btrfs_root *root;
374         int i;
375         int ret;
376         int err = 0;
377         u64 last = 0;
378
379         while(1) {
380                 ret = radix_tree_gang_lookup_tag(&info->fs_roots_radix,
381                                                  (void **)gang, last,
382                                                  ARRAY_SIZE(gang),
383                                                  BTRFS_ROOT_DEFRAG_TAG);
384                 if (ret == 0)
385                         break;
386                 for (i = 0; i < ret; i++) {
387                         root = gang[i];
388                         last = root->root_key.objectid + 1;
389                         btrfs_defrag_root(root, 1);
390                 }
391         }
392         btrfs_defrag_root(info->extent_root, 1);
393         return err;
394 }
395
396 static int drop_dirty_roots(struct btrfs_root *tree_root,
397                             struct list_head *list)
398 {
399         struct dirty_root *dirty;
400         struct btrfs_trans_handle *trans;
401         u64 num_blocks;
402         u64 blocks_used;
403         int ret = 0;
404         int err;
405
406         while(!list_empty(list)) {
407                 struct btrfs_root *root;
408
409                 mutex_lock(&tree_root->fs_info->fs_mutex);
410                 dirty = list_entry(list->next, struct dirty_root, list);
411                 list_del_init(&dirty->list);
412
413                 num_blocks = btrfs_root_blocks_used(&dirty->root->root_item);
414                 root = dirty->latest_root;
415
416                 while(1) {
417                         trans = btrfs_start_transaction(tree_root, 1);
418                         ret = btrfs_drop_snapshot(trans, dirty->root);
419                         if (ret != -EAGAIN) {
420                                 break;
421                         }
422
423                         err = btrfs_update_root(trans,
424                                         tree_root,
425                                         &dirty->root->root_key,
426                                         &dirty->root->root_item);
427                         if (err)
428                                 ret = err;
429                         ret = btrfs_end_transaction(trans, tree_root);
430                         BUG_ON(ret);
431                         mutex_unlock(&tree_root->fs_info->fs_mutex);
432
433                         btrfs_btree_balance_dirty(tree_root);
434                         schedule();
435
436                         mutex_lock(&tree_root->fs_info->fs_mutex);
437                 }
438                 BUG_ON(ret);
439
440                 num_blocks -= btrfs_root_blocks_used(&dirty->root->root_item);
441                 blocks_used = btrfs_root_blocks_used(&root->root_item);
442                 if (num_blocks) {
443                         record_root_in_trans(root);
444                         btrfs_set_root_blocks_used(&root->root_item,
445                                                    blocks_used - num_blocks);
446                 }
447                 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
448                 if (ret) {
449                         BUG();
450                         break;
451                 }
452                 ret = btrfs_end_transaction(trans, tree_root);
453                 BUG_ON(ret);
454
455                 kfree(dirty->root);
456                 kfree(dirty);
457                 mutex_unlock(&tree_root->fs_info->fs_mutex);
458                 btrfs_btree_balance_dirty(tree_root);
459                 schedule();
460         }
461         return ret;
462 }
463
464 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
465                              struct btrfs_root *root)
466 {
467         unsigned long joined = 0;
468         unsigned long timeout = 1;
469         struct btrfs_transaction *cur_trans;
470         struct btrfs_transaction *prev_trans = NULL;
471         struct list_head dirty_fs_roots;
472         struct radix_tree_root pinned_copy;
473         DEFINE_WAIT(wait);
474         int ret;
475
476         init_bit_radix(&pinned_copy);
477         INIT_LIST_HEAD(&dirty_fs_roots);
478
479         mutex_lock(&root->fs_info->trans_mutex);
480         if (trans->transaction->in_commit) {
481                 cur_trans = trans->transaction;
482                 trans->transaction->use_count++;
483                 mutex_unlock(&root->fs_info->trans_mutex);
484                 btrfs_end_transaction(trans, root);
485
486                 mutex_unlock(&root->fs_info->fs_mutex);
487                 ret = wait_for_commit(root, cur_trans);
488                 BUG_ON(ret);
489
490                 mutex_lock(&root->fs_info->trans_mutex);
491                 put_transaction(cur_trans);
492                 mutex_unlock(&root->fs_info->trans_mutex);
493
494                 mutex_lock(&root->fs_info->fs_mutex);
495                 return 0;
496         }
497         trans->transaction->in_commit = 1;
498         cur_trans = trans->transaction;
499         if (cur_trans->list.prev != &root->fs_info->trans_list) {
500                 prev_trans = list_entry(cur_trans->list.prev,
501                                         struct btrfs_transaction, list);
502                 if (!prev_trans->commit_done) {
503                         prev_trans->use_count++;
504                         mutex_unlock(&root->fs_info->fs_mutex);
505                         mutex_unlock(&root->fs_info->trans_mutex);
506
507                         wait_for_commit(root, prev_trans);
508
509                         mutex_lock(&root->fs_info->fs_mutex);
510                         mutex_lock(&root->fs_info->trans_mutex);
511                         put_transaction(prev_trans);
512                 }
513         }
514
515         do {
516                 joined = cur_trans->num_joined;
517                 WARN_ON(cur_trans != trans->transaction);
518                 prepare_to_wait(&cur_trans->writer_wait, &wait,
519                                 TASK_UNINTERRUPTIBLE);
520
521                 if (cur_trans->num_writers > 1)
522                         timeout = MAX_SCHEDULE_TIMEOUT;
523                 else
524                         timeout = 1;
525
526                 mutex_unlock(&root->fs_info->fs_mutex);
527                 mutex_unlock(&root->fs_info->trans_mutex);
528
529                 schedule_timeout(timeout);
530
531                 mutex_lock(&root->fs_info->fs_mutex);
532                 mutex_lock(&root->fs_info->trans_mutex);
533                 finish_wait(&cur_trans->writer_wait, &wait);
534         } while (cur_trans->num_writers > 1 ||
535                  (cur_trans->num_joined != joined));
536
537         WARN_ON(cur_trans != trans->transaction);
538         ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
539                               &dirty_fs_roots);
540         BUG_ON(ret);
541
542         ret = btrfs_commit_tree_roots(trans, root);
543         BUG_ON(ret);
544
545         cur_trans = root->fs_info->running_transaction;
546         root->fs_info->running_transaction = NULL;
547         btrfs_set_super_generation(&root->fs_info->super_copy,
548                                    cur_trans->transid);
549         btrfs_set_super_root(&root->fs_info->super_copy,
550                              bh_blocknr(root->fs_info->tree_root->node));
551         memcpy(root->fs_info->disk_super, &root->fs_info->super_copy,
552                sizeof(root->fs_info->super_copy));
553
554         btrfs_copy_pinned(root, &pinned_copy);
555
556         mutex_unlock(&root->fs_info->trans_mutex);
557         mutex_unlock(&root->fs_info->fs_mutex);
558         ret = btrfs_write_and_wait_transaction(trans, root);
559         BUG_ON(ret);
560         write_ctree_super(trans, root);
561         mutex_lock(&root->fs_info->fs_mutex);
562         btrfs_finish_extent_commit(trans, root, &pinned_copy);
563         mutex_lock(&root->fs_info->trans_mutex);
564         cur_trans->commit_done = 1;
565         root->fs_info->last_trans_committed = cur_trans->transid;
566         wake_up(&cur_trans->commit_wait);
567         put_transaction(cur_trans);
568         put_transaction(cur_trans);
569
570         if (root->fs_info->closing)
571                 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
572         else
573                 list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
574
575         mutex_unlock(&root->fs_info->trans_mutex);
576         kmem_cache_free(btrfs_trans_handle_cachep, trans);
577
578         if (root->fs_info->closing) {
579                 mutex_unlock(&root->fs_info->fs_mutex);
580                 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
581                 mutex_lock(&root->fs_info->fs_mutex);
582         }
583         return ret;
584 }
585
586 int btrfs_clean_old_snapshots(struct btrfs_root *root)
587 {
588         struct list_head dirty_roots;
589         INIT_LIST_HEAD(&dirty_roots);
590
591         mutex_lock(&root->fs_info->trans_mutex);
592         list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
593         mutex_unlock(&root->fs_info->trans_mutex);
594
595         if (!list_empty(&dirty_roots)) {
596                 drop_dirty_roots(root, &dirty_roots);
597         }
598         return 0;
599 }
600 void btrfs_transaction_cleaner(struct work_struct *work)
601 {
602         struct btrfs_fs_info *fs_info = container_of(work,
603                                                      struct btrfs_fs_info,
604                                                      trans_work.work);
605
606         struct btrfs_root *root = fs_info->tree_root;
607         struct btrfs_transaction *cur;
608         struct btrfs_trans_handle *trans;
609         unsigned long now;
610         unsigned long delay = HZ * 30;
611         int ret;
612
613         mutex_lock(&root->fs_info->fs_mutex);
614         mutex_lock(&root->fs_info->trans_mutex);
615         cur = root->fs_info->running_transaction;
616         if (!cur) {
617                 mutex_unlock(&root->fs_info->trans_mutex);
618                 goto out;
619         }
620         now = get_seconds();
621         if (now < cur->start_time || now - cur->start_time < 30) {
622                 mutex_unlock(&root->fs_info->trans_mutex);
623                 delay = HZ * 5;
624                 goto out;
625         }
626         mutex_unlock(&root->fs_info->trans_mutex);
627         btrfs_defrag_dirty_roots(root->fs_info);
628         trans = btrfs_start_transaction(root, 1);
629         ret = btrfs_commit_transaction(trans, root);
630 out:
631         mutex_unlock(&root->fs_info->fs_mutex);
632         btrfs_clean_old_snapshots(root);
633         btrfs_transaction_queue_work(root, delay);
634 }
635
636 void btrfs_transaction_queue_work(struct btrfs_root *root, int delay)
637 {
638         queue_delayed_work(trans_wq, &root->fs_info->trans_work, delay);
639 }
640
641 void btrfs_transaction_flush_work(struct btrfs_root *root)
642 {
643         cancel_rearming_delayed_workqueue(trans_wq, &root->fs_info->trans_work);
644         flush_workqueue(trans_wq);
645 }
646
647 void __init btrfs_init_transaction_sys(void)
648 {
649         trans_wq = create_workqueue("btrfs");
650 }
651
652 void __exit btrfs_exit_transaction_sys(void)
653 {
654         destroy_workqueue(trans_wq);
655 }
656