Btrfs: Add per-root block accounting and sysfs entries
[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, struct list_head *dead_list)
243 {
244         struct dirty_root *dirty;
245
246         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
247         if (!dirty)
248                 return -ENOMEM;
249         dirty->root = root;
250         list_add(&dirty->list, dead_list);
251         return 0;
252 }
253
254 static int add_dirty_roots(struct btrfs_trans_handle *trans,
255                            struct radix_tree_root *radix,
256                            struct list_head *list)
257 {
258         struct dirty_root *dirty;
259         struct btrfs_root *gang[8];
260         struct btrfs_root *root;
261         int i;
262         int ret;
263         int err = 0;
264         u32 refs;
265
266         while(1) {
267                 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
268                                                  ARRAY_SIZE(gang),
269                                                  BTRFS_ROOT_TRANS_TAG);
270                 if (ret == 0)
271                         break;
272                 for (i = 0; i < ret; i++) {
273                         root = gang[i];
274                         radix_tree_tag_clear(radix,
275                                      (unsigned long)root->root_key.objectid,
276                                      BTRFS_ROOT_TRANS_TAG);
277                         if (root->commit_root == root->node) {
278                                 WARN_ON(bh_blocknr(root->node) !=
279                                         btrfs_root_blocknr(&root->root_item));
280                                 brelse(root->commit_root);
281                                 root->commit_root = NULL;
282
283                                 /* make sure to update the root on disk
284                                  * so we get any updates to the block used
285                                  * counts
286                                  */
287                                 err = btrfs_update_root(trans,
288                                                 root->fs_info->tree_root,
289                                                 &root->root_key,
290                                                 &root->root_item);
291                                 continue;
292                         }
293                         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
294                         BUG_ON(!dirty);
295                         dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
296                         BUG_ON(!dirty->root);
297
298                         memset(&root->root_item.drop_progress, 0,
299                                sizeof(struct btrfs_disk_key));
300                         root->root_item.drop_level = 0;
301
302                         memcpy(dirty->root, root, sizeof(*root));
303                         dirty->root->node = root->commit_root;
304                         dirty->latest_root = root;
305                         root->commit_root = NULL;
306
307                         root->root_key.offset = root->fs_info->generation;
308                         btrfs_set_root_blocknr(&root->root_item,
309                                                bh_blocknr(root->node));
310                         err = btrfs_insert_root(trans, root->fs_info->tree_root,
311                                                 &root->root_key,
312                                                 &root->root_item);
313                         if (err)
314                                 break;
315
316                         refs = btrfs_root_refs(&dirty->root->root_item);
317                         btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
318                         err = btrfs_update_root(trans, root->fs_info->tree_root,
319                                                 &dirty->root->root_key,
320                                                 &dirty->root->root_item);
321
322                         BUG_ON(err);
323                         if (refs == 1) {
324                                 list_add(&dirty->list, list);
325                         } else {
326                                 WARN_ON(1);
327                                 kfree(dirty->root);
328                                 kfree(dirty);
329                         }
330                 }
331         }
332         return err;
333 }
334
335 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
336 {
337         struct btrfs_fs_info *info = root->fs_info;
338         int ret;
339         struct btrfs_trans_handle *trans;
340
341         if (root->defrag_running)
342                 return 0;
343
344         trans = btrfs_start_transaction(root, 1);
345         while (1) {
346                 root->defrag_running = 1;
347                 ret = btrfs_defrag_leaves(trans, root, cacheonly);
348                 btrfs_end_transaction(trans, root);
349                 mutex_unlock(&info->fs_mutex);
350
351                 btrfs_btree_balance_dirty(root);
352                 cond_resched();
353
354                 mutex_lock(&info->fs_mutex);
355                 trans = btrfs_start_transaction(root, 1);
356                 if (ret != -EAGAIN)
357                         break;
358         }
359         root->defrag_running = 0;
360         radix_tree_tag_clear(&info->fs_roots_radix,
361                      (unsigned long)root->root_key.objectid,
362                      BTRFS_ROOT_DEFRAG_TAG);
363         btrfs_end_transaction(trans, root);
364         return 0;
365 }
366
367 int btrfs_defrag_dirty_roots(struct btrfs_fs_info *info)
368 {
369         struct btrfs_root *gang[1];
370         struct btrfs_root *root;
371         int i;
372         int ret;
373         int err = 0;
374         u64 last = 0;
375
376         while(1) {
377                 ret = radix_tree_gang_lookup_tag(&info->fs_roots_radix,
378                                                  (void **)gang, last,
379                                                  ARRAY_SIZE(gang),
380                                                  BTRFS_ROOT_DEFRAG_TAG);
381                 if (ret == 0)
382                         break;
383                 for (i = 0; i < ret; i++) {
384                         root = gang[i];
385                         last = root->root_key.objectid + 1;
386                         btrfs_defrag_root(root, 1);
387                 }
388         }
389         btrfs_defrag_root(info->extent_root, 1);
390         return err;
391 }
392
393 static int drop_dirty_roots(struct btrfs_root *tree_root,
394                             struct list_head *list)
395 {
396         struct dirty_root *dirty;
397         struct btrfs_trans_handle *trans;
398         u64 num_blocks;
399         u64 blocks_used;
400         int ret = 0;
401         int err;
402
403         while(!list_empty(list)) {
404                 struct btrfs_root *root;
405
406                 mutex_lock(&tree_root->fs_info->fs_mutex);
407                 dirty = list_entry(list->next, struct dirty_root, list);
408                 list_del_init(&dirty->list);
409
410                 num_blocks = btrfs_root_blocks_used(&dirty->root->root_item);
411                 root = dirty->latest_root;
412
413                 while(1) {
414                         trans = btrfs_start_transaction(tree_root, 1);
415
416                         ret = btrfs_drop_snapshot(trans, dirty->root);
417                         if (ret != -EAGAIN) {
418                                 break;
419                         }
420
421                         err = btrfs_update_root(trans,
422                                         tree_root,
423                                         &dirty->root->root_key,
424                                         &dirty->root->root_item);
425                         if (err)
426                                 ret = err;
427                         ret = btrfs_end_transaction(trans, tree_root);
428                         BUG_ON(ret);
429                         mutex_unlock(&tree_root->fs_info->fs_mutex);
430
431                         btrfs_btree_balance_dirty(tree_root);
432                         schedule();
433
434                         mutex_lock(&tree_root->fs_info->fs_mutex);
435                 }
436                 BUG_ON(ret);
437
438                 num_blocks -= btrfs_root_blocks_used(&dirty->root->root_item);
439                 blocks_used = btrfs_root_blocks_used(&root->root_item);
440                 if (num_blocks) {
441                         record_root_in_trans(root);
442                         btrfs_set_root_blocks_used(&root->root_item,
443                                                    blocks_used - num_blocks);
444                 }
445                 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
446                 if (ret) {
447                         BUG();
448                         break;
449                 }
450                 ret = btrfs_end_transaction(trans, tree_root);
451                 BUG_ON(ret);
452
453                 kfree(dirty->root);
454                 kfree(dirty);
455                 mutex_unlock(&tree_root->fs_info->fs_mutex);
456                 btrfs_btree_balance_dirty(tree_root);
457                 schedule();
458         }
459         return ret;
460 }
461
462 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
463                              struct btrfs_root *root)
464 {
465         unsigned long joined = 0;
466         unsigned long timeout = 1;
467         struct btrfs_transaction *cur_trans;
468         struct btrfs_transaction *prev_trans = NULL;
469         struct list_head dirty_fs_roots;
470         struct radix_tree_root pinned_copy;
471         DEFINE_WAIT(wait);
472         int ret;
473
474         init_bit_radix(&pinned_copy);
475         INIT_LIST_HEAD(&dirty_fs_roots);
476
477         mutex_lock(&root->fs_info->trans_mutex);
478         if (trans->transaction->in_commit) {
479                 cur_trans = trans->transaction;
480                 trans->transaction->use_count++;
481                 mutex_unlock(&root->fs_info->trans_mutex);
482                 btrfs_end_transaction(trans, root);
483
484                 mutex_unlock(&root->fs_info->fs_mutex);
485                 ret = wait_for_commit(root, cur_trans);
486                 BUG_ON(ret);
487
488                 mutex_lock(&root->fs_info->trans_mutex);
489                 put_transaction(cur_trans);
490                 mutex_unlock(&root->fs_info->trans_mutex);
491
492                 mutex_lock(&root->fs_info->fs_mutex);
493                 return 0;
494         }
495         trans->transaction->in_commit = 1;
496         cur_trans = trans->transaction;
497         if (cur_trans->list.prev != &root->fs_info->trans_list) {
498                 prev_trans = list_entry(cur_trans->list.prev,
499                                         struct btrfs_transaction, list);
500                 if (!prev_trans->commit_done) {
501                         prev_trans->use_count++;
502                         mutex_unlock(&root->fs_info->fs_mutex);
503                         mutex_unlock(&root->fs_info->trans_mutex);
504
505                         wait_for_commit(root, prev_trans);
506
507                         mutex_lock(&root->fs_info->fs_mutex);
508                         mutex_lock(&root->fs_info->trans_mutex);
509                         put_transaction(prev_trans);
510                 }
511         }
512
513         do {
514                 joined = cur_trans->num_joined;
515                 WARN_ON(cur_trans != trans->transaction);
516                 prepare_to_wait(&cur_trans->writer_wait, &wait,
517                                 TASK_UNINTERRUPTIBLE);
518
519                 if (cur_trans->num_writers > 1)
520                         timeout = MAX_SCHEDULE_TIMEOUT;
521                 else
522                         timeout = 1;
523
524                 mutex_unlock(&root->fs_info->fs_mutex);
525                 mutex_unlock(&root->fs_info->trans_mutex);
526
527                 schedule_timeout(timeout);
528
529                 mutex_lock(&root->fs_info->fs_mutex);
530                 mutex_lock(&root->fs_info->trans_mutex);
531                 finish_wait(&cur_trans->writer_wait, &wait);
532         } while (cur_trans->num_writers > 1 ||
533                  (cur_trans->num_joined != joined));
534
535         WARN_ON(cur_trans != trans->transaction);
536         ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
537                               &dirty_fs_roots);
538         BUG_ON(ret);
539
540         ret = btrfs_commit_tree_roots(trans, root);
541         BUG_ON(ret);
542
543         cur_trans = root->fs_info->running_transaction;
544         root->fs_info->running_transaction = NULL;
545         btrfs_set_super_generation(&root->fs_info->super_copy,
546                                    cur_trans->transid);
547         btrfs_set_super_root(&root->fs_info->super_copy,
548                              bh_blocknr(root->fs_info->tree_root->node));
549         memcpy(root->fs_info->disk_super, &root->fs_info->super_copy,
550                sizeof(root->fs_info->super_copy));
551
552         btrfs_copy_pinned(root, &pinned_copy);
553
554         mutex_unlock(&root->fs_info->trans_mutex);
555         mutex_unlock(&root->fs_info->fs_mutex);
556         ret = btrfs_write_and_wait_transaction(trans, root);
557         BUG_ON(ret);
558         write_ctree_super(trans, root);
559         mutex_lock(&root->fs_info->fs_mutex);
560         btrfs_finish_extent_commit(trans, root, &pinned_copy);
561         mutex_lock(&root->fs_info->trans_mutex);
562         cur_trans->commit_done = 1;
563         root->fs_info->last_trans_committed = cur_trans->transid;
564         wake_up(&cur_trans->commit_wait);
565         put_transaction(cur_trans);
566         put_transaction(cur_trans);
567
568         if (root->fs_info->closing)
569                 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
570         else
571                 list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
572
573         mutex_unlock(&root->fs_info->trans_mutex);
574         kmem_cache_free(btrfs_trans_handle_cachep, trans);
575
576         if (root->fs_info->closing) {
577                 mutex_unlock(&root->fs_info->fs_mutex);
578                 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
579                 mutex_lock(&root->fs_info->fs_mutex);
580         }
581         return ret;
582 }
583
584 int btrfs_clean_old_snapshots(struct btrfs_root *root)
585 {
586         struct list_head dirty_roots;
587         INIT_LIST_HEAD(&dirty_roots);
588
589         mutex_lock(&root->fs_info->trans_mutex);
590         list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
591         mutex_unlock(&root->fs_info->trans_mutex);
592
593         if (!list_empty(&dirty_roots)) {
594                 drop_dirty_roots(root, &dirty_roots);
595         }
596         return 0;
597 }
598 void btrfs_transaction_cleaner(struct work_struct *work)
599 {
600         struct btrfs_fs_info *fs_info = container_of(work,
601                                                      struct btrfs_fs_info,
602                                                      trans_work.work);
603
604         struct btrfs_root *root = fs_info->tree_root;
605         struct btrfs_transaction *cur;
606         struct btrfs_trans_handle *trans;
607         unsigned long now;
608         unsigned long delay = HZ * 30;
609         int ret;
610
611         mutex_lock(&root->fs_info->fs_mutex);
612         mutex_lock(&root->fs_info->trans_mutex);
613         cur = root->fs_info->running_transaction;
614         if (!cur) {
615                 mutex_unlock(&root->fs_info->trans_mutex);
616                 goto out;
617         }
618         now = get_seconds();
619         if (now < cur->start_time || now - cur->start_time < 30) {
620                 mutex_unlock(&root->fs_info->trans_mutex);
621                 delay = HZ * 5;
622                 goto out;
623         }
624         mutex_unlock(&root->fs_info->trans_mutex);
625         btrfs_defrag_dirty_roots(root->fs_info);
626         trans = btrfs_start_transaction(root, 1);
627         ret = btrfs_commit_transaction(trans, root);
628 out:
629         mutex_unlock(&root->fs_info->fs_mutex);
630         btrfs_clean_old_snapshots(root);
631         btrfs_transaction_queue_work(root, delay);
632 }
633
634 void btrfs_transaction_queue_work(struct btrfs_root *root, int delay)
635 {
636         queue_delayed_work(trans_wq, &root->fs_info->trans_work, delay);
637 }
638
639 void btrfs_transaction_flush_work(struct btrfs_root *root)
640 {
641         cancel_rearming_delayed_workqueue(trans_wq, &root->fs_info->trans_work);
642         flush_workqueue(trans_wq);
643 }
644
645 void __init btrfs_init_transaction_sys(void)
646 {
647         trans_wq = create_workqueue("btrfs");
648 }
649
650 void __exit btrfs_exit_transaction_sys(void)
651 {
652         destroy_workqueue(trans_wq);
653 }
654