Btrfs: Fix extent_buffer and extent_state leaks
[deliverable/linux.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
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 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
70 extent_map_tree_init(&cur_trans->dirty_pages,
71 root->fs_info->btree_inode->i_mapping,
72 GFP_NOFS);
73 } else {
74 cur_trans->num_writers++;
75 cur_trans->num_joined++;
76 }
77
78 return 0;
79 }
80
81 static int record_root_in_trans(struct btrfs_root *root)
82 {
83 u64 running_trans_id = root->fs_info->running_transaction->transid;
84 if (root->ref_cows && root->last_trans < running_trans_id) {
85 WARN_ON(root == root->fs_info->extent_root);
86 if (root->root_item.refs != 0) {
87 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
88 (unsigned long)root->root_key.objectid,
89 BTRFS_ROOT_TRANS_TAG);
90 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
91 (unsigned long)root->root_key.objectid,
92 BTRFS_ROOT_DEFRAG_TAG);
93 root->commit_root = root->node;
94 extent_buffer_get(root->node);
95 } else {
96 WARN_ON(1);
97 }
98 root->last_trans = running_trans_id;
99 }
100 return 0;
101 }
102
103 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
104 int num_blocks)
105 {
106 struct btrfs_trans_handle *h =
107 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
108 int ret;
109
110 mutex_lock(&root->fs_info->trans_mutex);
111 ret = join_transaction(root);
112 BUG_ON(ret);
113
114 record_root_in_trans(root);
115 h->transid = root->fs_info->running_transaction->transid;
116 h->transaction = root->fs_info->running_transaction;
117 h->blocks_reserved = num_blocks;
118 h->blocks_used = 0;
119 h->block_group = NULL;
120 h->alloc_exclude_nr = 0;
121 h->alloc_exclude_start = 0;
122 root->fs_info->running_transaction->use_count++;
123 mutex_unlock(&root->fs_info->trans_mutex);
124 return h;
125 }
126
127 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
128 struct btrfs_root *root)
129 {
130 struct btrfs_transaction *cur_trans;
131
132 mutex_lock(&root->fs_info->trans_mutex);
133 cur_trans = root->fs_info->running_transaction;
134 WARN_ON(cur_trans != trans->transaction);
135 WARN_ON(cur_trans->num_writers < 1);
136 cur_trans->num_writers--;
137 if (waitqueue_active(&cur_trans->writer_wait))
138 wake_up(&cur_trans->writer_wait);
139 put_transaction(cur_trans);
140 mutex_unlock(&root->fs_info->trans_mutex);
141 memset(trans, 0, sizeof(*trans));
142 kmem_cache_free(btrfs_trans_handle_cachep, trans);
143 return 0;
144 }
145
146
147 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
148 struct btrfs_root *root)
149 {
150 int ret;
151 int err;
152 int werr = 0;
153 struct extent_map_tree *dirty_pages;
154 struct page *page;
155 struct inode *btree_inode = root->fs_info->btree_inode;
156 u64 start;
157 u64 end;
158 unsigned long index;
159
160 if (!trans || !trans->transaction) {
161 return filemap_write_and_wait(btree_inode->i_mapping);
162 }
163 dirty_pages = &trans->transaction->dirty_pages;
164 while(1) {
165 ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
166 EXTENT_DIRTY);
167 if (ret)
168 break;
169 clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
170 while(start <= end) {
171 index = start >> PAGE_CACHE_SHIFT;
172 start = (index + 1) << PAGE_CACHE_SHIFT;
173 page = find_lock_page(btree_inode->i_mapping, index);
174 if (!page)
175 continue;
176 if (PageWriteback(page)) {
177 if (PageDirty(page))
178 wait_on_page_writeback(page);
179 else {
180 unlock_page(page);
181 page_cache_release(page);
182 continue;
183 }
184 }
185 err = write_one_page(page, 0);
186 if (err)
187 werr = err;
188 page_cache_release(page);
189 }
190 }
191 err = filemap_fdatawait(btree_inode->i_mapping);
192 if (err)
193 werr = err;
194 return werr;
195 }
196
197 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
198 struct btrfs_root *root)
199 {
200 int ret;
201 u64 old_extent_block;
202 struct btrfs_fs_info *fs_info = root->fs_info;
203 struct btrfs_root *tree_root = fs_info->tree_root;
204 struct btrfs_root *extent_root = fs_info->extent_root;
205
206 btrfs_write_dirty_block_groups(trans, extent_root);
207 while(1) {
208 old_extent_block = btrfs_root_blocknr(&extent_root->root_item);
209 if (old_extent_block ==
210 extent_buffer_blocknr(extent_root->node))
211 break;
212 btrfs_set_root_blocknr(&extent_root->root_item,
213 extent_buffer_blocknr(extent_root->node));
214 ret = btrfs_update_root(trans, tree_root,
215 &extent_root->root_key,
216 &extent_root->root_item);
217 BUG_ON(ret);
218 btrfs_write_dirty_block_groups(trans, extent_root);
219 }
220 return 0;
221 }
222
223 static int wait_for_commit(struct btrfs_root *root,
224 struct btrfs_transaction *commit)
225 {
226 DEFINE_WAIT(wait);
227 mutex_lock(&root->fs_info->trans_mutex);
228 while(!commit->commit_done) {
229 prepare_to_wait(&commit->commit_wait, &wait,
230 TASK_UNINTERRUPTIBLE);
231 if (commit->commit_done)
232 break;
233 mutex_unlock(&root->fs_info->trans_mutex);
234 schedule();
235 mutex_lock(&root->fs_info->trans_mutex);
236 }
237 mutex_unlock(&root->fs_info->trans_mutex);
238 finish_wait(&commit->commit_wait, &wait);
239 return 0;
240 }
241
242 struct dirty_root {
243 struct list_head list;
244 struct btrfs_root *root;
245 struct btrfs_root *latest_root;
246 };
247
248 int btrfs_add_dead_root(struct btrfs_root *root,
249 struct btrfs_root *latest,
250 struct list_head *dead_list)
251 {
252 struct dirty_root *dirty;
253
254 dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
255 if (!dirty)
256 return -ENOMEM;
257 dirty->root = root;
258 dirty->latest_root = latest;
259 list_add(&dirty->list, dead_list);
260 return 0;
261 }
262
263 static int add_dirty_roots(struct btrfs_trans_handle *trans,
264 struct radix_tree_root *radix,
265 struct list_head *list)
266 {
267 struct dirty_root *dirty;
268 struct btrfs_root *gang[8];
269 struct btrfs_root *root;
270 int i;
271 int ret;
272 int err = 0;
273 u32 refs;
274
275 while(1) {
276 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
277 ARRAY_SIZE(gang),
278 BTRFS_ROOT_TRANS_TAG);
279 if (ret == 0)
280 break;
281 for (i = 0; i < ret; i++) {
282 root = gang[i];
283 radix_tree_tag_clear(radix,
284 (unsigned long)root->root_key.objectid,
285 BTRFS_ROOT_TRANS_TAG);
286 if (root->commit_root == root->node) {
287 WARN_ON(extent_buffer_blocknr(root->node) !=
288 btrfs_root_blocknr(&root->root_item));
289 free_extent_buffer(root->commit_root);
290 root->commit_root = NULL;
291
292 /* make sure to update the root on disk
293 * so we get any updates to the block used
294 * counts
295 */
296 err = btrfs_update_root(trans,
297 root->fs_info->tree_root,
298 &root->root_key,
299 &root->root_item);
300 continue;
301 }
302 dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
303 BUG_ON(!dirty);
304 dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
305 BUG_ON(!dirty->root);
306
307 memset(&root->root_item.drop_progress, 0,
308 sizeof(struct btrfs_disk_key));
309 root->root_item.drop_level = 0;
310
311 memcpy(dirty->root, root, sizeof(*root));
312 dirty->root->node = root->commit_root;
313 dirty->latest_root = root;
314 root->commit_root = NULL;
315
316 root->root_key.offset = root->fs_info->generation;
317 btrfs_set_root_blocknr(&root->root_item,
318 extent_buffer_blocknr(root->node));
319 err = btrfs_insert_root(trans, root->fs_info->tree_root,
320 &root->root_key,
321 &root->root_item);
322 if (err)
323 break;
324
325 refs = btrfs_root_refs(&dirty->root->root_item);
326 btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
327 err = btrfs_update_root(trans, root->fs_info->tree_root,
328 &dirty->root->root_key,
329 &dirty->root->root_item);
330
331 BUG_ON(err);
332 if (refs == 1) {
333 list_add(&dirty->list, list);
334 } else {
335 WARN_ON(1);
336 kfree(dirty->root);
337 kfree(dirty);
338 }
339 }
340 }
341 return err;
342 }
343
344 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
345 {
346 struct btrfs_fs_info *info = root->fs_info;
347 int ret;
348 struct btrfs_trans_handle *trans;
349 unsigned long nr;
350
351 if (root->defrag_running)
352 return 0;
353
354 trans = btrfs_start_transaction(root, 1);
355 while (0) {
356 root->defrag_running = 1;
357 ret = btrfs_defrag_leaves(trans, root, cacheonly);
358 nr = trans->blocks_used;
359 btrfs_end_transaction(trans, root);
360 mutex_unlock(&info->fs_mutex);
361
362 btrfs_btree_balance_dirty(info->tree_root, nr);
363 cond_resched();
364
365 mutex_lock(&info->fs_mutex);
366 trans = btrfs_start_transaction(root, 1);
367 if (ret != -EAGAIN)
368 break;
369 }
370 root->defrag_running = 0;
371 radix_tree_tag_clear(&info->fs_roots_radix,
372 (unsigned long)root->root_key.objectid,
373 BTRFS_ROOT_DEFRAG_TAG);
374 btrfs_end_transaction(trans, root);
375 return 0;
376 }
377
378 int btrfs_defrag_dirty_roots(struct btrfs_fs_info *info)
379 {
380 struct btrfs_root *gang[1];
381 struct btrfs_root *root;
382 int i;
383 int ret;
384 int err = 0;
385 u64 last = 0;
386
387 while(1) {
388 ret = radix_tree_gang_lookup_tag(&info->fs_roots_radix,
389 (void **)gang, last,
390 ARRAY_SIZE(gang),
391 BTRFS_ROOT_DEFRAG_TAG);
392 if (ret == 0)
393 break;
394 for (i = 0; i < ret; i++) {
395 root = gang[i];
396 last = root->root_key.objectid + 1;
397 btrfs_defrag_root(root, 1);
398 }
399 }
400 // btrfs_defrag_root(info->extent_root, 1);
401 return err;
402 }
403
404 static int drop_dirty_roots(struct btrfs_root *tree_root,
405 struct list_head *list)
406 {
407 struct dirty_root *dirty;
408 struct btrfs_trans_handle *trans;
409 unsigned long nr;
410 u64 num_blocks;
411 u64 blocks_used;
412 int ret = 0;
413 int err;
414
415 while(!list_empty(list)) {
416 struct btrfs_root *root;
417
418 mutex_lock(&tree_root->fs_info->fs_mutex);
419 dirty = list_entry(list->next, struct dirty_root, list);
420 list_del_init(&dirty->list);
421
422 num_blocks = btrfs_root_used(&dirty->root->root_item);
423 root = dirty->latest_root;
424
425 while(1) {
426 trans = btrfs_start_transaction(tree_root, 1);
427 ret = btrfs_drop_snapshot(trans, dirty->root);
428 if (ret != -EAGAIN) {
429 break;
430 }
431
432 err = btrfs_update_root(trans,
433 tree_root,
434 &dirty->root->root_key,
435 &dirty->root->root_item);
436 if (err)
437 ret = err;
438 nr = trans->blocks_used;
439 ret = btrfs_end_transaction(trans, tree_root);
440 BUG_ON(ret);
441 mutex_unlock(&tree_root->fs_info->fs_mutex);
442 btrfs_btree_balance_dirty(tree_root, nr);
443 schedule();
444
445 mutex_lock(&tree_root->fs_info->fs_mutex);
446 }
447 BUG_ON(ret);
448
449 num_blocks -= btrfs_root_used(&dirty->root->root_item);
450 blocks_used = btrfs_root_used(&root->root_item);
451 if (num_blocks) {
452 record_root_in_trans(root);
453 btrfs_set_root_used(&root->root_item,
454 blocks_used - num_blocks);
455 }
456 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
457 if (ret) {
458 BUG();
459 break;
460 }
461 nr = trans->blocks_used;
462 ret = btrfs_end_transaction(trans, tree_root);
463 BUG_ON(ret);
464
465 free_extent_buffer(dirty->root->node);
466 kfree(dirty->root);
467 kfree(dirty);
468 mutex_unlock(&tree_root->fs_info->fs_mutex);
469
470 btrfs_btree_balance_dirty(tree_root, nr);
471 schedule();
472 }
473 return ret;
474 }
475
476 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
477 struct btrfs_root *root)
478 {
479 unsigned long joined = 0;
480 unsigned long timeout = 1;
481 struct btrfs_transaction *cur_trans;
482 struct btrfs_transaction *prev_trans = NULL;
483 struct list_head dirty_fs_roots;
484 struct radix_tree_root pinned_copy;
485 DEFINE_WAIT(wait);
486 int ret;
487
488 init_bit_radix(&pinned_copy);
489 INIT_LIST_HEAD(&dirty_fs_roots);
490
491 mutex_lock(&root->fs_info->trans_mutex);
492 if (trans->transaction->in_commit) {
493 cur_trans = trans->transaction;
494 trans->transaction->use_count++;
495 mutex_unlock(&root->fs_info->trans_mutex);
496 btrfs_end_transaction(trans, root);
497
498 mutex_unlock(&root->fs_info->fs_mutex);
499 ret = wait_for_commit(root, cur_trans);
500 BUG_ON(ret);
501
502 mutex_lock(&root->fs_info->trans_mutex);
503 put_transaction(cur_trans);
504 mutex_unlock(&root->fs_info->trans_mutex);
505
506 mutex_lock(&root->fs_info->fs_mutex);
507 return 0;
508 }
509 trans->transaction->in_commit = 1;
510 cur_trans = trans->transaction;
511 if (cur_trans->list.prev != &root->fs_info->trans_list) {
512 prev_trans = list_entry(cur_trans->list.prev,
513 struct btrfs_transaction, list);
514 if (!prev_trans->commit_done) {
515 prev_trans->use_count++;
516 mutex_unlock(&root->fs_info->fs_mutex);
517 mutex_unlock(&root->fs_info->trans_mutex);
518
519 wait_for_commit(root, prev_trans);
520
521 mutex_lock(&root->fs_info->fs_mutex);
522 mutex_lock(&root->fs_info->trans_mutex);
523 put_transaction(prev_trans);
524 }
525 }
526
527 do {
528 joined = cur_trans->num_joined;
529 WARN_ON(cur_trans != trans->transaction);
530 prepare_to_wait(&cur_trans->writer_wait, &wait,
531 TASK_UNINTERRUPTIBLE);
532
533 if (cur_trans->num_writers > 1)
534 timeout = MAX_SCHEDULE_TIMEOUT;
535 else
536 timeout = 1;
537
538 mutex_unlock(&root->fs_info->fs_mutex);
539 mutex_unlock(&root->fs_info->trans_mutex);
540
541 schedule_timeout(timeout);
542
543 mutex_lock(&root->fs_info->fs_mutex);
544 mutex_lock(&root->fs_info->trans_mutex);
545 finish_wait(&cur_trans->writer_wait, &wait);
546 } while (cur_trans->num_writers > 1 ||
547 (cur_trans->num_joined != joined));
548
549 WARN_ON(cur_trans != trans->transaction);
550 ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
551 &dirty_fs_roots);
552 BUG_ON(ret);
553
554 ret = btrfs_commit_tree_roots(trans, root);
555 BUG_ON(ret);
556
557 cur_trans = root->fs_info->running_transaction;
558 root->fs_info->running_transaction = NULL;
559 btrfs_set_super_generation(&root->fs_info->super_copy,
560 cur_trans->transid);
561 btrfs_set_super_root(&root->fs_info->super_copy,
562 extent_buffer_blocknr(root->fs_info->tree_root->node));
563
564 write_extent_buffer(root->fs_info->sb_buffer,
565 &root->fs_info->super_copy, 0,
566 sizeof(root->fs_info->super_copy));
567
568 btrfs_copy_pinned(root, &pinned_copy);
569
570 mutex_unlock(&root->fs_info->trans_mutex);
571 mutex_unlock(&root->fs_info->fs_mutex);
572 ret = btrfs_write_and_wait_transaction(trans, root);
573 BUG_ON(ret);
574 write_ctree_super(trans, root);
575 mutex_lock(&root->fs_info->fs_mutex);
576 btrfs_finish_extent_commit(trans, root, &pinned_copy);
577 mutex_lock(&root->fs_info->trans_mutex);
578 cur_trans->commit_done = 1;
579 root->fs_info->last_trans_committed = cur_trans->transid;
580 wake_up(&cur_trans->commit_wait);
581 put_transaction(cur_trans);
582 put_transaction(cur_trans);
583
584 if (root->fs_info->closing)
585 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
586 else
587 list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
588
589 mutex_unlock(&root->fs_info->trans_mutex);
590 kmem_cache_free(btrfs_trans_handle_cachep, trans);
591
592 if (root->fs_info->closing) {
593 mutex_unlock(&root->fs_info->fs_mutex);
594 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
595 mutex_lock(&root->fs_info->fs_mutex);
596 }
597 return ret;
598 }
599
600 int btrfs_clean_old_snapshots(struct btrfs_root *root)
601 {
602 struct list_head dirty_roots;
603 INIT_LIST_HEAD(&dirty_roots);
604
605 mutex_lock(&root->fs_info->trans_mutex);
606 list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
607 mutex_unlock(&root->fs_info->trans_mutex);
608
609 if (!list_empty(&dirty_roots)) {
610 drop_dirty_roots(root, &dirty_roots);
611 }
612 return 0;
613 }
614 void btrfs_transaction_cleaner(struct work_struct *work)
615 {
616 struct btrfs_fs_info *fs_info = container_of(work,
617 struct btrfs_fs_info,
618 trans_work.work);
619
620 struct btrfs_root *root = fs_info->tree_root;
621 struct btrfs_transaction *cur;
622 struct btrfs_trans_handle *trans;
623 unsigned long now;
624 unsigned long delay = HZ * 30;
625 int ret;
626
627 mutex_lock(&root->fs_info->fs_mutex);
628 mutex_lock(&root->fs_info->trans_mutex);
629 cur = root->fs_info->running_transaction;
630 if (!cur) {
631 mutex_unlock(&root->fs_info->trans_mutex);
632 goto out;
633 }
634 now = get_seconds();
635 if (now < cur->start_time || now - cur->start_time < 30) {
636 mutex_unlock(&root->fs_info->trans_mutex);
637 delay = HZ * 5;
638 goto out;
639 }
640 mutex_unlock(&root->fs_info->trans_mutex);
641 btrfs_defrag_dirty_roots(root->fs_info);
642 trans = btrfs_start_transaction(root, 1);
643 ret = btrfs_commit_transaction(trans, root);
644 out:
645 mutex_unlock(&root->fs_info->fs_mutex);
646 btrfs_clean_old_snapshots(root);
647 btrfs_transaction_queue_work(root, delay);
648 }
649
650 void btrfs_transaction_queue_work(struct btrfs_root *root, int delay)
651 {
652 queue_delayed_work(trans_wq, &root->fs_info->trans_work, delay);
653 }
654
655 void btrfs_transaction_flush_work(struct btrfs_root *root)
656 {
657 cancel_rearming_delayed_workqueue(trans_wq, &root->fs_info->trans_work);
658 flush_workqueue(trans_wq);
659 }
660
661 void __init btrfs_init_transaction_sys(void)
662 {
663 trans_wq = create_workqueue("btrfs");
664 }
665
666 void __exit btrfs_exit_transaction_sys(void)
667 {
668 destroy_workqueue(trans_wq);
669 }
670
This page took 0.060497 seconds and 6 git commands to generate.