Btrfs: Allow tree blocks larger than the page size
[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_bytenr(&extent_root->root_item);
209 if (old_extent_block == extent_root->node->start)
210 break;
211 btrfs_set_root_bytenr(&extent_root->root_item,
212 extent_root->node->start);
213 btrfs_set_root_level(&extent_root->root_item,
214 btrfs_header_level(extent_root->node));
215 ret = btrfs_update_root(trans, tree_root,
216 &extent_root->root_key,
217 &extent_root->root_item);
218 BUG_ON(ret);
219 btrfs_write_dirty_block_groups(trans, extent_root);
220 }
221 return 0;
222 }
223
224 static int wait_for_commit(struct btrfs_root *root,
225 struct btrfs_transaction *commit)
226 {
227 DEFINE_WAIT(wait);
228 mutex_lock(&root->fs_info->trans_mutex);
229 while(!commit->commit_done) {
230 prepare_to_wait(&commit->commit_wait, &wait,
231 TASK_UNINTERRUPTIBLE);
232 if (commit->commit_done)
233 break;
234 mutex_unlock(&root->fs_info->trans_mutex);
235 schedule();
236 mutex_lock(&root->fs_info->trans_mutex);
237 }
238 mutex_unlock(&root->fs_info->trans_mutex);
239 finish_wait(&commit->commit_wait, &wait);
240 return 0;
241 }
242
243 struct dirty_root {
244 struct list_head list;
245 struct btrfs_root *root;
246 struct btrfs_root *latest_root;
247 };
248
249 int btrfs_add_dead_root(struct btrfs_root *root,
250 struct btrfs_root *latest,
251 struct list_head *dead_list)
252 {
253 struct dirty_root *dirty;
254
255 dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
256 if (!dirty)
257 return -ENOMEM;
258 dirty->root = root;
259 dirty->latest_root = latest;
260 list_add(&dirty->list, dead_list);
261 return 0;
262 }
263
264 static int add_dirty_roots(struct btrfs_trans_handle *trans,
265 struct radix_tree_root *radix,
266 struct list_head *list)
267 {
268 struct dirty_root *dirty;
269 struct btrfs_root *gang[8];
270 struct btrfs_root *root;
271 int i;
272 int ret;
273 int err = 0;
274 u32 refs;
275
276 while(1) {
277 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
278 ARRAY_SIZE(gang),
279 BTRFS_ROOT_TRANS_TAG);
280 if (ret == 0)
281 break;
282 for (i = 0; i < ret; i++) {
283 root = gang[i];
284 radix_tree_tag_clear(radix,
285 (unsigned long)root->root_key.objectid,
286 BTRFS_ROOT_TRANS_TAG);
287 if (root->commit_root == root->node) {
288 WARN_ON(root->node->start !=
289 btrfs_root_bytenr(&root->root_item));
290 free_extent_buffer(root->commit_root);
291 root->commit_root = NULL;
292
293 /* make sure to update the root on disk
294 * so we get any updates to the block used
295 * counts
296 */
297 err = btrfs_update_root(trans,
298 root->fs_info->tree_root,
299 &root->root_key,
300 &root->root_item);
301 continue;
302 }
303 dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
304 BUG_ON(!dirty);
305 dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
306 BUG_ON(!dirty->root);
307
308 memset(&root->root_item.drop_progress, 0,
309 sizeof(struct btrfs_disk_key));
310 root->root_item.drop_level = 0;
311
312 memcpy(dirty->root, root, sizeof(*root));
313 dirty->root->node = root->commit_root;
314 dirty->latest_root = root;
315 root->commit_root = NULL;
316
317 root->root_key.offset = root->fs_info->generation;
318 btrfs_set_root_bytenr(&root->root_item,
319 root->node->start);
320 btrfs_set_root_level(&root->root_item,
321 btrfs_header_level(root->node));
322 err = btrfs_insert_root(trans, root->fs_info->tree_root,
323 &root->root_key,
324 &root->root_item);
325 if (err)
326 break;
327
328 refs = btrfs_root_refs(&dirty->root->root_item);
329 btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
330 err = btrfs_update_root(trans, root->fs_info->tree_root,
331 &dirty->root->root_key,
332 &dirty->root->root_item);
333
334 BUG_ON(err);
335 if (refs == 1) {
336 list_add(&dirty->list, list);
337 } else {
338 WARN_ON(1);
339 kfree(dirty->root);
340 kfree(dirty);
341 }
342 }
343 }
344 return err;
345 }
346
347 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
348 {
349 struct btrfs_fs_info *info = root->fs_info;
350 int ret;
351 struct btrfs_trans_handle *trans;
352 unsigned long nr;
353
354 if (root->defrag_running)
355 return 0;
356
357 trans = btrfs_start_transaction(root, 1);
358 while (0) {
359 root->defrag_running = 1;
360 ret = btrfs_defrag_leaves(trans, root, cacheonly);
361 nr = trans->blocks_used;
362 btrfs_end_transaction(trans, root);
363 mutex_unlock(&info->fs_mutex);
364
365 btrfs_btree_balance_dirty(info->tree_root, nr);
366 cond_resched();
367
368 mutex_lock(&info->fs_mutex);
369 trans = btrfs_start_transaction(root, 1);
370 if (ret != -EAGAIN)
371 break;
372 }
373 root->defrag_running = 0;
374 radix_tree_tag_clear(&info->fs_roots_radix,
375 (unsigned long)root->root_key.objectid,
376 BTRFS_ROOT_DEFRAG_TAG);
377 btrfs_end_transaction(trans, root);
378 return 0;
379 }
380
381 int btrfs_defrag_dirty_roots(struct btrfs_fs_info *info)
382 {
383 struct btrfs_root *gang[1];
384 struct btrfs_root *root;
385 int i;
386 int ret;
387 int err = 0;
388 u64 last = 0;
389
390 while(1) {
391 ret = radix_tree_gang_lookup_tag(&info->fs_roots_radix,
392 (void **)gang, last,
393 ARRAY_SIZE(gang),
394 BTRFS_ROOT_DEFRAG_TAG);
395 if (ret == 0)
396 break;
397 for (i = 0; i < ret; i++) {
398 root = gang[i];
399 last = root->root_key.objectid + 1;
400 btrfs_defrag_root(root, 1);
401 }
402 }
403 // btrfs_defrag_root(info->extent_root, 1);
404 return err;
405 }
406
407 static int drop_dirty_roots(struct btrfs_root *tree_root,
408 struct list_head *list)
409 {
410 struct dirty_root *dirty;
411 struct btrfs_trans_handle *trans;
412 unsigned long nr;
413 u64 num_bytes;
414 u64 bytes_used;
415 int ret = 0;
416 int err;
417
418 while(!list_empty(list)) {
419 struct btrfs_root *root;
420
421 mutex_lock(&tree_root->fs_info->fs_mutex);
422 dirty = list_entry(list->next, struct dirty_root, list);
423 list_del_init(&dirty->list);
424
425 num_bytes = btrfs_root_used(&dirty->root->root_item);
426 root = dirty->latest_root;
427
428 while(1) {
429 trans = btrfs_start_transaction(tree_root, 1);
430 ret = btrfs_drop_snapshot(trans, dirty->root);
431 if (ret != -EAGAIN) {
432 break;
433 }
434
435 err = btrfs_update_root(trans,
436 tree_root,
437 &dirty->root->root_key,
438 &dirty->root->root_item);
439 if (err)
440 ret = err;
441 nr = trans->blocks_used;
442 ret = btrfs_end_transaction(trans, tree_root);
443 BUG_ON(ret);
444 mutex_unlock(&tree_root->fs_info->fs_mutex);
445 btrfs_btree_balance_dirty(tree_root, nr);
446 schedule();
447
448 mutex_lock(&tree_root->fs_info->fs_mutex);
449 }
450 BUG_ON(ret);
451
452 num_bytes -= btrfs_root_used(&dirty->root->root_item);
453 bytes_used = btrfs_root_used(&root->root_item);
454 if (num_bytes) {
455 record_root_in_trans(root);
456 btrfs_set_root_used(&root->root_item,
457 bytes_used - num_bytes);
458 }
459 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
460 if (ret) {
461 BUG();
462 break;
463 }
464 nr = trans->blocks_used;
465 ret = btrfs_end_transaction(trans, tree_root);
466 BUG_ON(ret);
467
468 free_extent_buffer(dirty->root->node);
469 kfree(dirty->root);
470 kfree(dirty);
471 mutex_unlock(&tree_root->fs_info->fs_mutex);
472
473 btrfs_btree_balance_dirty(tree_root, nr);
474 schedule();
475 }
476 return ret;
477 }
478
479 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
480 struct btrfs_root *root)
481 {
482 unsigned long joined = 0;
483 unsigned long timeout = 1;
484 struct btrfs_transaction *cur_trans;
485 struct btrfs_transaction *prev_trans = NULL;
486 struct list_head dirty_fs_roots;
487 struct extent_map_tree pinned_copy;
488 DEFINE_WAIT(wait);
489 int ret;
490
491 extent_map_tree_init(&pinned_copy,
492 root->fs_info->btree_inode->i_mapping, GFP_NOFS);
493 INIT_LIST_HEAD(&dirty_fs_roots);
494
495 mutex_lock(&root->fs_info->trans_mutex);
496 if (trans->transaction->in_commit) {
497 cur_trans = trans->transaction;
498 trans->transaction->use_count++;
499 mutex_unlock(&root->fs_info->trans_mutex);
500 btrfs_end_transaction(trans, root);
501
502 mutex_unlock(&root->fs_info->fs_mutex);
503 ret = wait_for_commit(root, cur_trans);
504 BUG_ON(ret);
505
506 mutex_lock(&root->fs_info->trans_mutex);
507 put_transaction(cur_trans);
508 mutex_unlock(&root->fs_info->trans_mutex);
509
510 mutex_lock(&root->fs_info->fs_mutex);
511 return 0;
512 }
513 trans->transaction->in_commit = 1;
514 cur_trans = trans->transaction;
515 if (cur_trans->list.prev != &root->fs_info->trans_list) {
516 prev_trans = list_entry(cur_trans->list.prev,
517 struct btrfs_transaction, list);
518 if (!prev_trans->commit_done) {
519 prev_trans->use_count++;
520 mutex_unlock(&root->fs_info->fs_mutex);
521 mutex_unlock(&root->fs_info->trans_mutex);
522
523 wait_for_commit(root, prev_trans);
524
525 mutex_lock(&root->fs_info->fs_mutex);
526 mutex_lock(&root->fs_info->trans_mutex);
527 put_transaction(prev_trans);
528 }
529 }
530
531 do {
532 joined = cur_trans->num_joined;
533 WARN_ON(cur_trans != trans->transaction);
534 prepare_to_wait(&cur_trans->writer_wait, &wait,
535 TASK_UNINTERRUPTIBLE);
536
537 if (cur_trans->num_writers > 1)
538 timeout = MAX_SCHEDULE_TIMEOUT;
539 else
540 timeout = 1;
541
542 mutex_unlock(&root->fs_info->fs_mutex);
543 mutex_unlock(&root->fs_info->trans_mutex);
544
545 schedule_timeout(timeout);
546
547 mutex_lock(&root->fs_info->fs_mutex);
548 mutex_lock(&root->fs_info->trans_mutex);
549 finish_wait(&cur_trans->writer_wait, &wait);
550 } while (cur_trans->num_writers > 1 ||
551 (cur_trans->num_joined != joined));
552
553 WARN_ON(cur_trans != trans->transaction);
554 ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
555 &dirty_fs_roots);
556 BUG_ON(ret);
557
558 ret = btrfs_commit_tree_roots(trans, root);
559 BUG_ON(ret);
560
561 cur_trans = root->fs_info->running_transaction;
562 root->fs_info->running_transaction = NULL;
563 btrfs_set_super_generation(&root->fs_info->super_copy,
564 cur_trans->transid);
565 btrfs_set_super_root(&root->fs_info->super_copy,
566 root->fs_info->tree_root->node->start);
567 btrfs_set_super_root_level(&root->fs_info->super_copy,
568 btrfs_header_level(root->fs_info->tree_root->node));
569
570 write_extent_buffer(root->fs_info->sb_buffer,
571 &root->fs_info->super_copy, 0,
572 sizeof(root->fs_info->super_copy));
573
574 btrfs_copy_pinned(root, &pinned_copy);
575
576 mutex_unlock(&root->fs_info->trans_mutex);
577 mutex_unlock(&root->fs_info->fs_mutex);
578 ret = btrfs_write_and_wait_transaction(trans, root);
579 BUG_ON(ret);
580 write_ctree_super(trans, root);
581 mutex_lock(&root->fs_info->fs_mutex);
582 btrfs_finish_extent_commit(trans, root, &pinned_copy);
583 mutex_lock(&root->fs_info->trans_mutex);
584 cur_trans->commit_done = 1;
585 root->fs_info->last_trans_committed = cur_trans->transid;
586 wake_up(&cur_trans->commit_wait);
587 put_transaction(cur_trans);
588 put_transaction(cur_trans);
589
590 if (root->fs_info->closing)
591 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
592 else
593 list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
594
595 mutex_unlock(&root->fs_info->trans_mutex);
596 kmem_cache_free(btrfs_trans_handle_cachep, trans);
597
598 if (root->fs_info->closing) {
599 mutex_unlock(&root->fs_info->fs_mutex);
600 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
601 mutex_lock(&root->fs_info->fs_mutex);
602 }
603 return ret;
604 }
605
606 int btrfs_clean_old_snapshots(struct btrfs_root *root)
607 {
608 struct list_head dirty_roots;
609 INIT_LIST_HEAD(&dirty_roots);
610
611 mutex_lock(&root->fs_info->trans_mutex);
612 list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
613 mutex_unlock(&root->fs_info->trans_mutex);
614
615 if (!list_empty(&dirty_roots)) {
616 drop_dirty_roots(root, &dirty_roots);
617 }
618 return 0;
619 }
620 void btrfs_transaction_cleaner(struct work_struct *work)
621 {
622 struct btrfs_fs_info *fs_info = container_of(work,
623 struct btrfs_fs_info,
624 trans_work.work);
625
626 struct btrfs_root *root = fs_info->tree_root;
627 struct btrfs_transaction *cur;
628 struct btrfs_trans_handle *trans;
629 unsigned long now;
630 unsigned long delay = HZ * 30;
631 int ret;
632
633 mutex_lock(&root->fs_info->fs_mutex);
634 mutex_lock(&root->fs_info->trans_mutex);
635 cur = root->fs_info->running_transaction;
636 if (!cur) {
637 mutex_unlock(&root->fs_info->trans_mutex);
638 goto out;
639 }
640 now = get_seconds();
641 if (now < cur->start_time || now - cur->start_time < 30) {
642 mutex_unlock(&root->fs_info->trans_mutex);
643 delay = HZ * 5;
644 goto out;
645 }
646 mutex_unlock(&root->fs_info->trans_mutex);
647 btrfs_defrag_dirty_roots(root->fs_info);
648 trans = btrfs_start_transaction(root, 1);
649 ret = btrfs_commit_transaction(trans, root);
650 out:
651 mutex_unlock(&root->fs_info->fs_mutex);
652 btrfs_clean_old_snapshots(root);
653 btrfs_transaction_queue_work(root, delay);
654 }
655
656 void btrfs_transaction_queue_work(struct btrfs_root *root, int delay)
657 {
658 queue_delayed_work(trans_wq, &root->fs_info->trans_work, delay);
659 }
660
661 void btrfs_transaction_flush_work(struct btrfs_root *root)
662 {
663 cancel_rearming_delayed_workqueue(trans_wq, &root->fs_info->trans_work);
664 flush_workqueue(trans_wq);
665 }
666
667 void __init btrfs_init_transaction_sys(void)
668 {
669 trans_wq = create_workqueue("btrfs");
670 }
671
672 void __exit btrfs_exit_transaction_sys(void)
673 {
674 destroy_workqueue(trans_wq);
675 }
676
This page took 0.158681 seconds and 5 git commands to generate.