xref: /openbmc/linux/fs/btrfs/transaction.c (revision d1310b2e0cd98eb1348553e69b73827b436dca7b)
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 		root->fs_info->last_alloc = 0;
61 		cur_trans->num_writers = 1;
62 		cur_trans->num_joined = 0;
63 		cur_trans->transid = root->fs_info->generation;
64 		init_waitqueue_head(&cur_trans->writer_wait);
65 		init_waitqueue_head(&cur_trans->commit_wait);
66 		cur_trans->in_commit = 0;
67 		cur_trans->use_count = 1;
68 		cur_trans->commit_done = 0;
69 		cur_trans->start_time = get_seconds();
70 		INIT_LIST_HEAD(&cur_trans->pending_snapshots);
71 		list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
72 		btrfs_ordered_inode_tree_init(&cur_trans->ordered_inode_tree);
73 		extent_io_tree_init(&cur_trans->dirty_pages,
74 				     root->fs_info->btree_inode->i_mapping,
75 				     GFP_NOFS);
76 	} else {
77 		cur_trans->num_writers++;
78 		cur_trans->num_joined++;
79 	}
80 
81 	return 0;
82 }
83 
84 static int record_root_in_trans(struct btrfs_root *root)
85 {
86 	u64 running_trans_id = root->fs_info->running_transaction->transid;
87 	if (root->ref_cows && root->last_trans < running_trans_id) {
88 		WARN_ON(root == root->fs_info->extent_root);
89 		if (root->root_item.refs != 0) {
90 			radix_tree_tag_set(&root->fs_info->fs_roots_radix,
91 				   (unsigned long)root->root_key.objectid,
92 				   BTRFS_ROOT_TRANS_TAG);
93 			radix_tree_tag_set(&root->fs_info->fs_roots_radix,
94 				   (unsigned long)root->root_key.objectid,
95 				   BTRFS_ROOT_DEFRAG_TAG);
96 			root->commit_root = root->node;
97 			extent_buffer_get(root->node);
98 		} else {
99 			WARN_ON(1);
100 		}
101 		root->last_trans = running_trans_id;
102 	}
103 	return 0;
104 }
105 
106 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
107 						   int num_blocks)
108 {
109 	struct btrfs_trans_handle *h =
110 		kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
111 	int ret;
112 
113 	mutex_lock(&root->fs_info->trans_mutex);
114 	ret = join_transaction(root);
115 	BUG_ON(ret);
116 
117 	record_root_in_trans(root);
118 	h->transid = root->fs_info->running_transaction->transid;
119 	h->transaction = root->fs_info->running_transaction;
120 	h->blocks_reserved = num_blocks;
121 	h->blocks_used = 0;
122 	h->block_group = NULL;
123 	h->alloc_exclude_nr = 0;
124 	h->alloc_exclude_start = 0;
125 	root->fs_info->running_transaction->use_count++;
126 	mutex_unlock(&root->fs_info->trans_mutex);
127 	return h;
128 }
129 
130 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
131 			  struct btrfs_root *root)
132 {
133 	struct btrfs_transaction *cur_trans;
134 
135 	mutex_lock(&root->fs_info->trans_mutex);
136 	cur_trans = root->fs_info->running_transaction;
137 	WARN_ON(cur_trans != trans->transaction);
138 	WARN_ON(cur_trans->num_writers < 1);
139 	cur_trans->num_writers--;
140 	if (waitqueue_active(&cur_trans->writer_wait))
141 		wake_up(&cur_trans->writer_wait);
142 	put_transaction(cur_trans);
143 	mutex_unlock(&root->fs_info->trans_mutex);
144 	memset(trans, 0, sizeof(*trans));
145 	kmem_cache_free(btrfs_trans_handle_cachep, trans);
146 	return 0;
147 }
148 
149 
150 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
151 				     struct btrfs_root *root)
152 {
153 	int ret;
154 	int err;
155 	int werr = 0;
156 	struct extent_io_tree *dirty_pages;
157 	struct page *page;
158 	struct inode *btree_inode = root->fs_info->btree_inode;
159 	u64 start;
160 	u64 end;
161 	unsigned long index;
162 
163 	if (!trans || !trans->transaction) {
164 		return filemap_write_and_wait(btree_inode->i_mapping);
165 	}
166 	dirty_pages = &trans->transaction->dirty_pages;
167 	while(1) {
168 		ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
169 					    EXTENT_DIRTY);
170 		if (ret)
171 			break;
172 		clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
173 		while(start <= end) {
174 			index = start >> PAGE_CACHE_SHIFT;
175 			start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
176 			page = find_lock_page(btree_inode->i_mapping, index);
177 			if (!page)
178 				continue;
179 			if (PageWriteback(page)) {
180 				if (PageDirty(page))
181 					wait_on_page_writeback(page);
182 				else {
183 					unlock_page(page);
184 					page_cache_release(page);
185 					continue;
186 				}
187 			}
188 			err = write_one_page(page, 0);
189 			if (err)
190 				werr = err;
191 			page_cache_release(page);
192 		}
193 	}
194 	err = filemap_fdatawait(btree_inode->i_mapping);
195 	if (err)
196 		werr = err;
197 	return werr;
198 }
199 
200 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
201 			    struct btrfs_root *root)
202 {
203 	int ret;
204 	u64 old_extent_block;
205 	struct btrfs_fs_info *fs_info = root->fs_info;
206 	struct btrfs_root *tree_root = fs_info->tree_root;
207 	struct btrfs_root *extent_root = fs_info->extent_root;
208 
209 	btrfs_write_dirty_block_groups(trans, extent_root);
210 	while(1) {
211 		old_extent_block = btrfs_root_bytenr(&extent_root->root_item);
212 		if (old_extent_block == extent_root->node->start)
213 			break;
214 		btrfs_set_root_bytenr(&extent_root->root_item,
215 				      extent_root->node->start);
216 		btrfs_set_root_level(&extent_root->root_item,
217 				     btrfs_header_level(extent_root->node));
218 		ret = btrfs_update_root(trans, tree_root,
219 					&extent_root->root_key,
220 					&extent_root->root_item);
221 		BUG_ON(ret);
222 		btrfs_write_dirty_block_groups(trans, extent_root);
223 	}
224 	return 0;
225 }
226 
227 static int wait_for_commit(struct btrfs_root *root,
228 			   struct btrfs_transaction *commit)
229 {
230 	DEFINE_WAIT(wait);
231 	mutex_lock(&root->fs_info->trans_mutex);
232 	while(!commit->commit_done) {
233 		prepare_to_wait(&commit->commit_wait, &wait,
234 				TASK_UNINTERRUPTIBLE);
235 		if (commit->commit_done)
236 			break;
237 		mutex_unlock(&root->fs_info->trans_mutex);
238 		schedule();
239 		mutex_lock(&root->fs_info->trans_mutex);
240 	}
241 	mutex_unlock(&root->fs_info->trans_mutex);
242 	finish_wait(&commit->commit_wait, &wait);
243 	return 0;
244 }
245 
246 struct dirty_root {
247 	struct list_head list;
248 	struct btrfs_root *root;
249 	struct btrfs_root *latest_root;
250 };
251 
252 int btrfs_add_dead_root(struct btrfs_root *root,
253 			struct btrfs_root *latest,
254 			struct list_head *dead_list)
255 {
256 	struct dirty_root *dirty;
257 
258 	dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
259 	if (!dirty)
260 		return -ENOMEM;
261 	dirty->root = root;
262 	dirty->latest_root = latest;
263 	list_add(&dirty->list, dead_list);
264 	return 0;
265 }
266 
267 static int add_dirty_roots(struct btrfs_trans_handle *trans,
268 			   struct radix_tree_root *radix,
269 			   struct list_head *list)
270 {
271 	struct dirty_root *dirty;
272 	struct btrfs_root *gang[8];
273 	struct btrfs_root *root;
274 	int i;
275 	int ret;
276 	int err = 0;
277 	u32 refs;
278 
279 	while(1) {
280 		ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
281 						 ARRAY_SIZE(gang),
282 						 BTRFS_ROOT_TRANS_TAG);
283 		if (ret == 0)
284 			break;
285 		for (i = 0; i < ret; i++) {
286 			root = gang[i];
287 			radix_tree_tag_clear(radix,
288 				     (unsigned long)root->root_key.objectid,
289 				     BTRFS_ROOT_TRANS_TAG);
290 			if (root->commit_root == root->node) {
291 				WARN_ON(root->node->start !=
292 					btrfs_root_bytenr(&root->root_item));
293 				free_extent_buffer(root->commit_root);
294 				root->commit_root = NULL;
295 
296 				/* make sure to update the root on disk
297 				 * so we get any updates to the block used
298 				 * counts
299 				 */
300 				err = btrfs_update_root(trans,
301 						root->fs_info->tree_root,
302 						&root->root_key,
303 						&root->root_item);
304 				continue;
305 			}
306 			dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
307 			BUG_ON(!dirty);
308 			dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
309 			BUG_ON(!dirty->root);
310 
311 			memset(&root->root_item.drop_progress, 0,
312 			       sizeof(struct btrfs_disk_key));
313 			root->root_item.drop_level = 0;
314 
315 			memcpy(dirty->root, root, sizeof(*root));
316 			dirty->root->node = root->commit_root;
317 			dirty->latest_root = root;
318 			root->commit_root = NULL;
319 
320 			root->root_key.offset = root->fs_info->generation;
321 			btrfs_set_root_bytenr(&root->root_item,
322 					      root->node->start);
323 			btrfs_set_root_level(&root->root_item,
324 					     btrfs_header_level(root->node));
325 			err = btrfs_insert_root(trans, root->fs_info->tree_root,
326 						&root->root_key,
327 						&root->root_item);
328 			if (err)
329 				break;
330 
331 			refs = btrfs_root_refs(&dirty->root->root_item);
332 			btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
333 			err = btrfs_update_root(trans, root->fs_info->tree_root,
334 						&dirty->root->root_key,
335 						&dirty->root->root_item);
336 
337 			BUG_ON(err);
338 			if (refs == 1) {
339 				list_add(&dirty->list, list);
340 			} else {
341 				WARN_ON(1);
342 				kfree(dirty->root);
343 				kfree(dirty);
344 			}
345 		}
346 	}
347 	return err;
348 }
349 
350 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
351 {
352 	struct btrfs_fs_info *info = root->fs_info;
353 	int ret;
354 	struct btrfs_trans_handle *trans;
355 	unsigned long nr;
356 
357 	if (root->defrag_running)
358 		return 0;
359 	trans = btrfs_start_transaction(root, 1);
360 	while (1) {
361 		root->defrag_running = 1;
362 		ret = btrfs_defrag_leaves(trans, root, cacheonly);
363 		nr = trans->blocks_used;
364 		btrfs_end_transaction(trans, root);
365 		mutex_unlock(&info->fs_mutex);
366 		btrfs_btree_balance_dirty(info->tree_root, nr);
367 		cond_resched();
368 
369 		mutex_lock(&info->fs_mutex);
370 		trans = btrfs_start_transaction(root, 1);
371 		if (ret != -EAGAIN)
372 			break;
373 	}
374 	root->defrag_running = 0;
375 	radix_tree_tag_clear(&info->fs_roots_radix,
376 		     (unsigned long)root->root_key.objectid,
377 		     BTRFS_ROOT_DEFRAG_TAG);
378 	btrfs_end_transaction(trans, root);
379 	return 0;
380 }
381 
382 int btrfs_defrag_dirty_roots(struct btrfs_fs_info *info)
383 {
384 	struct btrfs_root *gang[1];
385 	struct btrfs_root *root;
386 	int i;
387 	int ret;
388 	int err = 0;
389 	u64 last = 0;
390 
391 	while(1) {
392 		ret = radix_tree_gang_lookup_tag(&info->fs_roots_radix,
393 						 (void **)gang, last,
394 						 ARRAY_SIZE(gang),
395 						 BTRFS_ROOT_DEFRAG_TAG);
396 		if (ret == 0)
397 			break;
398 		for (i = 0; i < ret; i++) {
399 			root = gang[i];
400 			last = root->root_key.objectid + 1;
401 			btrfs_defrag_root(root, 1);
402 		}
403 	}
404 	btrfs_defrag_root(info->extent_root, 1);
405 	return err;
406 }
407 
408 static int drop_dirty_roots(struct btrfs_root *tree_root,
409 			    struct list_head *list)
410 {
411 	struct dirty_root *dirty;
412 	struct btrfs_trans_handle *trans;
413 	unsigned long nr;
414 	u64 num_bytes;
415 	u64 bytes_used;
416 	int ret = 0;
417 	int err;
418 
419 	while(!list_empty(list)) {
420 		struct btrfs_root *root;
421 
422 		mutex_lock(&tree_root->fs_info->fs_mutex);
423 		dirty = list_entry(list->next, struct dirty_root, list);
424 		list_del_init(&dirty->list);
425 
426 		num_bytes = btrfs_root_used(&dirty->root->root_item);
427 		root = dirty->latest_root;
428 		root->fs_info->throttles++;
429 
430 		while(1) {
431 			trans = btrfs_start_transaction(tree_root, 1);
432 			ret = btrfs_drop_snapshot(trans, dirty->root);
433 			if (ret != -EAGAIN) {
434 				break;
435 			}
436 
437 			err = btrfs_update_root(trans,
438 					tree_root,
439 					&dirty->root->root_key,
440 					&dirty->root->root_item);
441 			if (err)
442 				ret = err;
443 			nr = trans->blocks_used;
444 			ret = btrfs_end_transaction(trans, tree_root);
445 			BUG_ON(ret);
446 			mutex_unlock(&tree_root->fs_info->fs_mutex);
447 			btrfs_btree_balance_dirty(tree_root, nr);
448 			cond_resched();
449 			mutex_lock(&tree_root->fs_info->fs_mutex);
450 		}
451 		BUG_ON(ret);
452 		root->fs_info->throttles--;
453 
454 		num_bytes -= btrfs_root_used(&dirty->root->root_item);
455 		bytes_used = btrfs_root_used(&root->root_item);
456 		if (num_bytes) {
457 			record_root_in_trans(root);
458 			btrfs_set_root_used(&root->root_item,
459 					    bytes_used - num_bytes);
460 		}
461 		ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
462 		if (ret) {
463 			BUG();
464 			break;
465 		}
466 		nr = trans->blocks_used;
467 		ret = btrfs_end_transaction(trans, tree_root);
468 		BUG_ON(ret);
469 
470 		free_extent_buffer(dirty->root->node);
471 		kfree(dirty->root);
472 		kfree(dirty);
473 		mutex_unlock(&tree_root->fs_info->fs_mutex);
474 
475 		btrfs_btree_balance_dirty(tree_root, nr);
476 		cond_resched();
477 	}
478 	return ret;
479 }
480 
481 int btrfs_write_ordered_inodes(struct btrfs_trans_handle *trans,
482 				struct btrfs_root *root)
483 {
484 	struct btrfs_transaction *cur_trans = trans->transaction;
485 	struct inode *inode;
486 	u64 root_objectid = 0;
487 	u64 objectid = 0;
488 	int ret;
489 
490 	root->fs_info->throttles++;
491 	while(1) {
492 		ret = btrfs_find_first_ordered_inode(
493 				&cur_trans->ordered_inode_tree,
494 				&root_objectid, &objectid, &inode);
495 		if (!ret)
496 			break;
497 
498 		mutex_unlock(&root->fs_info->trans_mutex);
499 		mutex_unlock(&root->fs_info->fs_mutex);
500 
501 		if (S_ISREG(inode->i_mode))
502 			filemap_fdatawrite(inode->i_mapping);
503 		iput(inode);
504 
505 		mutex_lock(&root->fs_info->fs_mutex);
506 		mutex_lock(&root->fs_info->trans_mutex);
507 	}
508 	while(1) {
509 		root_objectid = 0;
510 		objectid = 0;
511 		ret = btrfs_find_del_first_ordered_inode(
512 				&cur_trans->ordered_inode_tree,
513 				&root_objectid, &objectid, &inode);
514 		if (!ret)
515 			break;
516 		mutex_unlock(&root->fs_info->trans_mutex);
517 		mutex_unlock(&root->fs_info->fs_mutex);
518 
519 		if (S_ISREG(inode->i_mode))
520 			filemap_write_and_wait(inode->i_mapping);
521 		atomic_dec(&inode->i_count);
522 		iput(inode);
523 
524 		mutex_lock(&root->fs_info->fs_mutex);
525 		mutex_lock(&root->fs_info->trans_mutex);
526 	}
527 	root->fs_info->throttles--;
528 	return 0;
529 }
530 
531 static int create_pending_snapshot(struct btrfs_trans_handle *trans,
532 				   struct btrfs_fs_info *fs_info,
533 				   struct btrfs_pending_snapshot *pending)
534 {
535 	struct btrfs_key key;
536 	struct btrfs_root_item new_root_item;
537 	struct btrfs_root *tree_root = fs_info->tree_root;
538 	struct btrfs_root *root = pending->root;
539 	struct extent_buffer *tmp;
540 	int ret;
541 	u64 objectid;
542 
543 	ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
544 	if (ret)
545 		goto fail;
546 
547 	memcpy(&new_root_item, &root->root_item, sizeof(new_root_item));
548 
549 	key.objectid = objectid;
550 	key.offset = 1;
551 	btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
552 
553 	extent_buffer_get(root->node);
554 	btrfs_cow_block(trans, root, root->node, NULL, 0, &tmp);
555 	free_extent_buffer(tmp);
556 
557 	btrfs_copy_root(trans, root, root->node, &tmp, objectid);
558 
559 	btrfs_set_root_bytenr(&new_root_item, tmp->start);
560 	btrfs_set_root_level(&new_root_item, btrfs_header_level(tmp));
561 	ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
562 				&new_root_item);
563 	free_extent_buffer(tmp);
564 	if (ret)
565 		goto fail;
566 
567 	/*
568 	 * insert the directory item
569 	 */
570 	key.offset = (u64)-1;
571 	ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
572 				    pending->name, strlen(pending->name),
573 				    root->fs_info->sb->s_root->d_inode->i_ino,
574 				    &key, BTRFS_FT_DIR);
575 
576 	if (ret)
577 		goto fail;
578 
579 	ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
580 			     pending->name, strlen(pending->name), objectid,
581 			     root->fs_info->sb->s_root->d_inode->i_ino);
582 fail:
583 	return ret;
584 }
585 
586 static int create_pending_snapshots(struct btrfs_trans_handle *trans,
587 				   struct btrfs_fs_info *fs_info)
588 {
589 	struct btrfs_pending_snapshot *pending;
590 	struct list_head *head = &trans->transaction->pending_snapshots;
591 	int ret;
592 
593 	while(!list_empty(head)) {
594 		pending = list_entry(head->next,
595 				     struct btrfs_pending_snapshot, list);
596 		ret = create_pending_snapshot(trans, fs_info, pending);
597 		BUG_ON(ret);
598 		list_del(&pending->list);
599 		kfree(pending->name);
600 		kfree(pending);
601 	}
602 	return 0;
603 }
604 
605 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
606 			     struct btrfs_root *root)
607 {
608 	unsigned long joined = 0;
609 	unsigned long timeout = 1;
610 	struct btrfs_transaction *cur_trans;
611 	struct btrfs_transaction *prev_trans = NULL;
612 	struct list_head dirty_fs_roots;
613 	struct extent_io_tree *pinned_copy;
614 	DEFINE_WAIT(wait);
615 	int ret;
616 
617 	INIT_LIST_HEAD(&dirty_fs_roots);
618 
619 	mutex_lock(&root->fs_info->trans_mutex);
620 	if (trans->transaction->in_commit) {
621 		cur_trans = trans->transaction;
622 		trans->transaction->use_count++;
623 		mutex_unlock(&root->fs_info->trans_mutex);
624 		btrfs_end_transaction(trans, root);
625 
626 		mutex_unlock(&root->fs_info->fs_mutex);
627 		ret = wait_for_commit(root, cur_trans);
628 		BUG_ON(ret);
629 
630 		mutex_lock(&root->fs_info->trans_mutex);
631 		put_transaction(cur_trans);
632 		mutex_unlock(&root->fs_info->trans_mutex);
633 
634 		mutex_lock(&root->fs_info->fs_mutex);
635 		return 0;
636 	}
637 
638 	pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
639 	if (!pinned_copy)
640 		return -ENOMEM;
641 
642 	extent_io_tree_init(pinned_copy,
643 			     root->fs_info->btree_inode->i_mapping, GFP_NOFS);
644 
645 	trans->transaction->in_commit = 1;
646 	cur_trans = trans->transaction;
647 	if (cur_trans->list.prev != &root->fs_info->trans_list) {
648 		prev_trans = list_entry(cur_trans->list.prev,
649 					struct btrfs_transaction, list);
650 		if (!prev_trans->commit_done) {
651 			prev_trans->use_count++;
652 			mutex_unlock(&root->fs_info->fs_mutex);
653 			mutex_unlock(&root->fs_info->trans_mutex);
654 
655 			wait_for_commit(root, prev_trans);
656 
657 			mutex_lock(&root->fs_info->fs_mutex);
658 			mutex_lock(&root->fs_info->trans_mutex);
659 			put_transaction(prev_trans);
660 		}
661 	}
662 
663 	do {
664 		joined = cur_trans->num_joined;
665 		WARN_ON(cur_trans != trans->transaction);
666 		prepare_to_wait(&cur_trans->writer_wait, &wait,
667 				TASK_UNINTERRUPTIBLE);
668 
669 		if (cur_trans->num_writers > 1)
670 			timeout = MAX_SCHEDULE_TIMEOUT;
671 		else
672 			timeout = 1;
673 
674 		mutex_unlock(&root->fs_info->fs_mutex);
675 		mutex_unlock(&root->fs_info->trans_mutex);
676 
677 		schedule_timeout(timeout);
678 
679 		mutex_lock(&root->fs_info->fs_mutex);
680 		mutex_lock(&root->fs_info->trans_mutex);
681 		finish_wait(&cur_trans->writer_wait, &wait);
682 		ret = btrfs_write_ordered_inodes(trans, root);
683 
684 	} while (cur_trans->num_writers > 1 ||
685 		 (cur_trans->num_joined != joined));
686 
687 	ret = create_pending_snapshots(trans, root->fs_info);
688 	BUG_ON(ret);
689 
690 	WARN_ON(cur_trans != trans->transaction);
691 
692 	ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
693 			      &dirty_fs_roots);
694 	BUG_ON(ret);
695 
696 	ret = btrfs_commit_tree_roots(trans, root);
697 	BUG_ON(ret);
698 
699 	cur_trans = root->fs_info->running_transaction;
700 	spin_lock(&root->fs_info->new_trans_lock);
701 	root->fs_info->running_transaction = NULL;
702 	spin_unlock(&root->fs_info->new_trans_lock);
703 	btrfs_set_super_generation(&root->fs_info->super_copy,
704 				   cur_trans->transid);
705 	btrfs_set_super_root(&root->fs_info->super_copy,
706 			     root->fs_info->tree_root->node->start);
707 	btrfs_set_super_root_level(&root->fs_info->super_copy,
708 			   btrfs_header_level(root->fs_info->tree_root->node));
709 
710 	write_extent_buffer(root->fs_info->sb_buffer,
711 			    &root->fs_info->super_copy, 0,
712 			    sizeof(root->fs_info->super_copy));
713 
714 	btrfs_copy_pinned(root, pinned_copy);
715 
716 	mutex_unlock(&root->fs_info->trans_mutex);
717 	mutex_unlock(&root->fs_info->fs_mutex);
718 	ret = btrfs_write_and_wait_transaction(trans, root);
719 	BUG_ON(ret);
720 	write_ctree_super(trans, root);
721 
722 	mutex_lock(&root->fs_info->fs_mutex);
723 	btrfs_finish_extent_commit(trans, root, pinned_copy);
724 	mutex_lock(&root->fs_info->trans_mutex);
725 
726 	kfree(pinned_copy);
727 
728 	cur_trans->commit_done = 1;
729 	root->fs_info->last_trans_committed = cur_trans->transid;
730 	wake_up(&cur_trans->commit_wait);
731 	put_transaction(cur_trans);
732 	put_transaction(cur_trans);
733 
734 	if (root->fs_info->closing)
735 		list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
736 	else
737 		list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
738 
739 	mutex_unlock(&root->fs_info->trans_mutex);
740 	kmem_cache_free(btrfs_trans_handle_cachep, trans);
741 
742 	if (root->fs_info->closing) {
743 		mutex_unlock(&root->fs_info->fs_mutex);
744 		drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
745 		mutex_lock(&root->fs_info->fs_mutex);
746 	}
747 	return ret;
748 }
749 
750 int btrfs_clean_old_snapshots(struct btrfs_root *root)
751 {
752 	struct list_head dirty_roots;
753 	INIT_LIST_HEAD(&dirty_roots);
754 
755 	mutex_lock(&root->fs_info->trans_mutex);
756 	list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
757 	mutex_unlock(&root->fs_info->trans_mutex);
758 
759 	if (!list_empty(&dirty_roots)) {
760 		drop_dirty_roots(root, &dirty_roots);
761 	}
762 	return 0;
763 }
764 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
765 void btrfs_transaction_cleaner(void *p)
766 #else
767 void btrfs_transaction_cleaner(struct work_struct *work)
768 #endif
769 {
770 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
771 	struct btrfs_fs_info *fs_info = p;
772 #else
773 	struct btrfs_fs_info *fs_info = container_of(work,
774 						     struct btrfs_fs_info,
775 						     trans_work.work);
776 
777 #endif
778 	struct btrfs_root *root = fs_info->tree_root;
779 	struct btrfs_transaction *cur;
780 	struct btrfs_trans_handle *trans;
781 	unsigned long now;
782 	unsigned long delay = HZ * 30;
783 	int ret;
784 
785 	mutex_lock(&root->fs_info->fs_mutex);
786 	mutex_lock(&root->fs_info->trans_mutex);
787 	cur = root->fs_info->running_transaction;
788 	if (!cur) {
789 		mutex_unlock(&root->fs_info->trans_mutex);
790 		goto out;
791 	}
792 	now = get_seconds();
793 	if (now < cur->start_time || now - cur->start_time < 30) {
794 		mutex_unlock(&root->fs_info->trans_mutex);
795 		delay = HZ * 5;
796 		goto out;
797 	}
798 	mutex_unlock(&root->fs_info->trans_mutex);
799 	btrfs_defrag_dirty_roots(root->fs_info);
800 	trans = btrfs_start_transaction(root, 1);
801 	ret = btrfs_commit_transaction(trans, root);
802 out:
803 	mutex_unlock(&root->fs_info->fs_mutex);
804 	btrfs_clean_old_snapshots(root);
805 	btrfs_transaction_queue_work(root, delay);
806 }
807 
808 void btrfs_transaction_queue_work(struct btrfs_root *root, int delay)
809 {
810 	queue_delayed_work(trans_wq, &root->fs_info->trans_work, delay);
811 }
812 
813 void btrfs_transaction_flush_work(struct btrfs_root *root)
814 {
815 	cancel_rearming_delayed_workqueue(trans_wq, &root->fs_info->trans_work);
816 	flush_workqueue(trans_wq);
817 }
818 
819 void __init btrfs_init_transaction_sys(void)
820 {
821 	trans_wq = create_workqueue("btrfs");
822 }
823 
824 void btrfs_exit_transaction_sys(void)
825 {
826 	destroy_workqueue(trans_wq);
827 }
828 
829