transaction.c 28.8 KB
Newer Older
Chris Mason's avatar
Chris Mason committed
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
/*
 * Copyright (C) 2007 Oracle.  All rights reserved.
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public
 * License v2 as published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * You should have received a copy of the GNU General Public
 * License along with this program; if not, write to the
 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
 * Boston, MA 021110-1307, USA.
 */

Chris Mason's avatar
Chris Mason committed
19
#include <linux/fs.h>
Chris Mason's avatar
Chris Mason committed
20
#include <linux/sched.h>
21
#include <linux/writeback.h>
22
#include <linux/pagemap.h>
23
#include <linux/blkdev.h>
Chris Mason's avatar
Chris Mason committed
24
25
26
#include "ctree.h"
#include "disk-io.h"
#include "transaction.h"
27
#include "locking.h"
28
#include "tree-log.h"
Chris Mason's avatar
Chris Mason committed
29

30
31
#define BTRFS_ROOT_TRANS_TAG 0

32
static noinline void put_transaction(struct btrfs_transaction *transaction)
Chris Mason's avatar
Chris Mason committed
33
{
Chris Mason's avatar
Chris Mason committed
34
	WARN_ON(transaction->use_count == 0);
Chris Mason's avatar
Chris Mason committed
35
	transaction->use_count--;
Chris Mason's avatar
Chris Mason committed
36
	if (transaction->use_count == 0) {
Chris Mason's avatar
Chris Mason committed
37
		list_del_init(&transaction->list);
Chris Mason's avatar
Chris Mason committed
38
39
		memset(transaction, 0, sizeof(*transaction));
		kmem_cache_free(btrfs_transaction_cachep, transaction);
Chris Mason's avatar
Chris Mason committed
40
	}
Chris Mason's avatar
Chris Mason committed
41
42
}

Chris Mason's avatar
Chris Mason committed
43
44
45
/*
 * either allocate a new transaction or hop into the existing one
 */
46
static noinline int join_transaction(struct btrfs_root *root)
Chris Mason's avatar
Chris Mason committed
47
48
49
50
{
	struct btrfs_transaction *cur_trans;
	cur_trans = root->fs_info->running_transaction;
	if (!cur_trans) {
Chris Mason's avatar
Chris Mason committed
51
52
		cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
					     GFP_NOFS);
Chris Mason's avatar
Chris Mason committed
53
		BUG_ON(!cur_trans);
54
		root->fs_info->generation++;
55
56
		cur_trans->num_writers = 1;
		cur_trans->num_joined = 0;
57
		cur_trans->transid = root->fs_info->generation;
Chris Mason's avatar
Chris Mason committed
58
59
60
		init_waitqueue_head(&cur_trans->writer_wait);
		init_waitqueue_head(&cur_trans->commit_wait);
		cur_trans->in_commit = 0;
61
		cur_trans->blocked = 0;
62
		cur_trans->use_count = 1;
Chris Mason's avatar
Chris Mason committed
63
		cur_trans->commit_done = 0;
Chris Mason's avatar
Chris Mason committed
64
		cur_trans->start_time = get_seconds();
65
66
67

		cur_trans->delayed_refs.root.rb_node = NULL;
		cur_trans->delayed_refs.num_entries = 0;
68
69
		cur_trans->delayed_refs.num_heads_ready = 0;
		cur_trans->delayed_refs.num_heads = 0;
70
		cur_trans->delayed_refs.flushing = 0;
71
		cur_trans->delayed_refs.run_delayed_start = 0;
72
73
		spin_lock_init(&cur_trans->delayed_refs.lock);

74
		INIT_LIST_HEAD(&cur_trans->pending_snapshots);
Chris Mason's avatar
Chris Mason committed
75
		list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
76
		extent_io_tree_init(&cur_trans->dirty_pages,
77
78
				     root->fs_info->btree_inode->i_mapping,
				     GFP_NOFS);
79
80
81
		spin_lock(&root->fs_info->new_trans_lock);
		root->fs_info->running_transaction = cur_trans;
		spin_unlock(&root->fs_info->new_trans_lock);
82
83
84
	} else {
		cur_trans->num_writers++;
		cur_trans->num_joined++;
Chris Mason's avatar
Chris Mason committed
85
	}
86

Chris Mason's avatar
Chris Mason committed
87
88
89
	return 0;
}

Chris Mason's avatar
Chris Mason committed
90
/*
91
92
93
94
 * this does all the record keeping required to make sure that a reference
 * counted root is properly recorded in a given transaction.  This is required
 * to make sure the old root from before we joined the transaction is deleted
 * when the transaction commits
Chris Mason's avatar
Chris Mason committed
95
 */
96
97
static noinline int record_root_in_trans(struct btrfs_trans_handle *trans,
					 struct btrfs_root *root)
98
{
99
	if (root->ref_cows && root->last_trans < trans->transid) {
100
		WARN_ON(root == root->fs_info->extent_root);
101
102
103
104
105
106
107
108
109
110
111
		WARN_ON(root->root_item.refs == 0);
		WARN_ON(root->commit_root != root->node);

		radix_tree_tag_set(&root->fs_info->fs_roots_radix,
			   (unsigned long)root->root_key.objectid,
			   BTRFS_ROOT_TRANS_TAG);
		root->last_trans = trans->transid;
		btrfs_init_reloc_root(trans, root);
	}
	return 0;
}
112

113
114
115
116
117
int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
			       struct btrfs_root *root)
{
	if (!root->ref_cows)
		return 0;
118

119
120
121
122
	mutex_lock(&root->fs_info->trans_mutex);
	if (root->last_trans == trans->transid) {
		mutex_unlock(&root->fs_info->trans_mutex);
		return 0;
123
	}
124
125
126

	record_root_in_trans(trans, root);
	mutex_unlock(&root->fs_info->trans_mutex);
127
128
129
	return 0;
}

Chris Mason's avatar
Chris Mason committed
130
131
132
133
/* wait for commit against the current transaction to become unblocked
 * when this is done, it is safe to start a new transaction, but the current
 * transaction might not be fully on disk.
 */
Chris Mason's avatar
Chris Mason committed
134
static void wait_current_trans(struct btrfs_root *root)
Chris Mason's avatar
Chris Mason committed
135
{
136
	struct btrfs_transaction *cur_trans;
Chris Mason's avatar
Chris Mason committed
137

138
	cur_trans = root->fs_info->running_transaction;
Chris Mason's avatar
Chris Mason committed
139
	if (cur_trans && cur_trans->blocked) {
140
141
		DEFINE_WAIT(wait);
		cur_trans->use_count++;
142
		while (1) {
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
			prepare_to_wait(&root->fs_info->transaction_wait, &wait,
					TASK_UNINTERRUPTIBLE);
			if (cur_trans->blocked) {
				mutex_unlock(&root->fs_info->trans_mutex);
				schedule();
				mutex_lock(&root->fs_info->trans_mutex);
				finish_wait(&root->fs_info->transaction_wait,
					    &wait);
			} else {
				finish_wait(&root->fs_info->transaction_wait,
					    &wait);
				break;
			}
		}
		put_transaction(cur_trans);
	}
Chris Mason's avatar
Chris Mason committed
159
160
}

161
static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
162
					     int num_blocks, int wait)
Chris Mason's avatar
Chris Mason committed
163
164
165
166
167
168
{
	struct btrfs_trans_handle *h =
		kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
	int ret;

	mutex_lock(&root->fs_info->trans_mutex);
Chris Mason's avatar
Chris Mason committed
169
170
	if (!root->fs_info->log_root_recovering &&
	    ((wait == 1 && !root->fs_info->open_ioctl_trans) || wait == 2))
Chris Mason's avatar
Chris Mason committed
171
		wait_current_trans(root);
Chris Mason's avatar
Chris Mason committed
172
173
	ret = join_transaction(root);
	BUG_ON(ret);
174

175
	h->transid = root->fs_info->running_transaction->transid;
Chris Mason's avatar
Chris Mason committed
176
177
178
	h->transaction = root->fs_info->running_transaction;
	h->blocks_reserved = num_blocks;
	h->blocks_used = 0;
179
	h->block_group = 0;
180
181
	h->alloc_exclude_nr = 0;
	h->alloc_exclude_start = 0;
182
	h->delayed_ref_updates = 0;
183

Chris Mason's avatar
Chris Mason committed
184
	root->fs_info->running_transaction->use_count++;
185
	record_root_in_trans(h, root);
Chris Mason's avatar
Chris Mason committed
186
187
188
189
	mutex_unlock(&root->fs_info->trans_mutex);
	return h;
}

190
191
192
struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
						   int num_blocks)
{
193
	return start_transaction(root, num_blocks, 1);
194
195
196
197
}
struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
						   int num_blocks)
{
198
	return start_transaction(root, num_blocks, 0);
199
200
}

201
202
203
204
205
206
struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
							 int num_blocks)
{
	return start_transaction(r, num_blocks, 2);
}

Chris Mason's avatar
Chris Mason committed
207
/* wait for a transaction commit to be fully complete */
208
209
210
211
212
static noinline int wait_for_commit(struct btrfs_root *root,
				    struct btrfs_transaction *commit)
{
	DEFINE_WAIT(wait);
	mutex_lock(&root->fs_info->trans_mutex);
213
	while (!commit->commit_done) {
214
215
216
217
218
219
220
221
222
223
224
225
226
		prepare_to_wait(&commit->commit_wait, &wait,
				TASK_UNINTERRUPTIBLE);
		if (commit->commit_done)
			break;
		mutex_unlock(&root->fs_info->trans_mutex);
		schedule();
		mutex_lock(&root->fs_info->trans_mutex);
	}
	mutex_unlock(&root->fs_info->trans_mutex);
	finish_wait(&commit->commit_wait, &wait);
	return 0;
}

227
#if 0
Chris Mason's avatar
Chris Mason committed
228
/*
229
230
 * rate limit against the drop_snapshot code.  This helps to slow down new
 * operations if the drop_snapshot code isn't able to keep up.
Chris Mason's avatar
Chris Mason committed
231
 */
Chris Mason's avatar
Chris Mason committed
232
static void throttle_on_drops(struct btrfs_root *root)
233
234
{
	struct btrfs_fs_info *info = root->fs_info;
Chris Mason's avatar
Chris Mason committed
235
	int harder_count = 0;
236

Chris Mason's avatar
Chris Mason committed
237
harder:
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
	if (atomic_read(&info->throttles)) {
		DEFINE_WAIT(wait);
		int thr;
		thr = atomic_read(&info->throttle_gen);

		do {
			prepare_to_wait(&info->transaction_throttle,
					&wait, TASK_UNINTERRUPTIBLE);
			if (!atomic_read(&info->throttles)) {
				finish_wait(&info->transaction_throttle, &wait);
				break;
			}
			schedule();
			finish_wait(&info->transaction_throttle, &wait);
		} while (thr == atomic_read(&info->throttle_gen));
Chris Mason's avatar
Chris Mason committed
253
254
255
256
257
258
259
260
261
262
263
264
265
		harder_count++;

		if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 &&
		    harder_count < 2)
			goto harder;

		if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 &&
		    harder_count < 10)
			goto harder;

		if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 &&
		    harder_count < 20)
			goto harder;
266
267
	}
}
268
#endif
269

Chris Mason's avatar
Chris Mason committed
270
271
272
void btrfs_throttle(struct btrfs_root *root)
{
	mutex_lock(&root->fs_info->trans_mutex);
273
274
	if (!root->fs_info->open_ioctl_trans)
		wait_current_trans(root);
Chris Mason's avatar
Chris Mason committed
275
276
277
	mutex_unlock(&root->fs_info->trans_mutex);
}

278
279
static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
			  struct btrfs_root *root, int throttle)
Chris Mason's avatar
Chris Mason committed
280
281
{
	struct btrfs_transaction *cur_trans;
282
	struct btrfs_fs_info *info = root->fs_info;
283
284
285
286
287
288
289
290
	int count = 0;

	while (count < 4) {
		unsigned long cur = trans->delayed_ref_updates;
		trans->delayed_ref_updates = 0;
		if (cur &&
		    trans->transaction->delayed_refs.num_heads_ready > 64) {
			trans->delayed_ref_updates = 0;
291
292
293
294
295
296
297

			/*
			 * do a full flush if the transaction is trying
			 * to close
			 */
			if (trans->transaction->delayed_refs.flushing)
				cur = 0;
298
299
300
301
302
			btrfs_run_delayed_refs(trans, root, cur);
		} else {
			break;
		}
		count++;
303
304
	}

305
306
	mutex_lock(&info->trans_mutex);
	cur_trans = info->running_transaction;
Chris Mason's avatar
Chris Mason committed
307
	WARN_ON(cur_trans != trans->transaction);
308
	WARN_ON(cur_trans->num_writers < 1);
Chris Mason's avatar
Chris Mason committed
309
	cur_trans->num_writers--;
310

Chris Mason's avatar
Chris Mason committed
311
312
313
	if (waitqueue_active(&cur_trans->writer_wait))
		wake_up(&cur_trans->writer_wait);
	put_transaction(cur_trans);
314
	mutex_unlock(&info->trans_mutex);
Chris Mason's avatar
Chris Mason committed
315
	memset(trans, 0, sizeof(*trans));
Chris Mason's avatar
Chris Mason committed
316
	kmem_cache_free(btrfs_trans_handle_cachep, trans);
317

Chris Mason's avatar
Chris Mason committed
318
319
320
	return 0;
}

321
322
323
324
325
326
327
328
329
330
331
332
int btrfs_end_transaction(struct btrfs_trans_handle *trans,
			  struct btrfs_root *root)
{
	return __btrfs_end_transaction(trans, root, 0);
}

int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
				   struct btrfs_root *root)
{
	return __btrfs_end_transaction(trans, root, 1);
}

Chris Mason's avatar
Chris Mason committed
333
334
335
336
337
/*
 * when btree blocks are allocated, they have some corresponding bits set for
 * them in one of two extent_io trees.  This is used to make sure all of
 * those extents are on disk for transaction or log commit
 */
338
339
int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
					struct extent_io_tree *dirty_pages)
Chris Mason's avatar
Chris Mason committed
340
{
341
	int ret;
342
	int err = 0;
343
344
345
	int werr = 0;
	struct page *page;
	struct inode *btree_inode = root->fs_info->btree_inode;
346
	u64 start = 0;
347
348
	u64 end;
	unsigned long index;
349

350
	while (1) {
351
		ret = find_first_extent_bit(dirty_pages, start, &start, &end,
352
353
					    EXTENT_DIRTY);
		if (ret)
354
			break;
355
		while (start <= end) {
356
357
			cond_resched();

358
			index = start >> PAGE_CACHE_SHIFT;
359
			start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
Chris Mason's avatar
Chris Mason committed
360
			page = find_get_page(btree_inode->i_mapping, index);
361
362
			if (!page)
				continue;
Chris Mason's avatar
Chris Mason committed
363
364
365
366
367
368
369
370

			btree_lock_page_hook(page);
			if (!page->mapping) {
				unlock_page(page);
				page_cache_release(page);
				continue;
			}

371
372
373
374
375
376
377
378
379
			if (PageWriteback(page)) {
				if (PageDirty(page))
					wait_on_page_writeback(page);
				else {
					unlock_page(page);
					page_cache_release(page);
					continue;
				}
			}
380
381
382
383
384
385
			err = write_one_page(page, 0);
			if (err)
				werr = err;
			page_cache_release(page);
		}
	}
386
	while (1) {
387
388
389
390
391
392
		ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
					    EXTENT_DIRTY);
		if (ret)
			break;

		clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
393
		while (start <= end) {
394
395
396
397
398
399
			index = start >> PAGE_CACHE_SHIFT;
			start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
			page = find_get_page(btree_inode->i_mapping, index);
			if (!page)
				continue;
			if (PageDirty(page)) {
Chris Mason's avatar
Chris Mason committed
400
401
				btree_lock_page_hook(page);
				wait_on_page_writeback(page);
402
403
404
405
				err = write_one_page(page, 0);
				if (err)
					werr = err;
			}
406
			wait_on_page_writeback(page);
407
408
409
410
			page_cache_release(page);
			cond_resched();
		}
	}
411
412
413
	if (err)
		werr = err;
	return werr;
Chris Mason's avatar
Chris Mason committed
414
415
}

416
417
418
419
420
421
422
423
424
425
426
427
int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
				     struct btrfs_root *root)
{
	if (!trans || !trans->transaction) {
		struct inode *btree_inode;
		btree_inode = root->fs_info->btree_inode;
		return filemap_write_and_wait(btree_inode->i_mapping);
	}
	return btrfs_write_and_wait_marked_extents(root,
					   &trans->transaction->dirty_pages);
}

Chris Mason's avatar
Chris Mason committed
428
429
430
431
432
433
434
435
436
437
/*
 * this is used to update the root pointer in the tree of tree roots.
 *
 * But, in the case of the extent allocation tree, updating the root
 * pointer may allocate blocks which may change the root of the extent
 * allocation tree.
 *
 * So, this loops and repeats and makes sure the cowonly root didn't
 * change while the root pointer was being updated in the metadata.
 */
438
439
static int update_cowonly_root(struct btrfs_trans_handle *trans,
			       struct btrfs_root *root)
Chris Mason's avatar
Chris Mason committed
440
441
{
	int ret;
442
443
	u64 old_root_bytenr;
	struct btrfs_root *tree_root = root->fs_info->tree_root;
Chris Mason's avatar
Chris Mason committed
444

445
	btrfs_write_dirty_block_groups(trans, root);
446
447
448

	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
	BUG_ON(ret);
449

450
	while (1) {
451
452
		old_root_bytenr = btrfs_root_bytenr(&root->root_item);
		if (old_root_bytenr == root->node->start)
Chris Mason's avatar
Chris Mason committed
453
			break;
454

455
		btrfs_set_root_node(&root->root_item, root->node);
Chris Mason's avatar
Chris Mason committed
456
		ret = btrfs_update_root(trans, tree_root,
457
458
					&root->root_key,
					&root->root_item);
Chris Mason's avatar
Chris Mason committed
459
		BUG_ON(ret);
460
		btrfs_write_dirty_block_groups(trans, root);
461
462
463

		ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
		BUG_ON(ret);
464
	}
465
466
	free_extent_buffer(root->commit_root);
	root->commit_root = btrfs_root_node(root);
467
468
469
	return 0;
}

Chris Mason's avatar
Chris Mason committed
470
471
472
/*
 * update all the cowonly tree roots on disk
 */
473
474
static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
					 struct btrfs_root *root)
475
476
477
{
	struct btrfs_fs_info *fs_info = root->fs_info;
	struct list_head *next;
478
	struct extent_buffer *eb;
479
	int ret;
480

481
482
	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
	BUG_ON(ret);
483

484
	eb = btrfs_lock_root_node(fs_info->tree_root);
485
	btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb);
486
487
	btrfs_tree_unlock(eb);
	free_extent_buffer(eb);
488

489
490
	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
	BUG_ON(ret);
491

492
	while (!list_empty(&fs_info->dirty_cowonly_roots)) {
493
494
495
		next = fs_info->dirty_cowonly_roots.next;
		list_del_init(next);
		root = list_entry(next, struct btrfs_root, dirty_list);
496

497
		update_cowonly_root(trans, root);
498
499
500

		ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
		BUG_ON(ret);
Chris Mason's avatar
Chris Mason committed
501
502
503
504
	}
	return 0;
}

Chris Mason's avatar
Chris Mason committed
505
506
507
508
509
/*
 * dead roots are old snapshots that need to be deleted.  This allocates
 * a dirty root struct and adds it into the list of dead roots that need to
 * be deleted
 */
510
int btrfs_add_dead_root(struct btrfs_root *root)
511
{
Yan Zheng's avatar
Yan Zheng committed
512
	mutex_lock(&root->fs_info->trans_mutex);
513
	list_add(&root->root_list, &root->fs_info->dead_roots);
Yan Zheng's avatar
Yan Zheng committed
514
	mutex_unlock(&root->fs_info->trans_mutex);
515
516
517
	return 0;
}

Chris Mason's avatar
Chris Mason committed
518
/*
519
 * update all the cowonly tree roots on disk
Chris Mason's avatar
Chris Mason committed
520
 */
521
522
static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
				    struct btrfs_root *root)
523
524
{
	struct btrfs_root *gang[8];
525
	struct btrfs_fs_info *fs_info = root->fs_info;
526
527
	int i;
	int ret;
528
529
	int err = 0;

530
	while (1) {
531
532
		ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
						 (void **)gang, 0,
533
534
535
536
537
538
						 ARRAY_SIZE(gang),
						 BTRFS_ROOT_TRANS_TAG);
		if (ret == 0)
			break;
		for (i = 0; i < ret; i++) {
			root = gang[i];
539
540
541
			radix_tree_tag_clear(&fs_info->fs_roots_radix,
					(unsigned long)root->root_key.objectid,
					BTRFS_ROOT_TRANS_TAG);
Yan Zheng's avatar
Yan Zheng committed
542

543
			btrfs_free_log(trans, root);
544
			btrfs_update_reloc_root(trans, root);
545

546
547
548
549
550
551
			if (root->commit_root != root->node) {
				free_extent_buffer(root->commit_root);
				root->commit_root = btrfs_root_node(root);
				btrfs_set_root_node(&root->root_item,
						    root->node);
			}
552
553

			err = btrfs_update_root(trans, fs_info->tree_root,
554
555
						&root->root_key,
						&root->root_item);
556
557
			if (err)
				break;
558
559
		}
	}
560
	return err;
561
562
}

Chris Mason's avatar
Chris Mason committed
563
564
565
566
/*
 * defrag a given btree.  If cacheonly == 1, this won't read from the disk,
 * otherwise every leaf in the btree is read and defragged.
 */
567
568
569
570
571
int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
{
	struct btrfs_fs_info *info = root->fs_info;
	int ret;
	struct btrfs_trans_handle *trans;
572
	unsigned long nr;
573

574
	smp_mb();
575
576
577
	if (root->defrag_running)
		return 0;
	trans = btrfs_start_transaction(root, 1);
578
	while (1) {
579
580
		root->defrag_running = 1;
		ret = btrfs_defrag_leaves(trans, root, cacheonly);
581
		nr = trans->blocks_used;
582
		btrfs_end_transaction(trans, root);
583
		btrfs_btree_balance_dirty(info->tree_root, nr);
584
585
586
		cond_resched();

		trans = btrfs_start_transaction(root, 1);
587
		if (root->fs_info->closing || ret != -EAGAIN)
588
589
590
			break;
	}
	root->defrag_running = 0;
591
	smp_mb();
592
593
594
595
	btrfs_end_transaction(trans, root);
	return 0;
}

596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
/*
 * when dropping snapshots, we generate a ton of delayed refs, and it makes
 * sense not to join the transaction while it is trying to flush the current
 * queue of delayed refs out.
 *
 * This is used by the drop snapshot code only
 */
static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info)
{
	DEFINE_WAIT(wait);

	mutex_lock(&info->trans_mutex);
	while (info->running_transaction &&
	       info->running_transaction->delayed_refs.flushing) {
		prepare_to_wait(&info->transaction_wait, &wait,
				TASK_UNINTERRUPTIBLE);
		mutex_unlock(&info->trans_mutex);
613

614
		schedule();
615

616
617
618
619
620
621
622
		mutex_lock(&info->trans_mutex);
		finish_wait(&info->transaction_wait, &wait);
	}
	mutex_unlock(&info->trans_mutex);
	return 0;
}

Chris Mason's avatar
Chris Mason committed
623
624
625
626
/*
 * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
 * all of them
 */
627
int btrfs_drop_dead_root(struct btrfs_root *root)
628
629
{
	struct btrfs_trans_handle *trans;
630
	struct btrfs_root *tree_root = root->fs_info->tree_root;
631
	unsigned long nr;
632
	int ret;
633

634
635
636
637
638
639
640
	while (1) {
		/*
		 * we don't want to jump in and create a bunch of
		 * delayed refs if the transaction is starting to close
		 */
		wait_transaction_pre_flush(tree_root->fs_info);
		trans = btrfs_start_transaction(tree_root, 1);
641

642
643
644
645
646
647
648
		/*
		 * we've joined a transaction, make sure it isn't
		 * closing right now
		 */
		if (trans->transaction->delayed_refs.flushing) {
			btrfs_end_transaction(trans, tree_root);
			continue;
649
		}
650

651
652
653
		ret = btrfs_drop_snapshot(trans, root);
		if (ret != -EAGAIN)
			break;
654

655
656
657
658
		ret = btrfs_update_root(trans, tree_root,
					&root->root_key,
					&root->root_item);
		if (ret)
659
			break;
660

661
		nr = trans->blocks_used;
662
663
		ret = btrfs_end_transaction(trans, tree_root);
		BUG_ON(ret);
664

665
		btrfs_btree_balance_dirty(tree_root, nr);
666
		cond_resched();
667
	}
668
669
670
671
672
673
674
675
676
677
678
679
680
681
	BUG_ON(ret);

	ret = btrfs_del_root(trans, tree_root, &root->root_key);
	BUG_ON(ret);

	nr = trans->blocks_used;
	ret = btrfs_end_transaction(trans, tree_root);
	BUG_ON(ret);

	free_extent_buffer(root->node);
	free_extent_buffer(root->commit_root);
	kfree(root);

	btrfs_btree_balance_dirty(tree_root, nr);
682
	return ret;
683
684
}

Chris Mason's avatar
Chris Mason committed
685
686
687
688
/*
 * new snapshots need to be created at a very specific time in the
 * transaction commit.  This does the actual creation
 */
689
static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
690
691
692
693
				   struct btrfs_fs_info *fs_info,
				   struct btrfs_pending_snapshot *pending)
{
	struct btrfs_key key;
694
	struct btrfs_root_item *new_root_item;
695
696
697
	struct btrfs_root *tree_root = fs_info->tree_root;
	struct btrfs_root *root = pending->root;
	struct extent_buffer *tmp;
698
	struct extent_buffer *old;
699
700
701
	int ret;
	u64 objectid;

702
703
704
705
706
	new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
	if (!new_root_item) {
		ret = -ENOMEM;
		goto fail;
	}
707
708
709
710
	ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
	if (ret)
		goto fail;

711
	record_root_in_trans(trans, root);
Yan Zheng's avatar
Yan Zheng committed
712
	btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
713
	memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
714
715

	key.objectid = objectid;
716
	key.offset = 0;
717
718
	btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);

719
	old = btrfs_lock_root_node(root);
720
	btrfs_cow_block(trans, root, old, NULL, 0, &old);
721
	btrfs_set_lock_blocking(old);
722

723
724
725
	btrfs_copy_root(trans, root, old, &tmp, objectid);
	btrfs_tree_unlock(old);
	free_extent_buffer(old);
726

727
	btrfs_set_root_node(new_root_item, tmp);
728
	ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
729
				new_root_item);
730
	btrfs_tree_unlock(tmp);
731
732
733
734
	free_extent_buffer(tmp);
	if (ret)
		goto fail;

735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
	key.offset = (u64)-1;
	memcpy(&pending->root_key, &key, sizeof(key));
fail:
	kfree(new_root_item);
	return ret;
}

static noinline int finish_pending_snapshot(struct btrfs_fs_info *fs_info,
				   struct btrfs_pending_snapshot *pending)
{
	int ret;
	int namelen;
	u64 index = 0;
	struct btrfs_trans_handle *trans;
	struct inode *parent_inode;
	struct inode *inode;
751
	struct btrfs_root *parent_root;
752

753
	parent_inode = pending->dentry->d_parent->d_inode;
754
	parent_root = BTRFS_I(parent_inode)->root;
755
	trans = btrfs_join_transaction(parent_root, 1);
756

757
758
759
	/*
	 * insert the directory item
	 */
760
	namelen = strlen(pending->name);
761
	ret = btrfs_set_inode_index(parent_inode, &index);
762
	ret = btrfs_insert_dir_item(trans, parent_root,
763
764
765
			    pending->name, namelen,
			    parent_inode->i_ino,
			    &pending->root_key, BTRFS_FT_DIR, index);
766
767
768

	if (ret)
		goto fail;
769

770
771
772
773
	btrfs_i_size_write(parent_inode, parent_inode->i_size + namelen * 2);
	ret = btrfs_update_inode(trans, parent_root, parent_inode);
	BUG_ON(ret);

774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
	/* add the backref first */
	ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root,
				 pending->root_key.objectid,
				 BTRFS_ROOT_BACKREF_KEY,
				 parent_root->root_key.objectid,
				 parent_inode->i_ino, index, pending->name,
				 namelen);

	BUG_ON(ret);

	/* now add the forward ref */
	ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root,
				 parent_root->root_key.objectid,
				 BTRFS_ROOT_REF_KEY,
				 pending->root_key.objectid,
				 parent_inode->i_ino, index, pending->name,
				 namelen);

792
793
	inode = btrfs_lookup_dentry(parent_inode, pending->dentry);
	d_instantiate(pending->dentry, inode);
794
fail:
795
	btrfs_end_transaction(trans, fs_info->fs_root);
796
797
798
	return ret;
}

Chris Mason's avatar
Chris Mason committed
799
800
801
/*
 * create all the snapshots we've scheduled for creation
 */
802
803
static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
					     struct btrfs_fs_info *fs_info)
804
805
806
807
808
{
	struct btrfs_pending_snapshot *pending;
	struct list_head *head = &trans->transaction->pending_snapshots;
	int ret;

809
	list_for_each_entry(pending, head, list) {
810
811
812
813
814
815
816
817
		ret = create_pending_snapshot(trans, fs_info, pending);
		BUG_ON(ret);
	}
	return 0;
}

static noinline int finish_pending_snapshots(struct btrfs_trans_handle *trans,
					     struct btrfs_fs_info *fs_info)
818
819
820
821
822
{
	struct btrfs_pending_snapshot *pending;
	struct list_head *head = &trans->transaction->pending_snapshots;
	int ret;

823
	while (!list_empty(head)) {
824
825
		pending = list_entry(head->next,
				     struct btrfs_pending_snapshot, list);
826
		ret = finish_pending_snapshot(fs_info, pending);
827
828
829
830
831
		BUG_ON(ret);
		list_del(&pending->list);
		kfree(pending->name);
		kfree(pending);
	}
Chris Mason's avatar
Chris Mason committed
832
833
834
	return 0;
}

835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
static void update_super_roots(struct btrfs_root *root)
{
	struct btrfs_root_item *root_item;
	struct btrfs_super_block *super;

	super = &root->fs_info->super_copy;

	root_item = &root->fs_info->chunk_root->root_item;
	super->chunk_root = root_item->bytenr;
	super->chunk_root_generation = root_item->generation;
	super->chunk_root_level = root_item->level;

	root_item = &root->fs_info->tree_root->root_item;
	super->root = root_item->bytenr;
	super->generation = root_item->generation;
	super->root_level = root_item->level;
}

Chris Mason's avatar
Chris Mason committed
853
854
855
int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
			     struct btrfs_root *root)
{
856
857
	unsigned long joined = 0;
	unsigned long timeout = 1;
Chris Mason's avatar
Chris Mason committed
858
	struct btrfs_transaction *cur_trans;
Chris Mason's avatar
Chris Mason committed
859
	struct btrfs_transaction *prev_trans = NULL;
860
	struct extent_io_tree *pinned_copy;
Chris Mason's avatar
Chris Mason committed
861
	DEFINE_WAIT(wait);
862
	int ret;
863
864
	int should_grow = 0;
	unsigned long now = get_seconds();
865
	int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
Chris Mason's avatar
Chris Mason committed
866

867
868
	btrfs_run_ordered_operations(root, 0);

869
870
871
872
873
874
	/* make a pass through all the delayed refs we have so far
	 * any runnings procs may add more while we are here
	 */
	ret = btrfs_run_delayed_refs(trans, root, 0);
	BUG_ON(ret);

875
	cur_trans = trans->transaction;
876
877
878
879
	/*
	 * set the flushing flag so procs in this transaction have to
	 * start sending their work down.
	 */
880
	cur_trans->delayed_refs.flushing = 1;
881

882
	ret = btrfs_run_delayed_refs(trans, root, 0);
883
884
	BUG_ON(ret);

Chris Mason's avatar
Chris Mason committed
885
	mutex_lock(&root->fs_info->trans_mutex);
886
887
	if (cur_trans->in_commit) {
		cur_trans->use_count++;
Chris Mason's avatar
Chris Mason committed
888
		mutex_unlock(&root->fs_info->trans_mutex);
Chris Mason's avatar
Chris Mason committed
889
		btrfs_end_transaction(trans, root);
Chris Mason's avatar
Chris Mason committed
890

Chris Mason's avatar
Chris Mason committed
891
892
		ret = wait_for_commit(root, cur_trans);
		BUG_ON(ret);
893
894

		mutex_lock(&root->fs_info->trans_mutex);
Chris Mason's avatar
Chris Mason committed
895
		put_transaction(cur_trans);
896
897
		mutex_unlock(&root->fs_info->trans_mutex);

Chris Mason's avatar
Chris Mason committed
898
899
		return 0;
	}
900
901
902
903
904

	pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
	if (!pinned_copy)
		return -ENOMEM;

905
	extent_io_tree_init(pinned_copy,
906
907
			     root->fs_info->btree_inode->i_mapping, GFP_NOFS);

Chris Mason's avatar
Chris Mason committed
908
	trans->transaction->in_commit = 1;
909
	trans->transaction->blocked = 1;
Chris Mason's avatar
Chris Mason committed
910
911
912
913
914
915
916
917
918
919
	if (cur_trans->list.prev != &root->fs_info->trans_list) {
		prev_trans = list_entry(cur_trans->list.prev,
					struct btrfs_transaction, list);
		if (!prev_trans->commit_done) {
			prev_trans->use_count++;
			mutex_unlock(&root->fs_info->trans_mutex);

			wait_for_commit(root, prev_trans);

			mutex_lock(&root->fs_info->trans_mutex);
920
			put_transaction(prev_trans);
Chris Mason's avatar
Chris Mason committed
921
922
		}
	}
923

924
925
926
	if (now < cur_trans->start_time || now - cur_trans->start_time < 1)
		should_grow = 1;

927
	do {
928
		int snap_pending = 0;
929
		joined = cur_trans->num_joined;
930
931
932
		if (!list_empty(&trans->transaction->pending_snapshots))
			snap_pending = 1;

Chris Mason's avatar
Chris Mason committed
933
		WARN_ON(cur_trans != trans->transaction);
934
		prepare_to_wait(&cur_trans->writer_wait, &wait,
Chris Mason's avatar
Chris Mason committed
935
				TASK_UNINTERRUPTIBLE);
936
937
938

		if (cur_trans->num_writers > 1)
			timeout = MAX_SCHEDULE_TIMEOUT;
939
		else if (should_grow)
940
941
			timeout = 1;

Chris Mason's avatar
Chris Mason committed
942
		mutex_unlock(&root->fs_info->trans_mutex);
943

944
945
946
		if (flush_on_commit || snap_pending) {
			if (flush_on_commit)
				btrfs_start_delalloc_inodes(root);
947
948
949
950
			ret = btrfs_wait_ordered_extents(root, 1);
			BUG_ON(ret);
		}

951
952
953
954
955
956
957
958
959
		/*
		 * rename don't use btrfs_join_transaction, so, once we
		 * set the transaction to blocked above, we aren't going
		 * to get any new ordered operations.  We can safely run
		 * it here and no for sure that nothing new will be added
		 * to the list
		 */
		btrfs_run_ordered_operations(root, 1);

960
961
962
		smp_mb();
		if (cur_trans->num_writers > 1 || should_grow)
			schedule_timeout(timeout);
963

Chris Mason's avatar
Chris Mason committed
964
		mutex_lock(&root->fs_info->trans_mutex);
965
966
		finish_wait(&cur_trans->writer_wait, &wait);
	} while (cur_trans->num_writers > 1 ||
967
		 (should_grow && cur_trans->num_joined != joined));
968

969
970
971
	ret = create_pending_snapshots(trans, root->fs_info);
	BUG_ON(ret);

972
973
974
	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
	BUG_ON(ret);

Chris Mason's avatar
Chris Mason committed
975
	WARN_ON(cur_trans != trans->transaction);
Chris Mason's avatar
Chris Mason committed
976

977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
	/* btrfs_commit_tree_roots is responsible for getting the
	 * various roots consistent with each other.  Every pointer
	 * in the tree of tree roots has to point to the most up to date
	 * root for every subvolume and other tree.  So, we have to keep
	 * the tree logging code from jumping in and changing any
	 * of the trees.
	 *
	 * At this point in the commit, there can't be any tree-log
	 * writers, but a little lower down we drop the trans mutex
	 * and let new people in.  By holding the tree_log_mutex
	 * from now until after the super is written, we avoid races
	 * with the tree-log code.
	 */
	mutex_lock(&root->fs_info->tree_log_mutex);

992
	ret = commit_fs_roots(trans, root);
993
994
	BUG_ON(ret);

995
	/* commit_fs_roots gets rid of all the tree log roots, it is now
996
997
998
999
	 * safe to free the root of tree log roots
	 */
	btrfs_free_log_root_tree(trans, root->fs_info);

1000
	ret = commit_cowonly_roots(trans, root);
For faster browsing, not all history is shown. View entire blame