transaction.c 37.3 KB
Newer Older
C
Chris Mason 已提交
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.
 */

C
Chris Mason 已提交
19
#include <linux/fs.h>
20
#include <linux/slab.h>
C
Chris Mason 已提交
21
#include <linux/sched.h>
22
#include <linux/writeback.h>
23
#include <linux/pagemap.h>
24
#include <linux/blkdev.h>
C
Chris Mason 已提交
25 26 27
#include "ctree.h"
#include "disk-io.h"
#include "transaction.h"
28
#include "locking.h"
29
#include "tree-log.h"
30
#include "inode-map.h"
C
Chris Mason 已提交
31

32 33
#define BTRFS_ROOT_TRANS_TAG 0

34
static noinline void put_transaction(struct btrfs_transaction *transaction)
C
Chris Mason 已提交
35
{
36 37
	WARN_ON(atomic_read(&transaction->use_count) == 0);
	if (atomic_dec_and_test(&transaction->use_count)) {
C
Chris Mason 已提交
38 39
		memset(transaction, 0, sizeof(*transaction));
		kmem_cache_free(btrfs_transaction_cachep, transaction);
C
Chris Mason 已提交
40
	}
C
Chris Mason 已提交
41 42
}

J
Josef Bacik 已提交
43 44 45 46 47 48
static noinline void switch_commit_root(struct btrfs_root *root)
{
	free_extent_buffer(root->commit_root);
	root->commit_root = btrfs_root_node(root);
}

C
Chris Mason 已提交
49 50 51
/*
 * either allocate a new transaction or hop into the existing one
 */
52
static noinline int join_transaction(struct btrfs_root *root)
C
Chris Mason 已提交
53 54 55 56
{
	struct btrfs_transaction *cur_trans;
	cur_trans = root->fs_info->running_transaction;
	if (!cur_trans) {
C
Chris Mason 已提交
57 58
		cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
					     GFP_NOFS);
T
Tsutomu Itoh 已提交
59 60
		if (!cur_trans)
			return -ENOMEM;
61
		root->fs_info->generation++;
62
		atomic_set(&cur_trans->num_writers, 1);
63
		cur_trans->num_joined = 0;
64
		cur_trans->transid = root->fs_info->generation;
C
Chris Mason 已提交
65 66 67
		init_waitqueue_head(&cur_trans->writer_wait);
		init_waitqueue_head(&cur_trans->commit_wait);
		cur_trans->in_commit = 0;
68
		cur_trans->blocked = 0;
69
		atomic_set(&cur_trans->use_count, 1);
C
Chris Mason 已提交
70
		cur_trans->commit_done = 0;
C
Chris Mason 已提交
71
		cur_trans->start_time = get_seconds();
72

73
		cur_trans->delayed_refs.root = RB_ROOT;
74
		cur_trans->delayed_refs.num_entries = 0;
75 76
		cur_trans->delayed_refs.num_heads_ready = 0;
		cur_trans->delayed_refs.num_heads = 0;
77
		cur_trans->delayed_refs.flushing = 0;
78
		cur_trans->delayed_refs.run_delayed_start = 0;
79 80
		spin_lock_init(&cur_trans->delayed_refs.lock);

81
		INIT_LIST_HEAD(&cur_trans->pending_snapshots);
C
Chris Mason 已提交
82
		list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
83
		extent_io_tree_init(&cur_trans->dirty_pages,
84 85
				     root->fs_info->btree_inode->i_mapping,
				     GFP_NOFS);
86 87 88
		spin_lock(&root->fs_info->new_trans_lock);
		root->fs_info->running_transaction = cur_trans;
		spin_unlock(&root->fs_info->new_trans_lock);
89
	} else {
90
		atomic_inc(&cur_trans->num_writers);
91
		cur_trans->num_joined++;
C
Chris Mason 已提交
92
	}
93

C
Chris Mason 已提交
94 95 96
	return 0;
}

C
Chris Mason 已提交
97
/*
C
Chris Mason 已提交
98 99 100 101
 * 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
C
Chris Mason 已提交
102
 */
103 104
static noinline int record_root_in_trans(struct btrfs_trans_handle *trans,
					 struct btrfs_root *root)
105
{
106
	if (root->ref_cows && root->last_trans < trans->transid) {
107
		WARN_ON(root == root->fs_info->extent_root);
108 109 110 111 112 113 114 115 116 117
		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;
}
118

119 120 121 122 123
int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
			       struct btrfs_root *root)
{
	if (!root->ref_cows)
		return 0;
124

125 126 127 128
	mutex_lock(&root->fs_info->trans_mutex);
	if (root->last_trans == trans->transid) {
		mutex_unlock(&root->fs_info->trans_mutex);
		return 0;
129
	}
130 131 132

	record_root_in_trans(trans, root);
	mutex_unlock(&root->fs_info->trans_mutex);
133 134 135
	return 0;
}

C
Chris Mason 已提交
136 137 138 139
/* 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.
 */
C
Chris Mason 已提交
140
static void wait_current_trans(struct btrfs_root *root)
C
Chris Mason 已提交
141
{
142
	struct btrfs_transaction *cur_trans;
C
Chris Mason 已提交
143

144
	cur_trans = root->fs_info->running_transaction;
C
Chris Mason 已提交
145
	if (cur_trans && cur_trans->blocked) {
146
		DEFINE_WAIT(wait);
147
		atomic_inc(&cur_trans->use_count);
C
Chris Mason 已提交
148
		while (1) {
149 150
			prepare_to_wait(&root->fs_info->transaction_wait, &wait,
					TASK_UNINTERRUPTIBLE);
151
			if (!cur_trans->blocked)
152
				break;
153 154 155
			mutex_unlock(&root->fs_info->trans_mutex);
			schedule();
			mutex_lock(&root->fs_info->trans_mutex);
156
		}
157
		finish_wait(&root->fs_info->transaction_wait, &wait);
158 159
		put_transaction(cur_trans);
	}
C
Chris Mason 已提交
160 161
}

162 163 164 165
enum btrfs_trans_type {
	TRANS_START,
	TRANS_JOIN,
	TRANS_USERSPACE,
166
	TRANS_JOIN_NOLOCK,
167 168
};

169 170 171 172 173 174 175 176 177
static int may_wait_transaction(struct btrfs_root *root, int type)
{
	if (!root->fs_info->log_root_recovering &&
	    ((type == TRANS_START && !root->fs_info->open_ioctl_trans) ||
	     type == TRANS_USERSPACE))
		return 1;
	return 0;
}

178
static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
179
						    u64 num_items, int type)
C
Chris Mason 已提交
180
{
181 182
	struct btrfs_trans_handle *h;
	struct btrfs_transaction *cur_trans;
183
	int retries = 0;
C
Chris Mason 已提交
184
	int ret;
L
liubo 已提交
185 186 187

	if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
		return ERR_PTR(-EROFS);
188 189 190 191
again:
	h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
	if (!h)
		return ERR_PTR(-ENOMEM);
C
Chris Mason 已提交
192

193 194
	if (type != TRANS_JOIN_NOLOCK)
		mutex_lock(&root->fs_info->trans_mutex);
195
	if (may_wait_transaction(root, type))
C
Chris Mason 已提交
196
		wait_current_trans(root);
197

C
Chris Mason 已提交
198
	ret = join_transaction(root);
T
Tsutomu Itoh 已提交
199
	if (ret < 0) {
200
		kmem_cache_free(btrfs_trans_handle_cachep, h);
T
Tsutomu Itoh 已提交
201 202 203 204
		if (type != TRANS_JOIN_NOLOCK)
			mutex_unlock(&root->fs_info->trans_mutex);
		return ERR_PTR(ret);
	}
205

206
	cur_trans = root->fs_info->running_transaction;
207
	atomic_inc(&cur_trans->use_count);
208 209
	if (type != TRANS_JOIN_NOLOCK)
		mutex_unlock(&root->fs_info->trans_mutex);
210 211 212

	h->transid = cur_trans->transid;
	h->transaction = cur_trans;
C
Chris Mason 已提交
213
	h->blocks_used = 0;
214
	h->block_group = 0;
215
	h->bytes_reserved = 0;
216
	h->delayed_ref_updates = 0;
217
	h->block_rsv = NULL;
218

219 220 221 222 223 224 225
	smp_mb();
	if (cur_trans->blocked && may_wait_transaction(root, type)) {
		btrfs_commit_transaction(h, root);
		goto again;
	}

	if (num_items > 0) {
226
		ret = btrfs_trans_reserve_metadata(h, root, num_items);
227 228
		if (ret == -EAGAIN && !retries) {
			retries++;
229 230
			btrfs_commit_transaction(h, root);
			goto again;
231 232 233 234 235 236
		} else if (ret == -EAGAIN) {
			/*
			 * We have already retried and got EAGAIN, so really we
			 * don't have space, so set ret to -ENOSPC.
			 */
			ret = -ENOSPC;
237
		}
238

239 240 241 242 243
		if (ret < 0) {
			btrfs_end_transaction(h, root);
			return ERR_PTR(ret);
		}
	}
J
Josef Bacik 已提交
244

245 246
	if (type != TRANS_JOIN_NOLOCK)
		mutex_lock(&root->fs_info->trans_mutex);
247
	record_root_in_trans(h, root);
248 249
	if (type != TRANS_JOIN_NOLOCK)
		mutex_unlock(&root->fs_info->trans_mutex);
250 251 252

	if (!current->journal_info && type != TRANS_USERSPACE)
		current->journal_info = h;
C
Chris Mason 已提交
253 254 255
	return h;
}

256
struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
257
						   int num_items)
258
{
259
	return start_transaction(root, num_items, TRANS_START);
260 261 262 263
}
struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
						   int num_blocks)
{
264
	return start_transaction(root, 0, TRANS_JOIN);
265 266
}

267 268 269 270 271 272
struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root,
							  int num_blocks)
{
	return start_transaction(root, 0, TRANS_JOIN_NOLOCK);
}

273 274 275
struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
							 int num_blocks)
{
276
	return start_transaction(r, 0, TRANS_USERSPACE);
277 278
}

C
Chris Mason 已提交
279
/* wait for a transaction commit to be fully complete */
280 281 282 283 284
static noinline int wait_for_commit(struct btrfs_root *root,
				    struct btrfs_transaction *commit)
{
	DEFINE_WAIT(wait);
	mutex_lock(&root->fs_info->trans_mutex);
C
Chris Mason 已提交
285
	while (!commit->commit_done) {
286 287 288 289 290 291 292 293 294 295 296 297 298
		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;
}

299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337
int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
{
	struct btrfs_transaction *cur_trans = NULL, *t;
	int ret;

	mutex_lock(&root->fs_info->trans_mutex);

	ret = 0;
	if (transid) {
		if (transid <= root->fs_info->last_trans_committed)
			goto out_unlock;

		/* find specified transaction */
		list_for_each_entry(t, &root->fs_info->trans_list, list) {
			if (t->transid == transid) {
				cur_trans = t;
				break;
			}
			if (t->transid > transid)
				break;
		}
		ret = -EINVAL;
		if (!cur_trans)
			goto out_unlock;  /* bad transid */
	} else {
		/* find newest transaction that is committing | committed */
		list_for_each_entry_reverse(t, &root->fs_info->trans_list,
					    list) {
			if (t->in_commit) {
				if (t->commit_done)
					goto out_unlock;
				cur_trans = t;
				break;
			}
		}
		if (!cur_trans)
			goto out_unlock;  /* nothing committing|committed */
	}

338
	atomic_inc(&cur_trans->use_count);
339 340 341 342 343 344 345 346 347 348 349 350
	mutex_unlock(&root->fs_info->trans_mutex);

	wait_for_commit(root, cur_trans);

	mutex_lock(&root->fs_info->trans_mutex);
	put_transaction(cur_trans);
	ret = 0;
out_unlock:
	mutex_unlock(&root->fs_info->trans_mutex);
	return ret;
}

351
#if 0
C
Chris Mason 已提交
352
/*
C
Chris Mason 已提交
353 354
 * 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.
C
Chris Mason 已提交
355
 */
C
Chris Mason 已提交
356
static void throttle_on_drops(struct btrfs_root *root)
357 358
{
	struct btrfs_fs_info *info = root->fs_info;
C
Chris Mason 已提交
359
	int harder_count = 0;
360

C
Chris Mason 已提交
361
harder:
362 363 364 365 366 367 368 369 370 371 372 373 374 375 376
	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));
C
Chris Mason 已提交
377 378 379 380 381 382 383 384 385 386 387 388 389
		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;
390 391
	}
}
392
#endif
393

C
Chris Mason 已提交
394 395 396
void btrfs_throttle(struct btrfs_root *root)
{
	mutex_lock(&root->fs_info->trans_mutex);
397 398
	if (!root->fs_info->open_ioctl_trans)
		wait_current_trans(root);
C
Chris Mason 已提交
399 400 401
	mutex_unlock(&root->fs_info->trans_mutex);
}

402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427
static int should_end_transaction(struct btrfs_trans_handle *trans,
				  struct btrfs_root *root)
{
	int ret;
	ret = btrfs_block_rsv_check(trans, root,
				    &root->fs_info->global_block_rsv, 0, 5);
	return ret ? 1 : 0;
}

int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
				 struct btrfs_root *root)
{
	struct btrfs_transaction *cur_trans = trans->transaction;
	int updates;

	if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
		return 1;

	updates = trans->delayed_ref_updates;
	trans->delayed_ref_updates = 0;
	if (updates)
		btrfs_run_delayed_refs(trans, root, updates);

	return should_end_transaction(trans, root);
}

428
static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
429
			  struct btrfs_root *root, int throttle, int lock)
C
Chris Mason 已提交
430
{
431
	struct btrfs_transaction *cur_trans = trans->transaction;
432
	struct btrfs_fs_info *info = root->fs_info;
433 434 435 436 437 438 439 440
	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;
441 442 443 444 445 446 447

			/*
			 * do a full flush if the transaction is trying
			 * to close
			 */
			if (trans->transaction->delayed_refs.flushing)
				cur = 0;
448 449 450 451 452
			btrfs_run_delayed_refs(trans, root, cur);
		} else {
			break;
		}
		count++;
453 454
	}

455 456
	btrfs_trans_release_metadata(trans, root);

457
	if (lock && !root->fs_info->open_ioctl_trans &&
458 459 460
	    should_end_transaction(trans, root))
		trans->transaction->blocked = 1;

461
	if (lock && cur_trans->blocked && !cur_trans->in_commit) {
462 463 464 465 466 467 468
		if (throttle)
			return btrfs_commit_transaction(trans, root);
		else
			wake_up_process(info->transaction_kthread);
	}

	WARN_ON(cur_trans != info->running_transaction);
469 470
	WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
	atomic_dec(&cur_trans->num_writers);
471

472
	smp_mb();
C
Chris Mason 已提交
473 474 475
	if (waitqueue_active(&cur_trans->writer_wait))
		wake_up(&cur_trans->writer_wait);
	put_transaction(cur_trans);
J
Josef Bacik 已提交
476 477 478

	if (current->journal_info == trans)
		current->journal_info = NULL;
C
Chris Mason 已提交
479
	memset(trans, 0, sizeof(*trans));
C
Chris Mason 已提交
480
	kmem_cache_free(btrfs_trans_handle_cachep, trans);
481

Y
Yan, Zheng 已提交
482 483 484
	if (throttle)
		btrfs_run_delayed_iputs(root);

C
Chris Mason 已提交
485 486 487
	return 0;
}

488 489 490
int btrfs_end_transaction(struct btrfs_trans_handle *trans,
			  struct btrfs_root *root)
{
491
	return __btrfs_end_transaction(trans, root, 0, 1);
492 493 494 495 496
}

int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
				   struct btrfs_root *root)
{
497 498 499 500 501 502 503
	return __btrfs_end_transaction(trans, root, 1, 1);
}

int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans,
				 struct btrfs_root *root)
{
	return __btrfs_end_transaction(trans, root, 0, 0);
504 505
}

C
Chris Mason 已提交
506 507 508
/*
 * 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
509
 * those extents are sent to disk but does not wait on them
C
Chris Mason 已提交
510
 */
511
int btrfs_write_marked_extents(struct btrfs_root *root,
512
			       struct extent_io_tree *dirty_pages, int mark)
C
Chris Mason 已提交
513
{
514
	int ret;
515
	int err = 0;
516 517 518
	int werr = 0;
	struct page *page;
	struct inode *btree_inode = root->fs_info->btree_inode;
519
	u64 start = 0;
520 521
	u64 end;
	unsigned long index;
522

C
Chris Mason 已提交
523
	while (1) {
524
		ret = find_first_extent_bit(dirty_pages, start, &start, &end,
525
					    mark);
526
		if (ret)
527
			break;
C
Chris Mason 已提交
528
		while (start <= end) {
529 530
			cond_resched();

531
			index = start >> PAGE_CACHE_SHIFT;
532
			start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
C
Chris Mason 已提交
533
			page = find_get_page(btree_inode->i_mapping, index);
534 535
			if (!page)
				continue;
C
Chris Mason 已提交
536 537 538 539 540 541 542 543

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

544 545 546 547 548 549 550 551 552
			if (PageWriteback(page)) {
				if (PageDirty(page))
					wait_on_page_writeback(page);
				else {
					unlock_page(page);
					page_cache_release(page);
					continue;
				}
			}
553 554 555 556 557 558
			err = write_one_page(page, 0);
			if (err)
				werr = err;
			page_cache_release(page);
		}
	}
559 560 561 562 563 564 565 566 567 568 569 570
	if (err)
		werr = err;
	return werr;
}

/*
 * 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.  We wait
 * on all the pages and clear them from the dirty pages state tree
 */
int btrfs_wait_marked_extents(struct btrfs_root *root,
571
			      struct extent_io_tree *dirty_pages, int mark)
572 573 574 575 576 577 578 579 580 581
{
	int ret;
	int err = 0;
	int werr = 0;
	struct page *page;
	struct inode *btree_inode = root->fs_info->btree_inode;
	u64 start = 0;
	u64 end;
	unsigned long index;

C
Chris Mason 已提交
582
	while (1) {
583 584
		ret = find_first_extent_bit(dirty_pages, start, &start, &end,
					    mark);
585 586 587
		if (ret)
			break;

588
		clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
C
Chris Mason 已提交
589
		while (start <= end) {
590 591 592 593 594 595
			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)) {
C
Chris Mason 已提交
596 597
				btree_lock_page_hook(page);
				wait_on_page_writeback(page);
598 599 600 601
				err = write_one_page(page, 0);
				if (err)
					werr = err;
			}
602
			wait_on_page_writeback(page);
603 604 605 606
			page_cache_release(page);
			cond_resched();
		}
	}
607 608 609
	if (err)
		werr = err;
	return werr;
C
Chris Mason 已提交
610 611
}

612 613 614 615 616 617
/*
 * 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
 */
int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
618
				struct extent_io_tree *dirty_pages, int mark)
619 620 621 622
{
	int ret;
	int ret2;

623 624
	ret = btrfs_write_marked_extents(root, dirty_pages, mark);
	ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
625 626 627
	return ret || ret2;
}

628 629 630 631 632 633 634 635 636
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,
637 638
					   &trans->transaction->dirty_pages,
					   EXTENT_DIRTY);
639 640
}

C
Chris Mason 已提交
641 642 643 644 645 646 647 648 649 650
/*
 * 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.
 */
651 652
static int update_cowonly_root(struct btrfs_trans_handle *trans,
			       struct btrfs_root *root)
C
Chris Mason 已提交
653 654
{
	int ret;
655
	u64 old_root_bytenr;
656
	u64 old_root_used;
657
	struct btrfs_root *tree_root = root->fs_info->tree_root;
C
Chris Mason 已提交
658

659
	old_root_used = btrfs_root_used(&root->root_item);
660
	btrfs_write_dirty_block_groups(trans, root);
661

C
Chris Mason 已提交
662
	while (1) {
663
		old_root_bytenr = btrfs_root_bytenr(&root->root_item);
664 665
		if (old_root_bytenr == root->node->start &&
		    old_root_used == btrfs_root_used(&root->root_item))
C
Chris Mason 已提交
666
			break;
667

668
		btrfs_set_root_node(&root->root_item, root->node);
C
Chris Mason 已提交
669
		ret = btrfs_update_root(trans, tree_root,
670 671
					&root->root_key,
					&root->root_item);
C
Chris Mason 已提交
672
		BUG_ON(ret);
673

674
		old_root_used = btrfs_root_used(&root->root_item);
675
		ret = btrfs_write_dirty_block_groups(trans, root);
676
		BUG_ON(ret);
677
	}
678 679 680 681

	if (root != root->fs_info->extent_root)
		switch_commit_root(root);

682 683 684
	return 0;
}

C
Chris Mason 已提交
685 686 687
/*
 * update all the cowonly tree roots on disk
 */
688 689
static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
					 struct btrfs_root *root)
690 691 692
{
	struct btrfs_fs_info *fs_info = root->fs_info;
	struct list_head *next;
693
	struct extent_buffer *eb;
694
	int ret;
695

696 697
	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
	BUG_ON(ret);
698

699
	eb = btrfs_lock_root_node(fs_info->tree_root);
700
	btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb);
701 702
	btrfs_tree_unlock(eb);
	free_extent_buffer(eb);
703

704 705
	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
	BUG_ON(ret);
706

C
Chris Mason 已提交
707
	while (!list_empty(&fs_info->dirty_cowonly_roots)) {
708 709 710
		next = fs_info->dirty_cowonly_roots.next;
		list_del_init(next);
		root = list_entry(next, struct btrfs_root, dirty_list);
711

712
		update_cowonly_root(trans, root);
C
Chris Mason 已提交
713
	}
714 715 716 717 718

	down_write(&fs_info->extent_commit_sem);
	switch_commit_root(fs_info->extent_root);
	up_write(&fs_info->extent_commit_sem);

C
Chris Mason 已提交
719 720 721
	return 0;
}

C
Chris Mason 已提交
722 723 724 725 726
/*
 * 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
 */
727
int btrfs_add_dead_root(struct btrfs_root *root)
728
{
Y
Yan Zheng 已提交
729
	mutex_lock(&root->fs_info->trans_mutex);
730
	list_add(&root->root_list, &root->fs_info->dead_roots);
Y
Yan Zheng 已提交
731
	mutex_unlock(&root->fs_info->trans_mutex);
732 733 734
	return 0;
}

C
Chris Mason 已提交
735
/*
736
 * update all the cowonly tree roots on disk
C
Chris Mason 已提交
737
 */
738 739
static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
				    struct btrfs_root *root)
740 741
{
	struct btrfs_root *gang[8];
742
	struct btrfs_fs_info *fs_info = root->fs_info;
743 744
	int i;
	int ret;
745 746
	int err = 0;

C
Chris Mason 已提交
747
	while (1) {
748 749
		ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
						 (void **)gang, 0,
750 751 752 753 754 755
						 ARRAY_SIZE(gang),
						 BTRFS_ROOT_TRANS_TAG);
		if (ret == 0)
			break;
		for (i = 0; i < ret; i++) {
			root = gang[i];
756 757 758
			radix_tree_tag_clear(&fs_info->fs_roots_radix,
					(unsigned long)root->root_key.objectid,
					BTRFS_ROOT_TRANS_TAG);
Y
Yan Zheng 已提交
759

760
			btrfs_free_log(trans, root);
761
			btrfs_update_reloc_root(trans, root);
762
			btrfs_orphan_commit_root(trans, root);
763

764
			if (root->commit_root != root->node) {
765
				mutex_lock(&root->fs_commit_mutex);
J
Josef Bacik 已提交
766
				switch_commit_root(root);
767 768 769
				btrfs_unpin_free_ino(root);
				mutex_unlock(&root->fs_commit_mutex);

770 771 772
				btrfs_set_root_node(&root->root_item,
						    root->node);
			}
773 774

			err = btrfs_update_root(trans, fs_info->tree_root,
775 776
						&root->root_key,
						&root->root_item);
777 778
			if (err)
				break;
779 780
		}
	}
781
	return err;
782 783
}

C
Chris Mason 已提交
784 785 786 787
/*
 * defrag a given btree.  If cacheonly == 1, this won't read from the disk,
 * otherwise every leaf in the btree is read and defragged.
 */
788 789 790 791
int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
{
	struct btrfs_fs_info *info = root->fs_info;
	struct btrfs_trans_handle *trans;
792
	int ret;
793
	unsigned long nr;
794

795
	if (xchg(&root->defrag_running, 1))
796
		return 0;
797

798
	while (1) {
799 800 801 802
		trans = btrfs_start_transaction(root, 0);
		if (IS_ERR(trans))
			return PTR_ERR(trans);

803
		ret = btrfs_defrag_leaves(trans, root, cacheonly);
804

805
		nr = trans->blocks_used;
806
		btrfs_end_transaction(trans, root);
807
		btrfs_btree_balance_dirty(info->tree_root, nr);
808 809
		cond_resched();

810
		if (root->fs_info->closing || ret != -EAGAIN)
811 812 813
			break;
	}
	root->defrag_running = 0;
814
	return ret;
815 816
}

817
#if 0
818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834
/*
 * 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);
835

836
		schedule();
837

838 839 840 841 842 843 844
		mutex_lock(&info->trans_mutex);
		finish_wait(&info->transaction_wait, &wait);
	}
	mutex_unlock(&info->trans_mutex);
	return 0;
}

C
Chris Mason 已提交
845 846 847 848
/*
 * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
 * all of them
 */
849
int btrfs_drop_dead_root(struct btrfs_root *root)
850 851
{
	struct btrfs_trans_handle *trans;
852
	struct btrfs_root *tree_root = root->fs_info->tree_root;
853
	unsigned long nr;
854
	int ret;
855

856 857 858 859 860 861 862
	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);
863

864 865 866 867 868 869 870
		/*
		 * 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;
871
		}
872

873 874 875
		ret = btrfs_drop_snapshot(trans, root);
		if (ret != -EAGAIN)
			break;
876

877 878 879 880
		ret = btrfs_update_root(trans, tree_root,
					&root->root_key,
					&root->root_item);
		if (ret)
881
			break;
882

883
		nr = trans->blocks_used;
884 885
		ret = btrfs_end_transaction(trans, tree_root);
		BUG_ON(ret);
886

887
		btrfs_btree_balance_dirty(tree_root, nr);
888
		cond_resched();
889
	}
890 891 892 893 894 895 896 897 898 899 900 901 902 903
	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);
904
	return ret;
905
}
906
#endif
907

C
Chris Mason 已提交
908 909 910 911
/*
 * new snapshots need to be created at a very specific time in the
 * transaction commit.  This does the actual creation
 */
912
static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
913 914 915 916
				   struct btrfs_fs_info *fs_info,
				   struct btrfs_pending_snapshot *pending)
{
	struct btrfs_key key;
917
	struct btrfs_root_item *new_root_item;
918 919
	struct btrfs_root *tree_root = fs_info->tree_root;
	struct btrfs_root *root = pending->root;
920 921
	struct btrfs_root *parent_root;
	struct inode *parent_inode;
922
	struct dentry *parent;
923
	struct dentry *dentry;
924
	struct extent_buffer *tmp;
925
	struct extent_buffer *old;
926
	int ret;
927
	u64 to_reserve = 0;
928
	u64 index = 0;
929
	u64 objectid;
L
Li Zefan 已提交
930
	u64 root_flags;
931

932 933
	new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
	if (!new_root_item) {
934
		pending->error = -ENOMEM;
935 936
		goto fail;
	}
937

938
	ret = btrfs_find_free_objectid(tree_root, &objectid);
939 940
	if (ret) {
		pending->error = ret;
941
		goto fail;
942
	}
943

944
	btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
945 946 947 948
	btrfs_orphan_pre_snapshot(trans, pending, &to_reserve);

	if (to_reserve > 0) {
		ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv,
949
					  to_reserve);
950 951 952 953 954 955
		if (ret) {
			pending->error = ret;
			goto fail;
		}
	}

956
	key.objectid = objectid;
957 958
	key.offset = (u64)-1;
	key.type = BTRFS_ROOT_ITEM_KEY;
959

960
	trans->block_rsv = &pending->block_rsv;
961

962
	dentry = pending->dentry;
963 964
	parent = dget_parent(dentry);
	parent_inode = parent->d_inode;
965
	parent_root = BTRFS_I(parent_inode)->root;
966
	record_root_in_trans(trans, parent_root);
967

968 969 970
	/*
	 * insert the directory item
	 */
971
	ret = btrfs_set_inode_index(parent_inode, &index);
972
	BUG_ON(ret);
973
	ret = btrfs_insert_dir_item(trans, parent_root,
974 975 976
				dentry->d_name.name, dentry->d_name.len,
				parent_inode->i_ino, &key,
				BTRFS_FT_DIR, index);
977
	BUG_ON(ret);
978

979 980
	btrfs_i_size_write(parent_inode, parent_inode->i_size +
					 dentry->d_name.len * 2);
981 982 983
	ret = btrfs_update_inode(trans, parent_root, parent_inode);
	BUG_ON(ret);

984 985 986
	record_root_in_trans(trans, root);
	btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
	memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
987
	btrfs_check_and_init_root_item(new_root_item);
988

L
Li Zefan 已提交
989 990 991 992 993 994 995
	root_flags = btrfs_root_flags(new_root_item);
	if (pending->readonly)
		root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
	else
		root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
	btrfs_set_root_flags(new_root_item, root_flags);

996 997 998 999 1000 1001 1002 1003 1004
	old = btrfs_lock_root_node(root);
	btrfs_cow_block(trans, root, old, NULL, 0, &old);
	btrfs_set_lock_blocking(old);

	btrfs_copy_root(trans, root, old, &tmp, objectid);
	btrfs_tree_unlock(old);
	free_extent_buffer(old);

	btrfs_set_root_node(new_root_item, tmp);
1005 1006 1007
	/* record when the snapshot was created in key.offset */
	key.offset = trans->transid;
	ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1008 1009
	btrfs_tree_unlock(tmp);
	free_extent_buffer(tmp);
1010
	BUG_ON(ret);
1011

1012 1013 1014 1015
	/*
	 * insert root back/forward references
	 */
	ret = btrfs_add_root_ref(trans, tree_root, objectid,
1016
				 parent_root->root_key.objectid,
1017 1018
				 parent_inode->i_ino, index,
				 dentry->d_name.name, dentry->d_name.len);
1019
	BUG_ON(ret);
1020
	dput(parent);
1021

1022 1023 1024
	key.offset = (u64)-1;
	pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
	BUG_ON(IS_ERR(pending->snap));
1025

1026
	btrfs_reloc_post_snapshot(trans, pending);
1027
	btrfs_orphan_post_snapshot(trans, pending);
1028
fail:
1029
	kfree(new_root_item);
1030 1031
	btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1);
	return 0;
1032 1033
}

C
Chris Mason 已提交
1034 1035 1036
/*
 * create all the snapshots we've scheduled for creation
 */
1037 1038
static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
					     struct btrfs_fs_info *fs_info)
1039 1040 1041 1042 1043
{
	struct btrfs_pending_snapshot *pending;
	struct list_head *head = &trans->transaction->pending_snapshots;
	int ret;

Q
Qinghuang Feng 已提交
1044
	list_for_each_entry(pending, head, list) {
1045 1046 1047 1048 1049 1050
		ret = create_pending_snapshot(trans, fs_info, pending);
		BUG_ON(ret);
	}
	return 0;
}

1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066
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;
1067 1068
	if (super->cache_generation != 0 || btrfs_test_opt(root, SPACE_CACHE))
		super->cache_generation = root_item->generation;
1069 1070
}

1071 1072 1073 1074 1075 1076 1077 1078 1079 1080
int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
{
	int ret = 0;
	spin_lock(&info->new_trans_lock);
	if (info->running_transaction)
		ret = info->running_transaction->in_commit;
	spin_unlock(&info->new_trans_lock);
	return ret;
}

1081 1082 1083 1084 1085 1086 1087 1088 1089 1090
int btrfs_transaction_blocked(struct btrfs_fs_info *info)
{
	int ret = 0;
	spin_lock(&info->new_trans_lock);
	if (info->running_transaction)
		ret = info->running_transaction->blocked;
	spin_unlock(&info->new_trans_lock);
	return ret;
}

S
Sage Weil 已提交
1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173
/*
 * wait for the current transaction commit to start and block subsequent
 * transaction joins
 */
static void wait_current_trans_commit_start(struct btrfs_root *root,
					    struct btrfs_transaction *trans)
{
	DEFINE_WAIT(wait);

	if (trans->in_commit)
		return;

	while (1) {
		prepare_to_wait(&root->fs_info->transaction_blocked_wait, &wait,
				TASK_UNINTERRUPTIBLE);
		if (trans->in_commit) {
			finish_wait(&root->fs_info->transaction_blocked_wait,
				    &wait);
			break;
		}
		mutex_unlock(&root->fs_info->trans_mutex);
		schedule();
		mutex_lock(&root->fs_info->trans_mutex);
		finish_wait(&root->fs_info->transaction_blocked_wait, &wait);
	}
}

/*
 * wait for the current transaction to start and then become unblocked.
 * caller holds ref.
 */
static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
					 struct btrfs_transaction *trans)
{
	DEFINE_WAIT(wait);

	if (trans->commit_done || (trans->in_commit && !trans->blocked))
		return;

	while (1) {
		prepare_to_wait(&root->fs_info->transaction_wait, &wait,
				TASK_UNINTERRUPTIBLE);
		if (trans->commit_done ||
		    (trans->in_commit && !trans->blocked)) {
			finish_wait(&root->fs_info->transaction_wait,
				    &wait);
			break;
		}
		mutex_unlock(&root->fs_info->trans_mutex);
		schedule();
		mutex_lock(&root->fs_info->trans_mutex);
		finish_wait(&root->fs_info->transaction_wait,
			    &wait);
	}
}

/*
 * commit transactions asynchronously. once btrfs_commit_transaction_async
 * returns, any subsequent transaction will not be allowed to join.
 */
struct btrfs_async_commit {
	struct btrfs_trans_handle *newtrans;
	struct btrfs_root *root;
	struct delayed_work work;
};

static void do_async_commit(struct work_struct *work)
{
	struct btrfs_async_commit *ac =
		container_of(work, struct btrfs_async_commit, work.work);

	btrfs_commit_transaction(ac->newtrans, ac->root);
	kfree(ac);
}

int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
				   struct btrfs_root *root,
				   int wait_for_unblock)
{
	struct btrfs_async_commit *ac;
	struct btrfs_transaction *cur_trans;

	ac = kmalloc(sizeof(*ac), GFP_NOFS);
T
Tsutomu Itoh 已提交
1174 1175
	if (!ac)
		return -ENOMEM;
S
Sage Weil 已提交
1176 1177 1178 1179

	INIT_DELAYED_WORK(&ac->work, do_async_commit);
	ac->root = root;
	ac->newtrans = btrfs_join_transaction(root, 0);
1180 1181 1182 1183 1184
	if (IS_ERR(ac->newtrans)) {
		int err = PTR_ERR(ac->newtrans);
		kfree(ac);
		return err;
	}
S
Sage Weil 已提交
1185 1186 1187 1188

	/* take transaction reference */
	mutex_lock(&root->fs_info->trans_mutex);
	cur_trans = trans->transaction;
1189
	atomic_inc(&cur_trans->use_count);
S
Sage Weil 已提交
1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213
	mutex_unlock(&root->fs_info->trans_mutex);

	btrfs_end_transaction(trans, root);
	schedule_delayed_work(&ac->work, 0);

	/* wait for transaction to start and unblock */
	mutex_lock(&root->fs_info->trans_mutex);
	if (wait_for_unblock)
		wait_current_trans_commit_start_and_unblock(root, cur_trans);
	else
		wait_current_trans_commit_start(root, cur_trans);
	put_transaction(cur_trans);
	mutex_unlock(&root->fs_info->trans_mutex);

	return 0;
}

/*
 * btrfs_transaction state sequence:
 *    in_commit = 0, blocked = 0  (initial)
 *    in_commit = 1, blocked = 1
 *    blocked = 0
 *    commit_done = 1
 */
C
Chris Mason 已提交
1214 1215 1216
int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
			     struct btrfs_root *root)
{
1217
	unsigned long joined = 0;
C
Chris Mason 已提交
1218
	struct btrfs_transaction *cur_trans;
C
Chris Mason 已提交
1219
	struct btrfs_transaction *prev_trans = NULL;
C
Chris Mason 已提交
1220
	DEFINE_WAIT(wait);
1221
	int ret;
1222 1223
	int should_grow = 0;
	unsigned long now = get_seconds();
1224
	int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
C
Chris Mason 已提交
1225

1226 1227
	btrfs_run_ordered_operations(root, 0);

1228 1229 1230 1231 1232 1233
	/* 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);

1234 1235
	btrfs_trans_release_metadata(trans, root);

1236
	cur_trans = trans->transaction;
1237 1238 1239 1240
	/*
	 * set the flushing flag so procs in this transaction have to
	 * start sending their work down.
	 */
1241
	cur_trans->delayed_refs.flushing = 1;
1242

1243
	ret = btrfs_run_delayed_refs(trans, root, 0);
1244 1245
	BUG_ON(ret);

C
Chris Mason 已提交
1246
	mutex_lock(&root->fs_info->trans_mutex);
1247
	if (cur_trans->in_commit) {
1248
		atomic_inc(&cur_trans->use_count);
C
Chris Mason 已提交
1249
		mutex_unlock(&root->fs_info->trans_mutex);
C
Chris Mason 已提交
1250
		btrfs_end_transaction(trans, root);
C
Chris Mason 已提交
1251

C
Chris Mason 已提交
1252 1253
		ret = wait_for_commit(root, cur_trans);
		BUG_ON(ret);
1254 1255

		mutex_lock(&root->fs_info->trans_mutex);
C
Chris Mason 已提交
1256
		put_transaction(cur_trans);
1257 1258
		mutex_unlock(&root->fs_info->trans_mutex);

C
Chris Mason 已提交
1259 1260
		return 0;
	}
1261

C
Chris Mason 已提交
1262
	trans->transaction->in_commit = 1;
1263
	trans->transaction->blocked = 1;
S
Sage Weil 已提交
1264 1265
	wake_up(&root->fs_info->transaction_blocked_wait);

C
Chris Mason 已提交
1266 1267 1268 1269
	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) {
1270
			atomic_inc(&prev_trans->use_count);
C
Chris Mason 已提交
1271 1272 1273 1274 1275
			mutex_unlock(&root->fs_info->trans_mutex);

			wait_for_commit(root, prev_trans);

			mutex_lock(&root->fs_info->trans_mutex);
1276
			put_transaction(prev_trans);
C
Chris Mason 已提交
1277 1278
		}
	}
1279

1280 1281 1282
	if (now < cur_trans->start_time || now - cur_trans->start_time < 1)
		should_grow = 1;

1283
	do {
1284
		int snap_pending = 0;
1285
		joined = cur_trans->num_joined;
1286 1287 1288
		if (!list_empty(&trans->transaction->pending_snapshots))
			snap_pending = 1;

C
Chris Mason 已提交
1289
		WARN_ON(cur_trans != trans->transaction);
C
Chris Mason 已提交
1290
		mutex_unlock(&root->fs_info->trans_mutex);
1291

1292
		if (flush_on_commit || snap_pending) {
Y
Yan, Zheng 已提交
1293 1294
			btrfs_start_delalloc_inodes(root, 1);
			ret = btrfs_wait_ordered_extents(root, 0, 1);
1295
			BUG_ON(ret);
1296 1297
		}

1298 1299 1300 1301 1302 1303 1304 1305 1306
		/*
		 * 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);

1307 1308 1309
		prepare_to_wait(&cur_trans->writer_wait, &wait,
				TASK_UNINTERRUPTIBLE);

1310
		smp_mb();
1311
		if (atomic_read(&cur_trans->num_writers) > 1)
1312 1313 1314
			schedule_timeout(MAX_SCHEDULE_TIMEOUT);
		else if (should_grow)
			schedule_timeout(1);
1315

C
Chris Mason 已提交
1316
		mutex_lock(&root->fs_info->trans_mutex);
1317
		finish_wait(&cur_trans->writer_wait, &wait);
1318
	} while (atomic_read(&cur_trans->num_writers) > 1 ||
1319
		 (should_grow && cur_trans->num_joined != joined));
1320

1321 1322 1323
	ret = create_pending_snapshots(trans, root->fs_info);
	BUG_ON(ret);

1324 1325 1326
	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
	BUG_ON(ret);

C
Chris Mason 已提交
1327
	WARN_ON(cur_trans != trans->transaction);
C
Chris Mason 已提交
1328

1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343
	/* 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);

1344
	ret = commit_fs_roots(trans, root);
1345 1346
	BUG_ON(ret);

1347
	/* commit_fs_roots gets rid of all the tree log roots, it is now
1348 1349 1350 1351
	 * safe to free the root of tree log roots
	 */
	btrfs_free_log_root_tree(trans, root->fs_info);

1352
	ret = commit_cowonly_roots(trans, root);
C
Chris Mason 已提交
1353
	BUG_ON(ret);
1354

1355 1356
	btrfs_prepare_extent_commit(trans, root);

C
Chris Mason 已提交
1357
	cur_trans = root->fs_info->running_transaction;
1358
	spin_lock(&root->fs_info->new_trans_lock);
C
Chris Mason 已提交
1359
	root->fs_info->running_transaction = NULL;
1360
	spin_unlock(&root->fs_info->new_trans_lock);
1361 1362 1363

	btrfs_set_root_node(&root->fs_info->tree_root->root_item,
			    root->fs_info->tree_root->node);
J
Josef Bacik 已提交
1364
	switch_commit_root(root->fs_info->tree_root);
1365 1366 1367

	btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
			    root->fs_info->chunk_root->node);
J
Josef Bacik 已提交
1368
	switch_commit_root(root->fs_info->chunk_root);
1369 1370

	update_super_roots(root);
1371 1372 1373 1374 1375 1376

	if (!root->fs_info->log_root_recovering) {
		btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
		btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0);
	}

1377 1378
	memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
	       sizeof(root->fs_info->super_copy));
C
Chris Mason 已提交
1379

1380
	trans->transaction->blocked = 0;
1381

1382
	wake_up(&root->fs_info->transaction_wait);
1383

C
Chris Mason 已提交
1384
	mutex_unlock(&root->fs_info->trans_mutex);
C
Chris Mason 已提交
1385 1386
	ret = btrfs_write_and_wait_transaction(trans, root);
	BUG_ON(ret);
Y
Yan Zheng 已提交
1387
	write_ctree_super(trans, root, 0);
1388

1389 1390 1391 1392 1393 1394
	/*
	 * the super is written, we can safely allow the tree-loggers
	 * to go about their business
	 */
	mutex_unlock(&root->fs_info->tree_log_mutex);

1395
	btrfs_finish_extent_commit(trans, root);
1396

Z
Zheng Yan 已提交
1397 1398
	mutex_lock(&root->fs_info->trans_mutex);

C
Chris Mason 已提交
1399
	cur_trans->commit_done = 1;
1400

1401
	root->fs_info->last_trans_committed = cur_trans->transid;
J
Josef Bacik 已提交
1402

C
Chris Mason 已提交
1403
	wake_up(&cur_trans->commit_wait);
1404

1405
	list_del_init(&cur_trans->list);
C
Chris Mason 已提交
1406
	put_transaction(cur_trans);
C
Chris Mason 已提交
1407
	put_transaction(cur_trans);
1408

1409 1410
	trace_btrfs_transaction_commit(root);

C
Chris Mason 已提交
1411
	mutex_unlock(&root->fs_info->trans_mutex);
1412

J
Josef Bacik 已提交
1413 1414 1415
	if (current->journal_info == trans)
		current->journal_info = NULL;

C
Chris Mason 已提交
1416
	kmem_cache_free(btrfs_trans_handle_cachep, trans);
Y
Yan, Zheng 已提交
1417 1418 1419 1420

	if (current != root->fs_info->transaction_kthread)
		btrfs_run_delayed_iputs(root);

C
Chris Mason 已提交
1421 1422 1423
	return ret;
}

C
Chris Mason 已提交
1424 1425 1426
/*
 * interface function to delete all the snapshots we have scheduled for deletion
 */
1427 1428
int btrfs_clean_old_snapshots(struct btrfs_root *root)
{
1429 1430 1431 1432 1433 1434
	LIST_HEAD(list);
	struct btrfs_fs_info *fs_info = root->fs_info;

	mutex_lock(&fs_info->trans_mutex);
	list_splice_init(&fs_info->dead_roots, &list);
	mutex_unlock(&fs_info->trans_mutex);
1435

1436 1437
	while (!list_empty(&list)) {
		root = list_entry(list.next, struct btrfs_root, root_list);
1438 1439 1440 1441
		list_del(&root->root_list);

		if (btrfs_header_backref_rev(root->node) <
		    BTRFS_MIXED_BACKREF_REV)
1442
			btrfs_drop_snapshot(root, NULL, 0);
1443
		else
1444
			btrfs_drop_snapshot(root, NULL, 1);
1445 1446 1447
	}
	return 0;
}