transaction.c 30.0 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>
C
Chris Mason 已提交
20
#include <linux/sched.h>
21
#include <linux/writeback.h>
22
#include <linux/pagemap.h>
23
#include <linux/blkdev.h>
C
Chris Mason 已提交
24 25 26
#include "ctree.h"
#include "disk-io.h"
#include "transaction.h"
27
#include "locking.h"
28
#include "tree-log.h"
C
Chris Mason 已提交
29

30 31
#define BTRFS_ROOT_TRANS_TAG 0

32
static noinline void put_transaction(struct btrfs_transaction *transaction)
C
Chris Mason 已提交
33
{
C
Chris Mason 已提交
34
	WARN_ON(transaction->use_count == 0);
C
Chris Mason 已提交
35
	transaction->use_count--;
C
Chris Mason 已提交
36
	if (transaction->use_count == 0) {
C
Chris Mason 已提交
37
		list_del_init(&transaction->list);
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);
C
Chris Mason 已提交
59
		BUG_ON(!cur_trans);
60
		root->fs_info->generation++;
61 62
		cur_trans->num_writers = 1;
		cur_trans->num_joined = 0;
63
		cur_trans->transid = root->fs_info->generation;
C
Chris Mason 已提交
64 65 66
		init_waitqueue_head(&cur_trans->writer_wait);
		init_waitqueue_head(&cur_trans->commit_wait);
		cur_trans->in_commit = 0;
67
		cur_trans->blocked = 0;
68
		cur_trans->use_count = 1;
C
Chris Mason 已提交
69
		cur_trans->commit_done = 0;
C
Chris Mason 已提交
70
		cur_trans->start_time = get_seconds();
71

72
		cur_trans->delayed_refs.root = RB_ROOT;
73
		cur_trans->delayed_refs.num_entries = 0;
74 75
		cur_trans->delayed_refs.num_heads_ready = 0;
		cur_trans->delayed_refs.num_heads = 0;
76
		cur_trans->delayed_refs.flushing = 0;
77
		cur_trans->delayed_refs.run_delayed_start = 0;
78 79
		spin_lock_init(&cur_trans->delayed_refs.lock);

80
		INIT_LIST_HEAD(&cur_trans->pending_snapshots);
C
Chris Mason 已提交
81
		list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
82
		extent_io_tree_init(&cur_trans->dirty_pages,
83 84
				     root->fs_info->btree_inode->i_mapping,
				     GFP_NOFS);
85 86 87
		spin_lock(&root->fs_info->new_trans_lock);
		root->fs_info->running_transaction = cur_trans;
		spin_unlock(&root->fs_info->new_trans_lock);
88 89 90
	} else {
		cur_trans->num_writers++;
		cur_trans->num_joined++;
C
Chris Mason 已提交
91
	}
92

C
Chris Mason 已提交
93 94 95
	return 0;
}

C
Chris Mason 已提交
96
/*
C
Chris Mason 已提交
97 98 99 100
 * 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
101
 */
102 103
static noinline int record_root_in_trans(struct btrfs_trans_handle *trans,
					 struct btrfs_root *root)
104
{
105
	if (root->ref_cows && root->last_trans < trans->transid) {
106
		WARN_ON(root == root->fs_info->extent_root);
107 108 109 110 111 112 113 114 115 116
		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;
}
117

118 119 120 121 122
int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
			       struct btrfs_root *root)
{
	if (!root->ref_cows)
		return 0;
123

124 125 126 127
	mutex_lock(&root->fs_info->trans_mutex);
	if (root->last_trans == trans->transid) {
		mutex_unlock(&root->fs_info->trans_mutex);
		return 0;
128
	}
129 130 131

	record_root_in_trans(trans, root);
	mutex_unlock(&root->fs_info->trans_mutex);
132 133 134
	return 0;
}

C
Chris Mason 已提交
135 136 137 138
/* 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
139
static void wait_current_trans(struct btrfs_root *root)
C
Chris Mason 已提交
140
{
141
	struct btrfs_transaction *cur_trans;
C
Chris Mason 已提交
142

143
	cur_trans = root->fs_info->running_transaction;
C
Chris Mason 已提交
144
	if (cur_trans && cur_trans->blocked) {
145 146
		DEFINE_WAIT(wait);
		cur_trans->use_count++;
C
Chris Mason 已提交
147
		while (1) {
148 149
			prepare_to_wait(&root->fs_info->transaction_wait, &wait,
					TASK_UNINTERRUPTIBLE);
150
			if (!cur_trans->blocked)
151
				break;
152 153 154
			mutex_unlock(&root->fs_info->trans_mutex);
			schedule();
			mutex_lock(&root->fs_info->trans_mutex);
155
		}
156
		finish_wait(&root->fs_info->transaction_wait, &wait);
157 158
		put_transaction(cur_trans);
	}
C
Chris Mason 已提交
159 160
}

161 162 163 164 165 166
enum btrfs_trans_type {
	TRANS_START,
	TRANS_JOIN,
	TRANS_USERSPACE,
};

167
static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
168
					     int num_blocks, int type)
C
Chris Mason 已提交
169 170 171 172 173 174
{
	struct btrfs_trans_handle *h =
		kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
	int ret;

	mutex_lock(&root->fs_info->trans_mutex);
C
Chris Mason 已提交
175
	if (!root->fs_info->log_root_recovering &&
176 177
	    ((type == TRANS_START && !root->fs_info->open_ioctl_trans) ||
	     type == TRANS_USERSPACE))
C
Chris Mason 已提交
178
		wait_current_trans(root);
C
Chris Mason 已提交
179 180
	ret = join_transaction(root);
	BUG_ON(ret);
181

182
	h->transid = root->fs_info->running_transaction->transid;
C
Chris Mason 已提交
183 184 185
	h->transaction = root->fs_info->running_transaction;
	h->blocks_reserved = num_blocks;
	h->blocks_used = 0;
186
	h->block_group = 0;
187 188
	h->alloc_exclude_nr = 0;
	h->alloc_exclude_start = 0;
189
	h->delayed_ref_updates = 0;
190

191
	if (!current->journal_info && type != TRANS_USERSPACE)
J
Josef Bacik 已提交
192 193
		current->journal_info = h;

C
Chris Mason 已提交
194
	root->fs_info->running_transaction->use_count++;
195
	record_root_in_trans(h, root);
C
Chris Mason 已提交
196 197 198 199
	mutex_unlock(&root->fs_info->trans_mutex);
	return h;
}

200 201 202
struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
						   int num_blocks)
{
203
	return start_transaction(root, num_blocks, TRANS_START);
204 205 206 207
}
struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
						   int num_blocks)
{
208
	return start_transaction(root, num_blocks, TRANS_JOIN);
209 210
}

211 212 213
struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
							 int num_blocks)
{
214
	return start_transaction(r, num_blocks, TRANS_USERSPACE);
215 216
}

C
Chris Mason 已提交
217
/* wait for a transaction commit to be fully complete */
218 219 220 221 222
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
223
	while (!commit->commit_done) {
224 225 226 227 228 229 230 231 232 233 234 235 236
		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;
}

237
#if 0
C
Chris Mason 已提交
238
/*
C
Chris Mason 已提交
239 240
 * 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
241
 */
C
Chris Mason 已提交
242
static void throttle_on_drops(struct btrfs_root *root)
243 244
{
	struct btrfs_fs_info *info = root->fs_info;
C
Chris Mason 已提交
245
	int harder_count = 0;
246

C
Chris Mason 已提交
247
harder:
248 249 250 251 252 253 254 255 256 257 258 259 260 261 262
	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
263 264 265 266 267 268 269 270 271 272 273 274 275
		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;
276 277
	}
}
278
#endif
279

C
Chris Mason 已提交
280 281 282
void btrfs_throttle(struct btrfs_root *root)
{
	mutex_lock(&root->fs_info->trans_mutex);
283 284
	if (!root->fs_info->open_ioctl_trans)
		wait_current_trans(root);
C
Chris Mason 已提交
285 286 287
	mutex_unlock(&root->fs_info->trans_mutex);
}

288 289
static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
			  struct btrfs_root *root, int throttle)
C
Chris Mason 已提交
290 291
{
	struct btrfs_transaction *cur_trans;
292
	struct btrfs_fs_info *info = root->fs_info;
293 294 295 296 297 298 299 300
	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;
301 302 303 304 305 306 307

			/*
			 * do a full flush if the transaction is trying
			 * to close
			 */
			if (trans->transaction->delayed_refs.flushing)
				cur = 0;
308 309 310 311 312
			btrfs_run_delayed_refs(trans, root, cur);
		} else {
			break;
		}
		count++;
313 314
	}

315 316
	mutex_lock(&info->trans_mutex);
	cur_trans = info->running_transaction;
C
Chris Mason 已提交
317
	WARN_ON(cur_trans != trans->transaction);
318
	WARN_ON(cur_trans->num_writers < 1);
C
Chris Mason 已提交
319
	cur_trans->num_writers--;
320

C
Chris Mason 已提交
321 322 323
	if (waitqueue_active(&cur_trans->writer_wait))
		wake_up(&cur_trans->writer_wait);
	put_transaction(cur_trans);
324
	mutex_unlock(&info->trans_mutex);
J
Josef Bacik 已提交
325 326 327

	if (current->journal_info == trans)
		current->journal_info = NULL;
C
Chris Mason 已提交
328
	memset(trans, 0, sizeof(*trans));
C
Chris Mason 已提交
329
	kmem_cache_free(btrfs_trans_handle_cachep, trans);
330

Y
Yan, Zheng 已提交
331 332 333
	if (throttle)
		btrfs_run_delayed_iputs(root);

C
Chris Mason 已提交
334 335 336
	return 0;
}

337 338 339 340 341 342 343 344 345 346 347 348
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);
}

C
Chris Mason 已提交
349 350 351
/*
 * 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
352
 * those extents are sent to disk but does not wait on them
C
Chris Mason 已提交
353
 */
354
int btrfs_write_marked_extents(struct btrfs_root *root,
355
			       struct extent_io_tree *dirty_pages, int mark)
C
Chris Mason 已提交
356
{
357
	int ret;
358
	int err = 0;
359 360 361
	int werr = 0;
	struct page *page;
	struct inode *btree_inode = root->fs_info->btree_inode;
362
	u64 start = 0;
363 364
	u64 end;
	unsigned long index;
365

C
Chris Mason 已提交
366
	while (1) {
367
		ret = find_first_extent_bit(dirty_pages, start, &start, &end,
368
					    mark);
369
		if (ret)
370
			break;
C
Chris Mason 已提交
371
		while (start <= end) {
372 373
			cond_resched();

374
			index = start >> PAGE_CACHE_SHIFT;
375
			start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
C
Chris Mason 已提交
376
			page = find_get_page(btree_inode->i_mapping, index);
377 378
			if (!page)
				continue;
C
Chris Mason 已提交
379 380 381 382 383 384 385 386

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

387 388 389 390 391 392 393 394 395
			if (PageWriteback(page)) {
				if (PageDirty(page))
					wait_on_page_writeback(page);
				else {
					unlock_page(page);
					page_cache_release(page);
					continue;
				}
			}
396 397 398 399 400 401
			err = write_one_page(page, 0);
			if (err)
				werr = err;
			page_cache_release(page);
		}
	}
402 403 404 405 406 407 408 409 410 411 412 413
	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,
414
			      struct extent_io_tree *dirty_pages, int mark)
415 416 417 418 419 420 421 422 423 424
{
	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
425
	while (1) {
426 427
		ret = find_first_extent_bit(dirty_pages, start, &start, &end,
					    mark);
428 429 430
		if (ret)
			break;

431
		clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
C
Chris Mason 已提交
432
		while (start <= end) {
433 434 435 436 437 438
			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
439 440
				btree_lock_page_hook(page);
				wait_on_page_writeback(page);
441 442 443 444
				err = write_one_page(page, 0);
				if (err)
					werr = err;
			}
445
			wait_on_page_writeback(page);
446 447 448 449
			page_cache_release(page);
			cond_resched();
		}
	}
450 451 452
	if (err)
		werr = err;
	return werr;
C
Chris Mason 已提交
453 454
}

455 456 457 458 459 460
/*
 * 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,
461
				struct extent_io_tree *dirty_pages, int mark)
462 463 464 465
{
	int ret;
	int ret2;

466 467
	ret = btrfs_write_marked_extents(root, dirty_pages, mark);
	ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
468 469 470
	return ret || ret2;
}

471 472 473 474 475 476 477 478 479
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,
480 481
					   &trans->transaction->dirty_pages,
					   EXTENT_DIRTY);
482 483
}

C
Chris Mason 已提交
484 485 486 487 488 489 490 491 492 493
/*
 * 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.
 */
494 495
static int update_cowonly_root(struct btrfs_trans_handle *trans,
			       struct btrfs_root *root)
C
Chris Mason 已提交
496 497
{
	int ret;
498
	u64 old_root_bytenr;
499
	u64 old_root_used;
500
	struct btrfs_root *tree_root = root->fs_info->tree_root;
C
Chris Mason 已提交
501

502
	old_root_used = btrfs_root_used(&root->root_item);
503
	btrfs_write_dirty_block_groups(trans, root);
504

C
Chris Mason 已提交
505
	while (1) {
506
		old_root_bytenr = btrfs_root_bytenr(&root->root_item);
507 508
		if (old_root_bytenr == root->node->start &&
		    old_root_used == btrfs_root_used(&root->root_item))
C
Chris Mason 已提交
509
			break;
510

511
		btrfs_set_root_node(&root->root_item, root->node);
C
Chris Mason 已提交
512
		ret = btrfs_update_root(trans, tree_root,
513 514
					&root->root_key,
					&root->root_item);
C
Chris Mason 已提交
515
		BUG_ON(ret);
516

517
		old_root_used = btrfs_root_used(&root->root_item);
518
		ret = btrfs_write_dirty_block_groups(trans, root);
519
		BUG_ON(ret);
520
	}
521 522 523 524

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

525 526 527
	return 0;
}

C
Chris Mason 已提交
528 529 530
/*
 * update all the cowonly tree roots on disk
 */
531 532
static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
					 struct btrfs_root *root)
533 534 535
{
	struct btrfs_fs_info *fs_info = root->fs_info;
	struct list_head *next;
536
	struct extent_buffer *eb;
537
	int ret;
538

539 540
	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
	BUG_ON(ret);
541

542
	eb = btrfs_lock_root_node(fs_info->tree_root);
543
	btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb);
544 545
	btrfs_tree_unlock(eb);
	free_extent_buffer(eb);
546

547 548
	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
	BUG_ON(ret);
549

C
Chris Mason 已提交
550
	while (!list_empty(&fs_info->dirty_cowonly_roots)) {
551 552 553
		next = fs_info->dirty_cowonly_roots.next;
		list_del_init(next);
		root = list_entry(next, struct btrfs_root, dirty_list);
554

555
		update_cowonly_root(trans, root);
C
Chris Mason 已提交
556
	}
557 558 559 560 561

	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
562 563 564
	return 0;
}

C
Chris Mason 已提交
565 566 567 568 569
/*
 * 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
 */
570
int btrfs_add_dead_root(struct btrfs_root *root)
571
{
Y
Yan Zheng 已提交
572
	mutex_lock(&root->fs_info->trans_mutex);
573
	list_add(&root->root_list, &root->fs_info->dead_roots);
Y
Yan Zheng 已提交
574
	mutex_unlock(&root->fs_info->trans_mutex);
575 576 577
	return 0;
}

C
Chris Mason 已提交
578
/*
579
 * update all the cowonly tree roots on disk
C
Chris Mason 已提交
580
 */
581 582
static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
				    struct btrfs_root *root)
583 584
{
	struct btrfs_root *gang[8];
585
	struct btrfs_fs_info *fs_info = root->fs_info;
586 587
	int i;
	int ret;
588 589
	int err = 0;

C
Chris Mason 已提交
590
	while (1) {
591 592
		ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
						 (void **)gang, 0,
593 594 595 596 597 598
						 ARRAY_SIZE(gang),
						 BTRFS_ROOT_TRANS_TAG);
		if (ret == 0)
			break;
		for (i = 0; i < ret; i++) {
			root = gang[i];
599 600 601
			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
602

603
			btrfs_free_log(trans, root);
604
			btrfs_update_reloc_root(trans, root);
605

606
			if (root->commit_root != root->node) {
J
Josef Bacik 已提交
607
				switch_commit_root(root);
608 609 610
				btrfs_set_root_node(&root->root_item,
						    root->node);
			}
611 612

			err = btrfs_update_root(trans, fs_info->tree_root,
613 614
						&root->root_key,
						&root->root_item);
615 616
			if (err)
				break;
617 618
		}
	}
619
	return err;
620 621
}

C
Chris Mason 已提交
622 623 624 625
/*
 * defrag a given btree.  If cacheonly == 1, this won't read from the disk,
 * otherwise every leaf in the btree is read and defragged.
 */
626 627 628 629 630
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;
631
	unsigned long nr;
632

633
	smp_mb();
634 635 636
	if (root->defrag_running)
		return 0;
	trans = btrfs_start_transaction(root, 1);
637
	while (1) {
638 639
		root->defrag_running = 1;
		ret = btrfs_defrag_leaves(trans, root, cacheonly);
640
		nr = trans->blocks_used;
641
		btrfs_end_transaction(trans, root);
642
		btrfs_btree_balance_dirty(info->tree_root, nr);
643 644 645
		cond_resched();

		trans = btrfs_start_transaction(root, 1);
646
		if (root->fs_info->closing || ret != -EAGAIN)
647 648 649
			break;
	}
	root->defrag_running = 0;
650
	smp_mb();
651 652 653 654
	btrfs_end_transaction(trans, root);
	return 0;
}

655
#if 0
656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672
/*
 * 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);
673

674
		schedule();
675

676 677 678 679 680 681 682
		mutex_lock(&info->trans_mutex);
		finish_wait(&info->transaction_wait, &wait);
	}
	mutex_unlock(&info->trans_mutex);
	return 0;
}

C
Chris Mason 已提交
683 684 685 686
/*
 * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
 * all of them
 */
687
int btrfs_drop_dead_root(struct btrfs_root *root)
688 689
{
	struct btrfs_trans_handle *trans;
690
	struct btrfs_root *tree_root = root->fs_info->tree_root;
691
	unsigned long nr;
692
	int ret;
693

694 695 696 697 698 699 700
	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);
701

702 703 704 705 706 707 708
		/*
		 * 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;
709
		}
710

711 712 713
		ret = btrfs_drop_snapshot(trans, root);
		if (ret != -EAGAIN)
			break;
714

715 716 717 718
		ret = btrfs_update_root(trans, tree_root,
					&root->root_key,
					&root->root_item);
		if (ret)
719
			break;
720

721
		nr = trans->blocks_used;
722 723
		ret = btrfs_end_transaction(trans, tree_root);
		BUG_ON(ret);
724

725
		btrfs_btree_balance_dirty(tree_root, nr);
726
		cond_resched();
727
	}
728 729 730 731 732 733 734 735 736 737 738 739 740 741
	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);
742
	return ret;
743
}
744
#endif
745

C
Chris Mason 已提交
746 747 748 749
/*
 * new snapshots need to be created at a very specific time in the
 * transaction commit.  This does the actual creation
 */
750
static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
751 752 753 754
				   struct btrfs_fs_info *fs_info,
				   struct btrfs_pending_snapshot *pending)
{
	struct btrfs_key key;
755
	struct btrfs_root_item *new_root_item;
756 757 758
	struct btrfs_root *tree_root = fs_info->tree_root;
	struct btrfs_root *root = pending->root;
	struct extent_buffer *tmp;
759
	struct extent_buffer *old;
760 761 762
	int ret;
	u64 objectid;

763 764 765 766 767
	new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
	if (!new_root_item) {
		ret = -ENOMEM;
		goto fail;
	}
768 769 770 771
	ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
	if (ret)
		goto fail;

772
	record_root_in_trans(trans, root);
Y
Yan Zheng 已提交
773
	btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
774
	memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
775 776

	key.objectid = objectid;
Y
Yan, Zheng 已提交
777 778
	/* record when the snapshot was created in key.offset */
	key.offset = trans->transid;
779 780
	btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);

781
	old = btrfs_lock_root_node(root);
782
	btrfs_cow_block(trans, root, old, NULL, 0, &old);
783
	btrfs_set_lock_blocking(old);
784

785 786 787
	btrfs_copy_root(trans, root, old, &tmp, objectid);
	btrfs_tree_unlock(old);
	free_extent_buffer(old);
788

789
	btrfs_set_root_node(new_root_item, tmp);
790
	ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
791
				new_root_item);
792
	btrfs_tree_unlock(tmp);
793 794 795 796
	free_extent_buffer(tmp);
	if (ret)
		goto fail;

797 798 799 800 801 802 803 804 805 806 807 808 809 810 811
	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;
812
	struct btrfs_root *parent_root;
813

814
	parent_inode = pending->dentry->d_parent->d_inode;
815
	parent_root = BTRFS_I(parent_inode)->root;
816
	trans = btrfs_join_transaction(parent_root, 1);
817

818 819 820
	/*
	 * insert the directory item
	 */
821
	namelen = strlen(pending->name);
822
	ret = btrfs_set_inode_index(parent_inode, &index);
823
	ret = btrfs_insert_dir_item(trans, parent_root,
824 825 826
			    pending->name, namelen,
			    parent_inode->i_ino,
			    &pending->root_key, BTRFS_FT_DIR, index);
827 828 829

	if (ret)
		goto fail;
830

831 832 833 834
	btrfs_i_size_write(parent_inode, parent_inode->i_size + namelen * 2);
	ret = btrfs_update_inode(trans, parent_root, parent_inode);
	BUG_ON(ret);

835 836 837 838 839 840 841 842
	ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root,
				 pending->root_key.objectid,
				 parent_root->root_key.objectid,
				 parent_inode->i_ino, index, pending->name,
				 namelen);

	BUG_ON(ret);

843
fail:
844
	btrfs_end_transaction(trans, fs_info->fs_root);
845 846 847
	return ret;
}

C
Chris Mason 已提交
848 849 850
/*
 * create all the snapshots we've scheduled for creation
 */
851 852
static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
					     struct btrfs_fs_info *fs_info)
853 854 855 856 857
{
	struct btrfs_pending_snapshot *pending;
	struct list_head *head = &trans->transaction->pending_snapshots;
	int ret;

Q
Qinghuang Feng 已提交
858
	list_for_each_entry(pending, head, list) {
859 860 861 862 863 864 865 866
		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)
867 868 869 870 871
{
	struct btrfs_pending_snapshot *pending;
	struct list_head *head = &trans->transaction->pending_snapshots;
	int ret;

C
Chris Mason 已提交
872
	while (!list_empty(head)) {
873 874
		pending = list_entry(head->next,
				     struct btrfs_pending_snapshot, list);
875
		ret = finish_pending_snapshot(fs_info, pending);
876 877 878 879 880
		BUG_ON(ret);
		list_del(&pending->list);
		kfree(pending->name);
		kfree(pending);
	}
C
Chris Mason 已提交
881 882 883
	return 0;
}

884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901
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;
}

902 903 904 905 906 907 908 909 910 911
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;
}

C
Chris Mason 已提交
912 913 914
int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
			     struct btrfs_root *root)
{
915 916
	unsigned long joined = 0;
	unsigned long timeout = 1;
C
Chris Mason 已提交
917
	struct btrfs_transaction *cur_trans;
C
Chris Mason 已提交
918
	struct btrfs_transaction *prev_trans = NULL;
C
Chris Mason 已提交
919
	DEFINE_WAIT(wait);
920
	int ret;
921 922
	int should_grow = 0;
	unsigned long now = get_seconds();
923
	int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
C
Chris Mason 已提交
924

925 926
	btrfs_run_ordered_operations(root, 0);

927 928 929 930 931 932
	/* 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);

933
	cur_trans = trans->transaction;
934 935 936 937
	/*
	 * set the flushing flag so procs in this transaction have to
	 * start sending their work down.
	 */
938
	cur_trans->delayed_refs.flushing = 1;
939

940
	ret = btrfs_run_delayed_refs(trans, root, 0);
941 942
	BUG_ON(ret);

C
Chris Mason 已提交
943
	mutex_lock(&root->fs_info->trans_mutex);
944 945
	if (cur_trans->in_commit) {
		cur_trans->use_count++;
C
Chris Mason 已提交
946
		mutex_unlock(&root->fs_info->trans_mutex);
C
Chris Mason 已提交
947
		btrfs_end_transaction(trans, root);
C
Chris Mason 已提交
948

C
Chris Mason 已提交
949 950
		ret = wait_for_commit(root, cur_trans);
		BUG_ON(ret);
951 952

		mutex_lock(&root->fs_info->trans_mutex);
C
Chris Mason 已提交
953
		put_transaction(cur_trans);
954 955
		mutex_unlock(&root->fs_info->trans_mutex);

C
Chris Mason 已提交
956 957
		return 0;
	}
958

C
Chris Mason 已提交
959
	trans->transaction->in_commit = 1;
960
	trans->transaction->blocked = 1;
C
Chris Mason 已提交
961 962 963 964 965 966 967 968 969 970
	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);
971
			put_transaction(prev_trans);
C
Chris Mason 已提交
972 973
		}
	}
974

975 976 977
	if (now < cur_trans->start_time || now - cur_trans->start_time < 1)
		should_grow = 1;

978
	do {
979
		int snap_pending = 0;
980
		joined = cur_trans->num_joined;
981 982 983
		if (!list_empty(&trans->transaction->pending_snapshots))
			snap_pending = 1;

C
Chris Mason 已提交
984
		WARN_ON(cur_trans != trans->transaction);
985
		prepare_to_wait(&cur_trans->writer_wait, &wait,
C
Chris Mason 已提交
986
				TASK_UNINTERRUPTIBLE);
987 988 989

		if (cur_trans->num_writers > 1)
			timeout = MAX_SCHEDULE_TIMEOUT;
990
		else if (should_grow)
991 992
			timeout = 1;

C
Chris Mason 已提交
993
		mutex_unlock(&root->fs_info->trans_mutex);
994

995
		if (flush_on_commit || snap_pending) {
Y
Yan, Zheng 已提交
996 997
			btrfs_start_delalloc_inodes(root, 1);
			ret = btrfs_wait_ordered_extents(root, 0, 1);
998
			BUG_ON(ret);
999 1000
		}

1001 1002 1003 1004 1005 1006 1007 1008 1009
		/*
		 * 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);

1010 1011 1012
		smp_mb();
		if (cur_trans->num_writers > 1 || should_grow)
			schedule_timeout(timeout);
1013

C
Chris Mason 已提交
1014
		mutex_lock(&root->fs_info->trans_mutex);
1015 1016
		finish_wait(&cur_trans->writer_wait, &wait);
	} while (cur_trans->num_writers > 1 ||
1017
		 (should_grow && cur_trans->num_joined != joined));
1018

1019 1020 1021
	ret = create_pending_snapshots(trans, root->fs_info);
	BUG_ON(ret);

1022 1023 1024
	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
	BUG_ON(ret);

C
Chris Mason 已提交
1025
	WARN_ON(cur_trans != trans->transaction);
C
Chris Mason 已提交
1026

1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041
	/* 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);

1042
	ret = commit_fs_roots(trans, root);
1043 1044
	BUG_ON(ret);

1045
	/* commit_fs_roots gets rid of all the tree log roots, it is now
1046 1047 1048 1049
	 * safe to free the root of tree log roots
	 */
	btrfs_free_log_root_tree(trans, root->fs_info);

1050
	ret = commit_cowonly_roots(trans, root);
C
Chris Mason 已提交
1051
	BUG_ON(ret);
1052

1053 1054
	btrfs_prepare_extent_commit(trans, root);

C
Chris Mason 已提交
1055
	cur_trans = root->fs_info->running_transaction;
1056
	spin_lock(&root->fs_info->new_trans_lock);
C
Chris Mason 已提交
1057
	root->fs_info->running_transaction = NULL;
1058
	spin_unlock(&root->fs_info->new_trans_lock);
1059 1060 1061

	btrfs_set_root_node(&root->fs_info->tree_root->root_item,
			    root->fs_info->tree_root->node);
J
Josef Bacik 已提交
1062
	switch_commit_root(root->fs_info->tree_root);
1063 1064 1065

	btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
			    root->fs_info->chunk_root->node);
J
Josef Bacik 已提交
1066
	switch_commit_root(root->fs_info->chunk_root);
1067 1068

	update_super_roots(root);
1069 1070 1071 1072 1073 1074

	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);
	}

1075 1076
	memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
	       sizeof(root->fs_info->super_copy));
C
Chris Mason 已提交
1077

1078
	trans->transaction->blocked = 0;
1079

1080
	wake_up(&root->fs_info->transaction_wait);
1081

C
Chris Mason 已提交
1082
	mutex_unlock(&root->fs_info->trans_mutex);
C
Chris Mason 已提交
1083 1084
	ret = btrfs_write_and_wait_transaction(trans, root);
	BUG_ON(ret);
Y
Yan Zheng 已提交
1085
	write_ctree_super(trans, root, 0);
1086

1087 1088 1089 1090 1091 1092
	/*
	 * the super is written, we can safely allow the tree-loggers
	 * to go about their business
	 */
	mutex_unlock(&root->fs_info->tree_log_mutex);

1093
	btrfs_finish_extent_commit(trans, root);
1094

1095 1096 1097
	/* do the directory inserts of any pending snapshot creations */
	finish_pending_snapshots(trans, root->fs_info);

Z
Zheng Yan 已提交
1098 1099
	mutex_lock(&root->fs_info->trans_mutex);

C
Chris Mason 已提交
1100
	cur_trans->commit_done = 1;
1101

1102
	root->fs_info->last_trans_committed = cur_trans->transid;
J
Josef Bacik 已提交
1103

C
Chris Mason 已提交
1104
	wake_up(&cur_trans->commit_wait);
1105

C
Chris Mason 已提交
1106
	put_transaction(cur_trans);
C
Chris Mason 已提交
1107
	put_transaction(cur_trans);
1108

C
Chris Mason 已提交
1109
	mutex_unlock(&root->fs_info->trans_mutex);
1110

J
Josef Bacik 已提交
1111 1112 1113
	if (current->journal_info == trans)
		current->journal_info = NULL;

C
Chris Mason 已提交
1114
	kmem_cache_free(btrfs_trans_handle_cachep, trans);
Y
Yan, Zheng 已提交
1115 1116 1117 1118

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

C
Chris Mason 已提交
1119 1120 1121
	return ret;
}

C
Chris Mason 已提交
1122 1123 1124
/*
 * interface function to delete all the snapshots we have scheduled for deletion
 */
1125 1126
int btrfs_clean_old_snapshots(struct btrfs_root *root)
{
1127 1128 1129 1130 1131 1132
	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);
1133

1134 1135
	while (!list_empty(&list)) {
		root = list_entry(list.next, struct btrfs_root, root_list);
1136 1137 1138 1139 1140 1141 1142
		list_del(&root->root_list);

		if (btrfs_header_backref_rev(root->node) <
		    BTRFS_MIXED_BACKREF_REV)
			btrfs_drop_snapshot(root, 0);
		else
			btrfs_drop_snapshot(root, 1);
1143 1144 1145
	}
	return 0;
}