transaction.c 44.5 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>
25
#include <linux/uuid.h>
C
Chris Mason 已提交
26 27 28
#include "ctree.h"
#include "disk-io.h"
#include "transaction.h"
29
#include "locking.h"
30
#include "tree-log.h"
31
#include "inode-map.h"
32
#include "volumes.h"
C
Chris Mason 已提交
33

34 35
#define BTRFS_ROOT_TRANS_TAG 0

36
void put_transaction(struct btrfs_transaction *transaction)
C
Chris Mason 已提交
37
{
38 39
	WARN_ON(atomic_read(&transaction->use_count) == 0);
	if (atomic_dec_and_test(&transaction->use_count)) {
J
Josef Bacik 已提交
40
		BUG_ON(!list_empty(&transaction->list));
41
		WARN_ON(transaction->delayed_refs.root.rb_node);
C
Chris Mason 已提交
42 43
		memset(transaction, 0, sizeof(*transaction));
		kmem_cache_free(btrfs_transaction_cachep, transaction);
C
Chris Mason 已提交
44
	}
C
Chris Mason 已提交
45 46
}

J
Josef Bacik 已提交
47 48 49 50 51 52
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
53 54 55
/*
 * either allocate a new transaction or hop into the existing one
 */
J
Josef Bacik 已提交
56
static noinline int join_transaction(struct btrfs_root *root, int nofail)
C
Chris Mason 已提交
57 58
{
	struct btrfs_transaction *cur_trans;
59
	struct btrfs_fs_info *fs_info = root->fs_info;
J
Josef Bacik 已提交
60

61
	spin_lock(&fs_info->trans_lock);
62
loop:
63
	/* The file system has been taken offline. No new transactions. */
64 65
	if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
		spin_unlock(&fs_info->trans_lock);
66 67 68
		return -EROFS;
	}

69
	if (fs_info->trans_no_join) {
J
Josef Bacik 已提交
70
		if (!nofail) {
71
			spin_unlock(&fs_info->trans_lock);
J
Josef Bacik 已提交
72 73 74 75
			return -EBUSY;
		}
	}

76
	cur_trans = fs_info->running_transaction;
J
Josef Bacik 已提交
77
	if (cur_trans) {
78
		if (cur_trans->aborted) {
79
			spin_unlock(&fs_info->trans_lock);
80
			return cur_trans->aborted;
81
		}
J
Josef Bacik 已提交
82
		atomic_inc(&cur_trans->use_count);
83
		atomic_inc(&cur_trans->num_writers);
84
		cur_trans->num_joined++;
85
		spin_unlock(&fs_info->trans_lock);
J
Josef Bacik 已提交
86
		return 0;
C
Chris Mason 已提交
87
	}
88
	spin_unlock(&fs_info->trans_lock);
J
Josef Bacik 已提交
89 90 91 92

	cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS);
	if (!cur_trans)
		return -ENOMEM;
93

94 95
	spin_lock(&fs_info->trans_lock);
	if (fs_info->running_transaction) {
96 97 98 99
		/*
		 * someone started a transaction after we unlocked.  Make sure
		 * to redo the trans_no_join checks above
		 */
J
Josef Bacik 已提交
100
		kmem_cache_free(btrfs_transaction_cachep, cur_trans);
101
		cur_trans = fs_info->running_transaction;
102
		goto loop;
103 104
	} else if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
		spin_unlock(&fs_info->trans_lock);
105 106
		kmem_cache_free(btrfs_transaction_cachep, cur_trans);
		return -EROFS;
C
Chris Mason 已提交
107
	}
108

J
Josef Bacik 已提交
109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128
	atomic_set(&cur_trans->num_writers, 1);
	cur_trans->num_joined = 0;
	init_waitqueue_head(&cur_trans->writer_wait);
	init_waitqueue_head(&cur_trans->commit_wait);
	cur_trans->in_commit = 0;
	cur_trans->blocked = 0;
	/*
	 * One for this trans handle, one so it will live on until we
	 * commit the transaction.
	 */
	atomic_set(&cur_trans->use_count, 2);
	cur_trans->commit_done = 0;
	cur_trans->start_time = get_seconds();

	cur_trans->delayed_refs.root = RB_ROOT;
	cur_trans->delayed_refs.num_entries = 0;
	cur_trans->delayed_refs.num_heads_ready = 0;
	cur_trans->delayed_refs.num_heads = 0;
	cur_trans->delayed_refs.flushing = 0;
	cur_trans->delayed_refs.run_delayed_start = 0;
129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146

	/*
	 * although the tree mod log is per file system and not per transaction,
	 * the log must never go across transaction boundaries.
	 */
	smp_mb();
	if (!list_empty(&fs_info->tree_mod_seq_list)) {
		printk(KERN_ERR "btrfs: tree_mod_seq_list not empty when "
			"creating a fresh transaction\n");
		WARN_ON(1);
	}
	if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) {
		printk(KERN_ERR "btrfs: tree_mod_log rb tree not empty when "
			"creating a fresh transaction\n");
		WARN_ON(1);
	}
	atomic_set(&fs_info->tree_mod_seq, 0);

J
Josef Bacik 已提交
147 148 149 150
	spin_lock_init(&cur_trans->commit_lock);
	spin_lock_init(&cur_trans->delayed_refs.lock);

	INIT_LIST_HEAD(&cur_trans->pending_snapshots);
151
	list_add_tail(&cur_trans->list, &fs_info->trans_list);
J
Josef Bacik 已提交
152
	extent_io_tree_init(&cur_trans->dirty_pages,
153 154 155 156
			     fs_info->btree_inode->i_mapping);
	fs_info->generation++;
	cur_trans->transid = fs_info->generation;
	fs_info->running_transaction = cur_trans;
157
	cur_trans->aborted = 0;
158
	spin_unlock(&fs_info->trans_lock);
159

C
Chris Mason 已提交
160 161 162
	return 0;
}

C
Chris Mason 已提交
163
/*
C
Chris Mason 已提交
164 165 166 167
 * 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
168
 */
C
Chris Mason 已提交
169
static int record_root_in_trans(struct btrfs_trans_handle *trans,
J
Josef Bacik 已提交
170
			       struct btrfs_root *root)
171
{
172
	if (root->ref_cows && root->last_trans < trans->transid) {
173
		WARN_ON(root == root->fs_info->extent_root);
174 175
		WARN_ON(root->commit_root != root->node);

C
Chris Mason 已提交
176 177 178 179 180 181 182 183 184 185 186 187
		/*
		 * see below for in_trans_setup usage rules
		 * we have the reloc mutex held now, so there
		 * is only one writer in this function
		 */
		root->in_trans_setup = 1;

		/* make sure readers find in_trans_setup before
		 * they find our root->last_trans update
		 */
		smp_wmb();

J
Josef Bacik 已提交
188 189 190 191 192
		spin_lock(&root->fs_info->fs_roots_radix_lock);
		if (root->last_trans == trans->transid) {
			spin_unlock(&root->fs_info->fs_roots_radix_lock);
			return 0;
		}
193 194 195
		radix_tree_tag_set(&root->fs_info->fs_roots_radix,
			   (unsigned long)root->root_key.objectid,
			   BTRFS_ROOT_TRANS_TAG);
J
Josef Bacik 已提交
196
		spin_unlock(&root->fs_info->fs_roots_radix_lock);
C
Chris Mason 已提交
197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217
		root->last_trans = trans->transid;

		/* this is pretty tricky.  We don't want to
		 * take the relocation lock in btrfs_record_root_in_trans
		 * unless we're really doing the first setup for this root in
		 * this transaction.
		 *
		 * Normally we'd use root->last_trans as a flag to decide
		 * if we want to take the expensive mutex.
		 *
		 * But, we have to set root->last_trans before we
		 * init the relocation root, otherwise, we trip over warnings
		 * in ctree.c.  The solution used here is to flag ourselves
		 * with root->in_trans_setup.  When this is 1, we're still
		 * fixing up the reloc trees and everyone must wait.
		 *
		 * When this is zero, they can trust root->last_trans and fly
		 * through btrfs_record_root_in_trans without having to take the
		 * lock.  smp_wmb() makes sure that all the writes above are
		 * done before we pop in the zero below
		 */
218
		btrfs_init_reloc_root(trans, root);
C
Chris Mason 已提交
219 220
		smp_wmb();
		root->in_trans_setup = 0;
221 222 223
	}
	return 0;
}
224

C
Chris Mason 已提交
225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247

int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
			       struct btrfs_root *root)
{
	if (!root->ref_cows)
		return 0;

	/*
	 * see record_root_in_trans for comments about in_trans_setup usage
	 * and barriers
	 */
	smp_rmb();
	if (root->last_trans == trans->transid &&
	    !root->in_trans_setup)
		return 0;

	mutex_lock(&root->fs_info->reloc_mutex);
	record_root_in_trans(trans, root);
	mutex_unlock(&root->fs_info->reloc_mutex);

	return 0;
}

C
Chris Mason 已提交
248 249 250 251
/* 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
252
static void wait_current_trans(struct btrfs_root *root)
C
Chris Mason 已提交
253
{
254
	struct btrfs_transaction *cur_trans;
C
Chris Mason 已提交
255

J
Josef Bacik 已提交
256
	spin_lock(&root->fs_info->trans_lock);
257
	cur_trans = root->fs_info->running_transaction;
C
Chris Mason 已提交
258
	if (cur_trans && cur_trans->blocked) {
259
		atomic_inc(&cur_trans->use_count);
J
Josef Bacik 已提交
260
		spin_unlock(&root->fs_info->trans_lock);
L
Li Zefan 已提交
261 262 263

		wait_event(root->fs_info->transaction_wait,
			   !cur_trans->blocked);
264
		put_transaction(cur_trans);
J
Josef Bacik 已提交
265 266
	} else {
		spin_unlock(&root->fs_info->trans_lock);
267
	}
C
Chris Mason 已提交
268 269
}

270 271 272 273
enum btrfs_trans_type {
	TRANS_START,
	TRANS_JOIN,
	TRANS_USERSPACE,
274
	TRANS_JOIN_NOLOCK,
275 276
};

277 278
static int may_wait_transaction(struct btrfs_root *root, int type)
{
J
Josef Bacik 已提交
279 280 281 282 283 284 285 286
	if (root->fs_info->log_root_recovering)
		return 0;

	if (type == TRANS_USERSPACE)
		return 1;

	if (type == TRANS_START &&
	    !atomic_read(&root->fs_info->open_ioctl_trans))
287
		return 1;
J
Josef Bacik 已提交
288

289 290 291
	return 0;
}

292
static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
293 294
						    u64 num_items, int type,
						    int noflush)
C
Chris Mason 已提交
295
{
296 297
	struct btrfs_trans_handle *h;
	struct btrfs_transaction *cur_trans;
298
	u64 num_bytes = 0;
C
Chris Mason 已提交
299
	int ret;
300
	u64 qgroup_reserved = 0;
L
liubo 已提交
301 302 303

	if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
		return ERR_PTR(-EROFS);
304 305 306 307 308 309 310 311 312

	if (current->journal_info) {
		WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK);
		h = current->journal_info;
		h->use_count++;
		h->orig_rsv = h->block_rsv;
		h->block_rsv = NULL;
		goto got_it;
	}
313 314 315 316 317 318

	/*
	 * Do the reservation before we join the transaction so we can do all
	 * the appropriate flushing if need be.
	 */
	if (num_items > 0 && root != root->fs_info->chunk_root) {
319 320 321 322 323 324 325 326
		if (root->fs_info->quota_enabled &&
		    is_fstree(root->root_key.objectid)) {
			qgroup_reserved = num_items * root->leafsize;
			ret = btrfs_qgroup_reserve(root, qgroup_reserved);
			if (ret)
				return ERR_PTR(ret);
		}

327
		num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
328 329 330 331 332 333 334 335
		if (noflush)
			ret = btrfs_block_rsv_add_noflush(root,
						&root->fs_info->trans_block_rsv,
						num_bytes);
		else
			ret = btrfs_block_rsv_add(root,
						&root->fs_info->trans_block_rsv,
						num_bytes);
336 337 338
		if (ret)
			return ERR_PTR(ret);
	}
339 340 341 342
again:
	h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
	if (!h)
		return ERR_PTR(-ENOMEM);
C
Chris Mason 已提交
343

344 345
	sb_start_intwrite(root->fs_info->sb);

346
	if (may_wait_transaction(root, type))
C
Chris Mason 已提交
347
		wait_current_trans(root);
348

J
Josef Bacik 已提交
349 350 351 352 353 354
	do {
		ret = join_transaction(root, type == TRANS_JOIN_NOLOCK);
		if (ret == -EBUSY)
			wait_current_trans(root);
	} while (ret == -EBUSY);

T
Tsutomu Itoh 已提交
355
	if (ret < 0) {
356
		sb_end_intwrite(root->fs_info->sb);
357
		kmem_cache_free(btrfs_trans_handle_cachep, h);
T
Tsutomu Itoh 已提交
358 359
		return ERR_PTR(ret);
	}
360

361 362 363 364
	cur_trans = root->fs_info->running_transaction;

	h->transid = cur_trans->transid;
	h->transaction = cur_trans;
C
Chris Mason 已提交
365
	h->blocks_used = 0;
366
	h->bytes_reserved = 0;
367
	h->root = root;
368
	h->delayed_ref_updates = 0;
369
	h->use_count = 1;
370
	h->adding_csums = 0;
371
	h->block_rsv = NULL;
372
	h->orig_rsv = NULL;
373
	h->aborted = 0;
374
	h->qgroup_reserved = qgroup_reserved;
375 376
	h->delayed_ref_elem.seq = 0;
	INIT_LIST_HEAD(&h->qgroup_ref_list);
377

378 379 380 381 382 383
	smp_mb();
	if (cur_trans->blocked && may_wait_transaction(root, type)) {
		btrfs_commit_transaction(h, root);
		goto again;
	}

384
	if (num_bytes) {
J
Josef Bacik 已提交
385
		trace_btrfs_space_reservation(root->fs_info, "transaction",
386
					      h->transid, num_bytes, 1);
387 388
		h->block_rsv = &root->fs_info->trans_block_rsv;
		h->bytes_reserved = num_bytes;
389
	}
J
Josef Bacik 已提交
390

391
got_it:
J
Josef Bacik 已提交
392
	btrfs_record_root_in_trans(h, root);
393 394 395

	if (!current->journal_info && type != TRANS_USERSPACE)
		current->journal_info = h;
C
Chris Mason 已提交
396 397 398
	return h;
}

399
struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
400
						   int num_items)
401
{
402
	return start_transaction(root, num_items, TRANS_START, 0);
403
}
404 405 406 407 408 409 410

struct btrfs_trans_handle *btrfs_start_transaction_noflush(
					struct btrfs_root *root, int num_items)
{
	return start_transaction(root, num_items, TRANS_START, 1);
}

411
struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
412
{
413
	return start_transaction(root, 0, TRANS_JOIN, 0);
414 415
}

416
struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
417
{
418
	return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0);
419 420
}

421
struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
422
{
423
	return start_transaction(root, 0, TRANS_USERSPACE, 0);
424 425
}

C
Chris Mason 已提交
426
/* wait for a transaction commit to be fully complete */
427
static noinline void wait_for_commit(struct btrfs_root *root,
428 429
				    struct btrfs_transaction *commit)
{
L
Li Zefan 已提交
430
	wait_event(commit->commit_wait, commit->commit_done);
431 432
}

433 434 435 436 437 438 439 440
int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
{
	struct btrfs_transaction *cur_trans = NULL, *t;
	int ret;

	ret = 0;
	if (transid) {
		if (transid <= root->fs_info->last_trans_committed)
J
Josef Bacik 已提交
441
			goto out;
442 443

		/* find specified transaction */
J
Josef Bacik 已提交
444
		spin_lock(&root->fs_info->trans_lock);
445 446 447
		list_for_each_entry(t, &root->fs_info->trans_list, list) {
			if (t->transid == transid) {
				cur_trans = t;
J
Josef Bacik 已提交
448
				atomic_inc(&cur_trans->use_count);
449 450 451 452 453
				break;
			}
			if (t->transid > transid)
				break;
		}
J
Josef Bacik 已提交
454
		spin_unlock(&root->fs_info->trans_lock);
455 456
		ret = -EINVAL;
		if (!cur_trans)
J
Josef Bacik 已提交
457
			goto out;  /* bad transid */
458 459
	} else {
		/* find newest transaction that is committing | committed */
J
Josef Bacik 已提交
460
		spin_lock(&root->fs_info->trans_lock);
461 462 463 464
		list_for_each_entry_reverse(t, &root->fs_info->trans_list,
					    list) {
			if (t->in_commit) {
				if (t->commit_done)
465
					break;
466
				cur_trans = t;
J
Josef Bacik 已提交
467
				atomic_inc(&cur_trans->use_count);
468 469 470
				break;
			}
		}
J
Josef Bacik 已提交
471
		spin_unlock(&root->fs_info->trans_lock);
472
		if (!cur_trans)
J
Josef Bacik 已提交
473
			goto out;  /* nothing committing|committed */
474 475 476 477 478 479
	}

	wait_for_commit(root, cur_trans);

	put_transaction(cur_trans);
	ret = 0;
J
Josef Bacik 已提交
480
out:
481 482 483
	return ret;
}

C
Chris Mason 已提交
484 485
void btrfs_throttle(struct btrfs_root *root)
{
J
Josef Bacik 已提交
486
	if (!atomic_read(&root->fs_info->open_ioctl_trans))
487
		wait_current_trans(root);
C
Chris Mason 已提交
488 489
}

490 491 492 493
static int should_end_transaction(struct btrfs_trans_handle *trans,
				  struct btrfs_root *root)
{
	int ret;
494 495

	ret = btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
496 497 498 499 500 501 502 503
	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;
504
	int err;
505

J
Josef Bacik 已提交
506
	smp_mb();
507 508 509 510 511
	if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
		return 1;

	updates = trans->delayed_ref_updates;
	trans->delayed_ref_updates = 0;
512 513 514 515 516
	if (updates) {
		err = btrfs_run_delayed_refs(trans, root, updates);
		if (err) /* Error code will also eval true */
			return err;
	}
517 518 519 520

	return should_end_transaction(trans, root);
}

521
static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
522
			  struct btrfs_root *root, int throttle, int lock)
C
Chris Mason 已提交
523
{
524
	struct btrfs_transaction *cur_trans = trans->transaction;
525
	struct btrfs_fs_info *info = root->fs_info;
526
	int count = 0;
527
	int err = 0;
528

529 530 531 532 533
	if (--trans->use_count) {
		trans->block_rsv = trans->orig_rsv;
		return 0;
	}

534 535 536 537 538
	/*
	 * do the qgroup accounting as early as possible
	 */
	err = btrfs_delayed_refs_qgroup_accounting(trans, info);

539
	btrfs_trans_release_metadata(trans, root);
540
	trans->block_rsv = NULL;
541 542 543 544 545
	/*
	 * the same root has to be passed to start_transaction and
	 * end_transaction. Subvolume quota depends on this.
	 */
	WARN_ON(trans->root != root);
546 547 548 549 550 551

	if (trans->qgroup_reserved) {
		btrfs_qgroup_free(root, trans->qgroup_reserved);
		trans->qgroup_reserved = 0;
	}

552
	while (count < 2) {
553 554 555 556 557 558 559 560 561 562
		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;
			btrfs_run_delayed_refs(trans, root, cur);
		} else {
			break;
		}
		count++;
563
	}
564 565
	btrfs_trans_release_metadata(trans, root);
	trans->block_rsv = NULL;
566

J
Josef Bacik 已提交
567 568
	if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
	    should_end_transaction(trans, root)) {
569
		trans->transaction->blocked = 1;
J
Josef Bacik 已提交
570 571
		smp_wmb();
	}
572

573
	if (lock && cur_trans->blocked && !cur_trans->in_commit) {
574 575 576 577 578 579 580
		if (throttle) {
			/*
			 * We may race with somebody else here so end up having
			 * to call end_transaction on ourselves again, so inc
			 * our use_count.
			 */
			trans->use_count++;
581
			return btrfs_commit_transaction(trans, root);
582
		} else {
583
			wake_up_process(info->transaction_kthread);
584
		}
585 586
	}

587 588
	sb_end_intwrite(root->fs_info->sb);

589
	WARN_ON(cur_trans != info->running_transaction);
590 591
	WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
	atomic_dec(&cur_trans->num_writers);
592

593
	smp_mb();
C
Chris Mason 已提交
594 595 596
	if (waitqueue_active(&cur_trans->writer_wait))
		wake_up(&cur_trans->writer_wait);
	put_transaction(cur_trans);
J
Josef Bacik 已提交
597 598 599

	if (current->journal_info == trans)
		current->journal_info = NULL;
600

Y
Yan, Zheng 已提交
601 602 603
	if (throttle)
		btrfs_run_delayed_iputs(root);

604 605
	if (trans->aborted ||
	    root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
606
		err = -EIO;
607
	}
608
	assert_qgroups_uptodate(trans);
609

610 611 612
	memset(trans, 0, sizeof(*trans));
	kmem_cache_free(btrfs_trans_handle_cachep, trans);
	return err;
C
Chris Mason 已提交
613 614
}

615 616 617
int btrfs_end_transaction(struct btrfs_trans_handle *trans,
			  struct btrfs_root *root)
{
618 619 620 621 622 623
	int ret;

	ret = __btrfs_end_transaction(trans, root, 0, 1);
	if (ret)
		return ret;
	return 0;
624 625 626 627 628
}

int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
				   struct btrfs_root *root)
{
629 630 631 632 633 634
	int ret;

	ret = __btrfs_end_transaction(trans, root, 1, 1);
	if (ret)
		return ret;
	return 0;
635 636 637 638 639
}

int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans,
				 struct btrfs_root *root)
{
640 641 642 643 644 645 646 647 648 649 650 651
	int ret;

	ret = __btrfs_end_transaction(trans, root, 0, 0);
	if (ret)
		return ret;
	return 0;
}

int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans,
				struct btrfs_root *root)
{
	return __btrfs_end_transaction(trans, root, 1, 1);
652 653
}

C
Chris Mason 已提交
654 655 656
/*
 * 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
657
 * those extents are sent to disk but does not wait on them
C
Chris Mason 已提交
658
 */
659
int btrfs_write_marked_extents(struct btrfs_root *root,
660
			       struct extent_io_tree *dirty_pages, int mark)
C
Chris Mason 已提交
661
{
662
	int err = 0;
663
	int werr = 0;
J
Josef Bacik 已提交
664
	struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
665
	u64 start = 0;
666
	u64 end;
667

J
Josef Bacik 已提交
668 669 670 671 672 673 674 675 676
	while (!find_first_extent_bit(dirty_pages, start, &start, &end,
				      mark)) {
		convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, mark,
				   GFP_NOFS);
		err = filemap_fdatawrite_range(mapping, start, end);
		if (err)
			werr = err;
		cond_resched();
		start = end + 1;
677
	}
678 679 680 681 682 683 684 685 686 687 688 689
	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,
690
			      struct extent_io_tree *dirty_pages, int mark)
691 692 693
{
	int err = 0;
	int werr = 0;
J
Josef Bacik 已提交
694
	struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
695 696
	u64 start = 0;
	u64 end;
697

J
Josef Bacik 已提交
698 699 700 701 702 703 704 705
	while (!find_first_extent_bit(dirty_pages, start, &start, &end,
				      EXTENT_NEED_WAIT)) {
		clear_extent_bits(dirty_pages, start, end, EXTENT_NEED_WAIT, GFP_NOFS);
		err = filemap_fdatawait_range(mapping, start, end);
		if (err)
			werr = err;
		cond_resched();
		start = end + 1;
706
	}
707 708 709
	if (err)
		werr = err;
	return werr;
C
Chris Mason 已提交
710 711
}

712 713 714 715 716 717
/*
 * 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,
718
				struct extent_io_tree *dirty_pages, int mark)
719 720 721 722
{
	int ret;
	int ret2;

723 724
	ret = btrfs_write_marked_extents(root, dirty_pages, mark);
	ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
725 726 727 728 729 730

	if (ret)
		return ret;
	if (ret2)
		return ret2;
	return 0;
731 732
}

733 734 735 736 737 738 739 740 741
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,
742 743
					   &trans->transaction->dirty_pages,
					   EXTENT_DIRTY);
744 745
}

C
Chris Mason 已提交
746 747 748 749 750 751 752 753 754 755
/*
 * 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.
 */
756 757
static int update_cowonly_root(struct btrfs_trans_handle *trans,
			       struct btrfs_root *root)
C
Chris Mason 已提交
758 759
{
	int ret;
760
	u64 old_root_bytenr;
761
	u64 old_root_used;
762
	struct btrfs_root *tree_root = root->fs_info->tree_root;
C
Chris Mason 已提交
763

764
	old_root_used = btrfs_root_used(&root->root_item);
765
	btrfs_write_dirty_block_groups(trans, root);
766

C
Chris Mason 已提交
767
	while (1) {
768
		old_root_bytenr = btrfs_root_bytenr(&root->root_item);
769 770
		if (old_root_bytenr == root->node->start &&
		    old_root_used == btrfs_root_used(&root->root_item))
C
Chris Mason 已提交
771
			break;
772

773
		btrfs_set_root_node(&root->root_item, root->node);
C
Chris Mason 已提交
774
		ret = btrfs_update_root(trans, tree_root,
775 776
					&root->root_key,
					&root->root_item);
777 778
		if (ret)
			return ret;
779

780
		old_root_used = btrfs_root_used(&root->root_item);
781
		ret = btrfs_write_dirty_block_groups(trans, root);
782 783
		if (ret)
			return ret;
784
	}
785 786 787 788

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

789 790 791
	return 0;
}

C
Chris Mason 已提交
792 793
/*
 * update all the cowonly tree roots on disk
794 795 796 797
 *
 * The error handling in this function may not be obvious. Any of the
 * failures will cause the file system to go offline. We still need
 * to clean up the delayed refs.
C
Chris Mason 已提交
798
 */
799 800
static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
					 struct btrfs_root *root)
801 802 803
{
	struct btrfs_fs_info *fs_info = root->fs_info;
	struct list_head *next;
804
	struct extent_buffer *eb;
805
	int ret;
806

807
	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
808 809
	if (ret)
		return ret;
810

811
	eb = btrfs_lock_root_node(fs_info->tree_root);
812 813
	ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
			      0, &eb);
814 815
	btrfs_tree_unlock(eb);
	free_extent_buffer(eb);
816

817 818 819
	if (ret)
		return ret;

820
	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
821 822
	if (ret)
		return ret;
823

824 825 826
	ret = btrfs_run_dev_stats(trans, root->fs_info);
	BUG_ON(ret);

827 828 829 830 831 832 833
	ret = btrfs_run_qgroups(trans, root->fs_info);
	BUG_ON(ret);

	/* run_qgroups might have added some more refs */
	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
	BUG_ON(ret);

C
Chris Mason 已提交
834
	while (!list_empty(&fs_info->dirty_cowonly_roots)) {
835 836 837
		next = fs_info->dirty_cowonly_roots.next;
		list_del_init(next);
		root = list_entry(next, struct btrfs_root, dirty_list);
838

839 840 841
		ret = update_cowonly_root(trans, root);
		if (ret)
			return ret;
C
Chris Mason 已提交
842
	}
843 844 845 846 847

	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
848 849 850
	return 0;
}

C
Chris Mason 已提交
851 852 853 854 855
/*
 * 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
 */
856
int btrfs_add_dead_root(struct btrfs_root *root)
857
{
J
Josef Bacik 已提交
858
	spin_lock(&root->fs_info->trans_lock);
859
	list_add(&root->root_list, &root->fs_info->dead_roots);
J
Josef Bacik 已提交
860
	spin_unlock(&root->fs_info->trans_lock);
861 862 863
	return 0;
}

C
Chris Mason 已提交
864
/*
865
 * update all the cowonly tree roots on disk
C
Chris Mason 已提交
866
 */
867 868
static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
				    struct btrfs_root *root)
869 870
{
	struct btrfs_root *gang[8];
871
	struct btrfs_fs_info *fs_info = root->fs_info;
872 873
	int i;
	int ret;
874 875
	int err = 0;

J
Josef Bacik 已提交
876
	spin_lock(&fs_info->fs_roots_radix_lock);
C
Chris Mason 已提交
877
	while (1) {
878 879
		ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
						 (void **)gang, 0,
880 881 882 883 884 885
						 ARRAY_SIZE(gang),
						 BTRFS_ROOT_TRANS_TAG);
		if (ret == 0)
			break;
		for (i = 0; i < ret; i++) {
			root = gang[i];
886 887 888
			radix_tree_tag_clear(&fs_info->fs_roots_radix,
					(unsigned long)root->root_key.objectid,
					BTRFS_ROOT_TRANS_TAG);
J
Josef Bacik 已提交
889
			spin_unlock(&fs_info->fs_roots_radix_lock);
Y
Yan Zheng 已提交
890

891
			btrfs_free_log(trans, root);
892
			btrfs_update_reloc_root(trans, root);
893
			btrfs_orphan_commit_root(trans, root);
894

895 896
			btrfs_save_ino_cache(root, trans);

897 898 899 900
			/* see comments in should_cow_block() */
			root->force_cow = 0;
			smp_wmb();

901
			if (root->commit_root != root->node) {
902
				mutex_lock(&root->fs_commit_mutex);
J
Josef Bacik 已提交
903
				switch_commit_root(root);
904 905 906
				btrfs_unpin_free_ino(root);
				mutex_unlock(&root->fs_commit_mutex);

907 908 909
				btrfs_set_root_node(&root->root_item,
						    root->node);
			}
910 911

			err = btrfs_update_root(trans, fs_info->tree_root,
912 913
						&root->root_key,
						&root->root_item);
J
Josef Bacik 已提交
914
			spin_lock(&fs_info->fs_roots_radix_lock);
915 916
			if (err)
				break;
917 918
		}
	}
J
Josef Bacik 已提交
919
	spin_unlock(&fs_info->fs_roots_radix_lock);
920
	return err;
921 922
}

C
Chris Mason 已提交
923 924 925 926
/*
 * defrag a given btree.  If cacheonly == 1, this won't read from the disk,
 * otherwise every leaf in the btree is read and defragged.
 */
927 928 929 930
int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
{
	struct btrfs_fs_info *info = root->fs_info;
	struct btrfs_trans_handle *trans;
931
	int ret;
932
	unsigned long nr;
933

934
	if (xchg(&root->defrag_running, 1))
935
		return 0;
936

937
	while (1) {
938 939 940 941
		trans = btrfs_start_transaction(root, 0);
		if (IS_ERR(trans))
			return PTR_ERR(trans);

942
		ret = btrfs_defrag_leaves(trans, root, cacheonly);
943

944
		nr = trans->blocks_used;
945
		btrfs_end_transaction(trans, root);
946
		btrfs_btree_balance_dirty(info->tree_root, nr);
947 948
		cond_resched();

949
		if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
950 951 952
			break;
	}
	root->defrag_running = 0;
953
	return ret;
954 955
}

C
Chris Mason 已提交
956 957 958 959
/*
 * new snapshots need to be created at a very specific time in the
 * transaction commit.  This does the actual creation
 */
960
static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
961 962 963 964
				   struct btrfs_fs_info *fs_info,
				   struct btrfs_pending_snapshot *pending)
{
	struct btrfs_key key;
965
	struct btrfs_root_item *new_root_item;
966 967
	struct btrfs_root *tree_root = fs_info->tree_root;
	struct btrfs_root *root = pending->root;
968
	struct btrfs_root *parent_root;
L
Liu Bo 已提交
969
	struct btrfs_block_rsv *rsv;
970
	struct inode *parent_inode;
971 972
	struct btrfs_path *path;
	struct btrfs_dir_item *dir_item;
973
	struct dentry *parent;
974
	struct dentry *dentry;
975
	struct extent_buffer *tmp;
976
	struct extent_buffer *old;
977
	struct timespec cur_time = CURRENT_TIME;
978
	int ret;
979
	u64 to_reserve = 0;
980
	u64 index = 0;
981
	u64 objectid;
L
Li Zefan 已提交
982
	u64 root_flags;
983
	uuid_le new_uuid;
984

985 986 987 988 989 990
	path = btrfs_alloc_path();
	if (!path) {
		ret = pending->error = -ENOMEM;
		goto path_alloc_fail;
	}

991 992
	new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
	if (!new_root_item) {
993
		ret = pending->error = -ENOMEM;
994
		goto root_item_alloc_fail;
995
	}
996

997
	ret = btrfs_find_free_objectid(tree_root, &objectid);
998 999
	if (ret) {
		pending->error = ret;
1000
		goto no_free_objectid;
1001
	}
1002

1003
	btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
1004 1005

	if (to_reserve > 0) {
1006 1007
		ret = btrfs_block_rsv_add_noflush(root, &pending->block_rsv,
						  to_reserve);
1008 1009
		if (ret) {
			pending->error = ret;
1010
			goto no_free_objectid;
1011 1012 1013
		}
	}

A
Arne Jansen 已提交
1014 1015 1016 1017
	ret = btrfs_qgroup_inherit(trans, fs_info, root->root_key.objectid,
				   objectid, pending->inherit);
	if (ret) {
		pending->error = ret;
1018
		goto no_free_objectid;
A
Arne Jansen 已提交
1019 1020
	}

1021
	key.objectid = objectid;
1022 1023
	key.offset = (u64)-1;
	key.type = BTRFS_ROOT_ITEM_KEY;
1024

1025
	rsv = trans->block_rsv;
1026
	trans->block_rsv = &pending->block_rsv;
1027

1028
	dentry = pending->dentry;
1029 1030
	parent = dget_parent(dentry);
	parent_inode = parent->d_inode;
1031
	parent_root = BTRFS_I(parent_inode)->root;
C
Chris Mason 已提交
1032
	record_root_in_trans(trans, parent_root);
1033

1034 1035 1036
	/*
	 * insert the directory item
	 */
1037
	ret = btrfs_set_inode_index(parent_inode, &index);
1038
	BUG_ON(ret); /* -ENOMEM */
1039 1040 1041 1042 1043 1044 1045

	/* check if there is a file/dir which has the same name. */
	dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
					 btrfs_ino(parent_inode),
					 dentry->d_name.name,
					 dentry->d_name.len, 0);
	if (dir_item != NULL && !IS_ERR(dir_item)) {
1046 1047
		pending->error = -EEXIST;
		goto fail;
1048 1049
	} else if (IS_ERR(dir_item)) {
		ret = PTR_ERR(dir_item);
1050
		goto abort_trans;
1051
	}
1052
	btrfs_release_path(path);
1053

1054 1055 1056 1057 1058 1059 1060
	/*
	 * pull in the delayed directory update
	 * and the delayed inode item
	 * otherwise we corrupt the FS during
	 * snapshot
	 */
	ret = btrfs_run_delayed_items(trans, root);
1061 1062
	if (ret)	/* Transaction aborted */
		goto abort_trans;
1063

C
Chris Mason 已提交
1064
	record_root_in_trans(trans, root);
1065 1066
	btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
	memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1067
	btrfs_check_and_init_root_item(new_root_item);
1068

L
Li Zefan 已提交
1069 1070 1071 1072 1073 1074 1075
	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);

1076 1077 1078 1079 1080 1081 1082
	btrfs_set_root_generation_v2(new_root_item,
			trans->transid);
	uuid_le_gen(&new_uuid);
	memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
	memcpy(new_root_item->parent_uuid, root->root_item.uuid,
			BTRFS_UUID_SIZE);
	new_root_item->otime.sec = cpu_to_le64(cur_time.tv_sec);
1083
	new_root_item->otime.nsec = cpu_to_le32(cur_time.tv_nsec);
1084 1085 1086 1087 1088 1089
	btrfs_set_root_otransid(new_root_item, trans->transid);
	memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
	memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
	btrfs_set_root_stransid(new_root_item, 0);
	btrfs_set_root_rtransid(new_root_item, 0);

1090
	old = btrfs_lock_root_node(root);
1091
	ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
1092 1093 1094
	if (ret) {
		btrfs_tree_unlock(old);
		free_extent_buffer(old);
1095
		goto abort_trans;
1096
	}
1097

1098 1099
	btrfs_set_lock_blocking(old);

1100
	ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1101
	/* clean up in any case */
1102 1103
	btrfs_tree_unlock(old);
	free_extent_buffer(old);
1104
	if (ret)
1105
		goto abort_trans;
1106

1107 1108 1109 1110
	/* see comments in should_cow_block() */
	root->force_cow = 1;
	smp_wmb();

1111
	btrfs_set_root_node(new_root_item, tmp);
1112 1113 1114
	/* record when the snapshot was created in key.offset */
	key.offset = trans->transid;
	ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1115 1116
	btrfs_tree_unlock(tmp);
	free_extent_buffer(tmp);
1117
	if (ret)
1118
		goto abort_trans;
1119

1120 1121 1122 1123
	/*
	 * insert root back/forward references
	 */
	ret = btrfs_add_root_ref(trans, tree_root, objectid,
1124
				 parent_root->root_key.objectid,
L
Li Zefan 已提交
1125
				 btrfs_ino(parent_inode), index,
1126
				 dentry->d_name.name, dentry->d_name.len);
1127
	if (ret)
1128
		goto abort_trans;
1129

1130 1131
	key.offset = (u64)-1;
	pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1132 1133
	if (IS_ERR(pending->snap)) {
		ret = PTR_ERR(pending->snap);
1134
		goto abort_trans;
1135
	}
1136

1137 1138 1139
	ret = btrfs_reloc_post_snapshot(trans, pending);
	if (ret)
		goto abort_trans;
1140 1141 1142 1143

	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
	if (ret)
		goto abort_trans;
1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159

	ret = btrfs_insert_dir_item(trans, parent_root,
				    dentry->d_name.name, dentry->d_name.len,
				    parent_inode, &key,
				    BTRFS_FT_DIR, index);
	/* We have check then name at the beginning, so it is impossible. */
	BUG_ON(ret == -EEXIST);
	if (ret)
		goto abort_trans;

	btrfs_i_size_write(parent_inode, parent_inode->i_size +
					 dentry->d_name.len * 2);
	parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
	ret = btrfs_update_inode(trans, parent_root, parent_inode);
	if (ret)
		goto abort_trans;
1160
fail:
1161
	dput(parent);
L
Liu Bo 已提交
1162
	trans->block_rsv = rsv;
1163 1164 1165
no_free_objectid:
	kfree(new_root_item);
root_item_alloc_fail:
1166 1167
	btrfs_free_path(path);
path_alloc_fail:
1168
	btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1);
1169 1170 1171 1172 1173
	return ret;

abort_trans:
	btrfs_abort_transaction(trans, root, ret);
	goto fail;
1174 1175
}

C
Chris Mason 已提交
1176 1177 1178
/*
 * create all the snapshots we've scheduled for creation
 */
1179 1180
static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
					     struct btrfs_fs_info *fs_info)
1181 1182 1183 1184
{
	struct btrfs_pending_snapshot *pending;
	struct list_head *head = &trans->transaction->pending_snapshots;

1185 1186
	list_for_each_entry(pending, head, list)
		create_pending_snapshot(trans, fs_info, pending);
1187 1188 1189
	return 0;
}

1190 1191 1192 1193 1194
static void update_super_roots(struct btrfs_root *root)
{
	struct btrfs_root_item *root_item;
	struct btrfs_super_block *super;

1195
	super = root->fs_info->super_copy;
1196 1197 1198 1199 1200 1201 1202 1203 1204 1205

	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;
1206
	if (btrfs_test_opt(root, SPACE_CACHE))
1207
		super->cache_generation = root_item->generation;
1208 1209
}

1210 1211 1212
int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
{
	int ret = 0;
J
Josef Bacik 已提交
1213
	spin_lock(&info->trans_lock);
1214 1215
	if (info->running_transaction)
		ret = info->running_transaction->in_commit;
J
Josef Bacik 已提交
1216
	spin_unlock(&info->trans_lock);
1217 1218 1219
	return ret;
}

1220 1221 1222
int btrfs_transaction_blocked(struct btrfs_fs_info *info)
{
	int ret = 0;
J
Josef Bacik 已提交
1223
	spin_lock(&info->trans_lock);
1224 1225
	if (info->running_transaction)
		ret = info->running_transaction->blocked;
J
Josef Bacik 已提交
1226
	spin_unlock(&info->trans_lock);
1227 1228 1229
	return ret;
}

S
Sage Weil 已提交
1230 1231 1232 1233 1234 1235 1236
/*
 * 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)
{
L
Li Zefan 已提交
1237
	wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit);
S
Sage Weil 已提交
1238 1239 1240 1241 1242 1243 1244 1245 1246
}

/*
 * 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)
{
L
Li Zefan 已提交
1247 1248
	wait_event(root->fs_info->transaction_wait,
		   trans->commit_done || (trans->in_commit && !trans->blocked));
S
Sage Weil 已提交
1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265
}

/*
 * 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);

1266 1267 1268 1269 1270 1271 1272 1273
	/*
	 * We've got freeze protection passed with the transaction.
	 * Tell lockdep about it.
	 */
	rwsem_acquire_read(
		&ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
		0, 1, _THIS_IP_);

1274 1275
	current->journal_info = ac->newtrans;

S
Sage Weil 已提交
1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287
	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
1288 1289
	if (!ac)
		return -ENOMEM;
S
Sage Weil 已提交
1290 1291 1292

	INIT_DELAYED_WORK(&ac->work, do_async_commit);
	ac->root = root;
1293
	ac->newtrans = btrfs_join_transaction(root);
1294 1295 1296 1297 1298
	if (IS_ERR(ac->newtrans)) {
		int err = PTR_ERR(ac->newtrans);
		kfree(ac);
		return err;
	}
S
Sage Weil 已提交
1299 1300 1301

	/* take transaction reference */
	cur_trans = trans->transaction;
1302
	atomic_inc(&cur_trans->use_count);
S
Sage Weil 已提交
1303 1304

	btrfs_end_transaction(trans, root);
1305 1306 1307 1308 1309 1310 1311 1312

	/*
	 * Tell lockdep we've released the freeze rwsem, since the
	 * async commit thread will be the one to unlock it.
	 */
	rwsem_release(&root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
		      1, _THIS_IP_);

S
Sage Weil 已提交
1313 1314 1315 1316 1317 1318 1319 1320
	schedule_delayed_work(&ac->work, 0);

	/* wait for transaction to start and unblock */
	if (wait_for_unblock)
		wait_current_trans_commit_start_and_unblock(root, cur_trans);
	else
		wait_current_trans_commit_start(root, cur_trans);

1321 1322 1323 1324
	if (current->journal_info == trans)
		current->journal_info = NULL;

	put_transaction(cur_trans);
S
Sage Weil 已提交
1325 1326 1327
	return 0;
}

1328 1329

static void cleanup_transaction(struct btrfs_trans_handle *trans,
1330
				struct btrfs_root *root, int err)
1331 1332 1333 1334 1335
{
	struct btrfs_transaction *cur_trans = trans->transaction;

	WARN_ON(trans->use_count > 1);

1336 1337
	btrfs_abort_transaction(trans, root, err);

1338 1339
	spin_lock(&root->fs_info->trans_lock);
	list_del_init(&cur_trans->list);
1340 1341 1342 1343
	if (cur_trans == root->fs_info->running_transaction) {
		root->fs_info->running_transaction = NULL;
		root->fs_info->trans_no_join = 0;
	}
1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360
	spin_unlock(&root->fs_info->trans_lock);

	btrfs_cleanup_one_transaction(trans->transaction, root);

	put_transaction(cur_trans);
	put_transaction(cur_trans);

	trace_btrfs_transaction_commit(root);

	btrfs_scrub_continue(root);

	if (current->journal_info == trans)
		current->journal_info = NULL;

	kmem_cache_free(btrfs_trans_handle_cachep, trans);
}

S
Sage Weil 已提交
1361 1362 1363 1364 1365 1366 1367
/*
 * 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
1368 1369 1370
int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
			     struct btrfs_root *root)
{
1371
	unsigned long joined = 0;
1372
	struct btrfs_transaction *cur_trans = trans->transaction;
C
Chris Mason 已提交
1373
	struct btrfs_transaction *prev_trans = NULL;
C
Chris Mason 已提交
1374
	DEFINE_WAIT(wait);
1375
	int ret = -EIO;
1376 1377
	int should_grow = 0;
	unsigned long now = get_seconds();
1378
	int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
C
Chris Mason 已提交
1379

1380 1381
	btrfs_run_ordered_operations(root, 0);

1382 1383 1384
	if (cur_trans->aborted)
		goto cleanup_transaction;

1385 1386 1387 1388
	/* 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);
1389 1390
	if (ret)
		goto cleanup_transaction;
1391

1392 1393 1394
	btrfs_trans_release_metadata(trans, root);
	trans->block_rsv = NULL;

1395
	cur_trans = trans->transaction;
1396

1397 1398 1399 1400
	/*
	 * set the flushing flag so procs in this transaction have to
	 * start sending their work down.
	 */
1401
	cur_trans->delayed_refs.flushing = 1;
1402

1403
	ret = btrfs_run_delayed_refs(trans, root, 0);
1404 1405
	if (ret)
		goto cleanup_transaction;
1406

J
Josef Bacik 已提交
1407
	spin_lock(&cur_trans->commit_lock);
1408
	if (cur_trans->in_commit) {
J
Josef Bacik 已提交
1409
		spin_unlock(&cur_trans->commit_lock);
1410
		atomic_inc(&cur_trans->use_count);
1411
		ret = btrfs_end_transaction(trans, root);
C
Chris Mason 已提交
1412

1413
		wait_for_commit(root, cur_trans);
1414

C
Chris Mason 已提交
1415
		put_transaction(cur_trans);
1416

1417
		return ret;
C
Chris Mason 已提交
1418
	}
1419

C
Chris Mason 已提交
1420
	trans->transaction->in_commit = 1;
1421
	trans->transaction->blocked = 1;
J
Josef Bacik 已提交
1422
	spin_unlock(&cur_trans->commit_lock);
S
Sage Weil 已提交
1423 1424
	wake_up(&root->fs_info->transaction_blocked_wait);

J
Josef Bacik 已提交
1425
	spin_lock(&root->fs_info->trans_lock);
C
Chris Mason 已提交
1426 1427 1428 1429
	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) {
1430
			atomic_inc(&prev_trans->use_count);
J
Josef Bacik 已提交
1431
			spin_unlock(&root->fs_info->trans_lock);
C
Chris Mason 已提交
1432 1433 1434

			wait_for_commit(root, prev_trans);

1435
			put_transaction(prev_trans);
J
Josef Bacik 已提交
1436 1437
		} else {
			spin_unlock(&root->fs_info->trans_lock);
C
Chris Mason 已提交
1438
		}
J
Josef Bacik 已提交
1439 1440
	} else {
		spin_unlock(&root->fs_info->trans_lock);
C
Chris Mason 已提交
1441
	}
1442

1443 1444
	if (!btrfs_test_opt(root, SSD) &&
	    (now < cur_trans->start_time || now - cur_trans->start_time < 1))
1445 1446
		should_grow = 1;

1447
	do {
1448
		int snap_pending = 0;
J
Josef Bacik 已提交
1449

1450
		joined = cur_trans->num_joined;
1451 1452 1453
		if (!list_empty(&trans->transaction->pending_snapshots))
			snap_pending = 1;

C
Chris Mason 已提交
1454
		WARN_ON(cur_trans != trans->transaction);
1455

1456
		if (flush_on_commit || snap_pending) {
Y
Yan, Zheng 已提交
1457
			btrfs_start_delalloc_inodes(root, 1);
1458
			btrfs_wait_ordered_extents(root, 0, 1);
1459 1460
		}

1461
		ret = btrfs_run_delayed_items(trans, root);
1462 1463
		if (ret)
			goto cleanup_transaction;
1464

1465 1466 1467 1468 1469 1470 1471
		/*
		 * running the delayed items may have added new refs. account
		 * them now so that they hinder processing of more delayed refs
		 * as little as possible.
		 */
		btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);

1472 1473 1474 1475 1476 1477 1478 1479 1480
		/*
		 * 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);

1481 1482 1483
		prepare_to_wait(&cur_trans->writer_wait, &wait,
				TASK_UNINTERRUPTIBLE);

1484
		if (atomic_read(&cur_trans->num_writers) > 1)
1485 1486 1487
			schedule_timeout(MAX_SCHEDULE_TIMEOUT);
		else if (should_grow)
			schedule_timeout(1);
1488 1489

		finish_wait(&cur_trans->writer_wait, &wait);
1490
	} while (atomic_read(&cur_trans->num_writers) > 1 ||
1491
		 (should_grow && cur_trans->num_joined != joined));
1492

1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503
	/*
	 * Ok now we need to make sure to block out any other joins while we
	 * commit the transaction.  We could have started a join before setting
	 * no_join so make sure to wait for num_writers to == 1 again.
	 */
	spin_lock(&root->fs_info->trans_lock);
	root->fs_info->trans_no_join = 1;
	spin_unlock(&root->fs_info->trans_lock);
	wait_event(cur_trans->writer_wait,
		   atomic_read(&cur_trans->num_writers) == 1);

C
Chris Mason 已提交
1504 1505 1506 1507 1508 1509 1510
	/*
	 * the reloc mutex makes sure that we stop
	 * the balancing code from coming in and moving
	 * extents around in the middle of the commit
	 */
	mutex_lock(&root->fs_info->reloc_mutex);

1511 1512 1513 1514 1515 1516
	/*
	 * We needn't worry about the delayed items because we will
	 * deal with them in create_pending_snapshot(), which is the
	 * core function of the snapshot creation.
	 */
	ret = create_pending_snapshots(trans, root->fs_info);
1517 1518 1519 1520
	if (ret) {
		mutex_unlock(&root->fs_info->reloc_mutex);
		goto cleanup_transaction;
	}
1521

1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532
	/*
	 * We insert the dir indexes of the snapshots and update the inode
	 * of the snapshots' parents after the snapshot creation, so there
	 * are some delayed items which are not dealt with. Now deal with
	 * them.
	 *
	 * We needn't worry that this operation will corrupt the snapshots,
	 * because all the tree which are snapshoted will be forced to COW
	 * the nodes and leaves.
	 */
	ret = btrfs_run_delayed_items(trans, root);
1533 1534 1535 1536
	if (ret) {
		mutex_unlock(&root->fs_info->reloc_mutex);
		goto cleanup_transaction;
	}
1537

1538
	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1539 1540 1541 1542
	if (ret) {
		mutex_unlock(&root->fs_info->reloc_mutex);
		goto cleanup_transaction;
	}
1543

1544 1545 1546 1547 1548 1549
	/*
	 * make sure none of the code above managed to slip in a
	 * delayed item
	 */
	btrfs_assert_delayed_root_empty(root);

C
Chris Mason 已提交
1550
	WARN_ON(cur_trans != trans->transaction);
C
Chris Mason 已提交
1551

A
Arne Jansen 已提交
1552
	btrfs_scrub_pause(root);
1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567
	/* 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);

1568
	ret = commit_fs_roots(trans, root);
1569 1570
	if (ret) {
		mutex_unlock(&root->fs_info->tree_log_mutex);
1571
		mutex_unlock(&root->fs_info->reloc_mutex);
1572 1573
		goto cleanup_transaction;
	}
1574

1575
	/* commit_fs_roots gets rid of all the tree log roots, it is now
1576 1577 1578 1579
	 * safe to free the root of tree log roots
	 */
	btrfs_free_log_root_tree(trans, root->fs_info);

1580
	ret = commit_cowonly_roots(trans, root);
1581 1582
	if (ret) {
		mutex_unlock(&root->fs_info->tree_log_mutex);
1583
		mutex_unlock(&root->fs_info->reloc_mutex);
1584 1585
		goto cleanup_transaction;
	}
1586

1587 1588
	btrfs_prepare_extent_commit(trans, root);

C
Chris Mason 已提交
1589
	cur_trans = root->fs_info->running_transaction;
1590 1591 1592

	btrfs_set_root_node(&root->fs_info->tree_root->root_item,
			    root->fs_info->tree_root->node);
J
Josef Bacik 已提交
1593
	switch_commit_root(root->fs_info->tree_root);
1594 1595 1596

	btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
			    root->fs_info->chunk_root->node);
J
Josef Bacik 已提交
1597
	switch_commit_root(root->fs_info->chunk_root);
1598

1599
	assert_qgroups_uptodate(trans);
1600
	update_super_roots(root);
1601 1602

	if (!root->fs_info->log_root_recovering) {
1603 1604
		btrfs_set_super_log_root(root->fs_info->super_copy, 0);
		btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
1605 1606
	}

1607 1608
	memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
	       sizeof(*root->fs_info->super_copy));
C
Chris Mason 已提交
1609

1610
	trans->transaction->blocked = 0;
J
Josef Bacik 已提交
1611 1612 1613 1614
	spin_lock(&root->fs_info->trans_lock);
	root->fs_info->running_transaction = NULL;
	root->fs_info->trans_no_join = 0;
	spin_unlock(&root->fs_info->trans_lock);
C
Chris Mason 已提交
1615
	mutex_unlock(&root->fs_info->reloc_mutex);
1616

1617
	wake_up(&root->fs_info->transaction_wait);
1618

C
Chris Mason 已提交
1619
	ret = btrfs_write_and_wait_transaction(trans, root);
1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631
	if (ret) {
		btrfs_error(root->fs_info, ret,
			    "Error while writing out transaction.");
		mutex_unlock(&root->fs_info->tree_log_mutex);
		goto cleanup_transaction;
	}

	ret = write_ctree_super(trans, root, 0);
	if (ret) {
		mutex_unlock(&root->fs_info->tree_log_mutex);
		goto cleanup_transaction;
	}
1632

1633 1634 1635 1636 1637 1638
	/*
	 * the super is written, we can safely allow the tree-loggers
	 * to go about their business
	 */
	mutex_unlock(&root->fs_info->tree_log_mutex);

1639
	btrfs_finish_extent_commit(trans, root);
1640

C
Chris Mason 已提交
1641
	cur_trans->commit_done = 1;
1642

1643
	root->fs_info->last_trans_committed = cur_trans->transid;
J
Josef Bacik 已提交
1644

C
Chris Mason 已提交
1645
	wake_up(&cur_trans->commit_wait);
1646

J
Josef Bacik 已提交
1647
	spin_lock(&root->fs_info->trans_lock);
1648
	list_del_init(&cur_trans->list);
J
Josef Bacik 已提交
1649 1650
	spin_unlock(&root->fs_info->trans_lock);

C
Chris Mason 已提交
1651
	put_transaction(cur_trans);
C
Chris Mason 已提交
1652
	put_transaction(cur_trans);
1653

1654 1655
	sb_end_intwrite(root->fs_info->sb);

1656 1657
	trace_btrfs_transaction_commit(root);

A
Arne Jansen 已提交
1658 1659
	btrfs_scrub_continue(root);

J
Josef Bacik 已提交
1660 1661 1662
	if (current->journal_info == trans)
		current->journal_info = NULL;

C
Chris Mason 已提交
1663
	kmem_cache_free(btrfs_trans_handle_cachep, trans);
Y
Yan, Zheng 已提交
1664 1665 1666 1667

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

C
Chris Mason 已提交
1668
	return ret;
1669 1670

cleanup_transaction:
1671 1672
	btrfs_trans_release_metadata(trans, root);
	trans->block_rsv = NULL;
1673 1674 1675 1676
	btrfs_printk(root->fs_info, "Skipping commit of aborted transaction.\n");
//	WARN_ON(1);
	if (current->journal_info == trans)
		current->journal_info = NULL;
1677
	cleanup_transaction(trans, root, ret);
1678 1679

	return ret;
C
Chris Mason 已提交
1680 1681
}

C
Chris Mason 已提交
1682 1683 1684
/*
 * interface function to delete all the snapshots we have scheduled for deletion
 */
1685 1686
int btrfs_clean_old_snapshots(struct btrfs_root *root)
{
1687 1688 1689
	LIST_HEAD(list);
	struct btrfs_fs_info *fs_info = root->fs_info;

J
Josef Bacik 已提交
1690
	spin_lock(&fs_info->trans_lock);
1691
	list_splice_init(&fs_info->dead_roots, &list);
J
Josef Bacik 已提交
1692
	spin_unlock(&fs_info->trans_lock);
1693

1694
	while (!list_empty(&list)) {
1695 1696
		int ret;

1697
		root = list_entry(list.next, struct btrfs_root, root_list);
1698 1699
		list_del(&root->root_list);

1700 1701
		btrfs_kill_all_delayed_nodes(root);

1702 1703
		if (btrfs_header_backref_rev(root->node) <
		    BTRFS_MIXED_BACKREF_REV)
1704
			ret = btrfs_drop_snapshot(root, NULL, 0, 0);
1705
		else
1706 1707
			ret =btrfs_drop_snapshot(root, NULL, 1, 0);
		BUG_ON(ret < 0);
1708 1709 1710
	}
	return 0;
}