transaction.c 45.1 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
	TRANS_JOIN_FREEZE,
276 277
};

278 279
static int may_wait_transaction(struct btrfs_root *root, int type)
{
J
Josef Bacik 已提交
280 281 282 283 284 285 286 287
	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))
288
		return 1;
J
Josef Bacik 已提交
289

290 291 292
	return 0;
}

293
static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
294 295
						    u64 num_items, int type,
						    int noflush)
C
Chris Mason 已提交
296
{
297 298
	struct btrfs_trans_handle *h;
	struct btrfs_transaction *cur_trans;
299
	u64 num_bytes = 0;
C
Chris Mason 已提交
300
	int ret;
301
	u64 qgroup_reserved = 0;
L
liubo 已提交
302 303 304

	if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
		return ERR_PTR(-EROFS);
305 306 307 308 309 310 311 312 313

	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;
	}
314 315 316 317 318 319

	/*
	 * 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) {
320 321 322 323 324 325 326 327
		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);
		}

328
		num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
329 330 331 332 333 334 335 336
		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);
337 338 339
		if (ret)
			return ERR_PTR(ret);
	}
340 341 342 343
again:
	h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
	if (!h)
		return ERR_PTR(-ENOMEM);
C
Chris Mason 已提交
344

345 346 347 348 349
	if (!__sb_start_write(root->fs_info->sb, SB_FREEZE_FS, false)) {
		if (type == TRANS_JOIN_FREEZE)
			return ERR_PTR(-EPERM);
		sb_start_intwrite(root->fs_info->sb);
	}
350

351
	if (may_wait_transaction(root, type))
C
Chris Mason 已提交
352
		wait_current_trans(root);
353

J
Josef Bacik 已提交
354 355 356 357 358 359
	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
360
	if (ret < 0) {
361
		sb_end_intwrite(root->fs_info->sb);
362
		kmem_cache_free(btrfs_trans_handle_cachep, h);
T
Tsutomu Itoh 已提交
363 364
		return ERR_PTR(ret);
	}
365

366 367 368 369
	cur_trans = root->fs_info->running_transaction;

	h->transid = cur_trans->transid;
	h->transaction = cur_trans;
C
Chris Mason 已提交
370
	h->blocks_used = 0;
371
	h->bytes_reserved = 0;
372
	h->root = root;
373
	h->delayed_ref_updates = 0;
374
	h->use_count = 1;
375
	h->adding_csums = 0;
376
	h->block_rsv = NULL;
377
	h->orig_rsv = NULL;
378
	h->aborted = 0;
379
	h->qgroup_reserved = qgroup_reserved;
380 381
	h->delayed_ref_elem.seq = 0;
	INIT_LIST_HEAD(&h->qgroup_ref_list);
382
	INIT_LIST_HEAD(&h->new_bgs);
383

384 385 386 387 388 389
	smp_mb();
	if (cur_trans->blocked && may_wait_transaction(root, type)) {
		btrfs_commit_transaction(h, root);
		goto again;
	}

390
	if (num_bytes) {
J
Josef Bacik 已提交
391
		trace_btrfs_space_reservation(root->fs_info, "transaction",
392
					      h->transid, num_bytes, 1);
393 394
		h->block_rsv = &root->fs_info->trans_block_rsv;
		h->bytes_reserved = num_bytes;
395
	}
J
Josef Bacik 已提交
396

397
got_it:
J
Josef Bacik 已提交
398
	btrfs_record_root_in_trans(h, root);
399 400 401

	if (!current->journal_info && type != TRANS_USERSPACE)
		current->journal_info = h;
C
Chris Mason 已提交
402 403 404
	return h;
}

405
struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
406
						   int num_items)
407
{
408
	return start_transaction(root, num_items, TRANS_START, 0);
409
}
410 411 412 413 414 415 416

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

417
struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
418
{
419
	return start_transaction(root, 0, TRANS_JOIN, 0);
420 421
}

422
struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
423
{
424
	return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0);
425 426
}

427
struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
428
{
429
	return start_transaction(root, 0, TRANS_USERSPACE, 0);
430 431
}

432 433 434 435 436
struct btrfs_trans_handle *btrfs_join_transaction_freeze(struct btrfs_root *root)
{
	return start_transaction(root, 0, TRANS_JOIN_FREEZE, 0);
}

C
Chris Mason 已提交
437
/* wait for a transaction commit to be fully complete */
438
static noinline void wait_for_commit(struct btrfs_root *root,
439 440
				    struct btrfs_transaction *commit)
{
L
Li Zefan 已提交
441
	wait_event(commit->commit_wait, commit->commit_done);
442 443
}

444 445 446 447 448 449 450 451
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
452
			goto out;
453 454

		/* find specified transaction */
J
Josef Bacik 已提交
455
		spin_lock(&root->fs_info->trans_lock);
456 457 458
		list_for_each_entry(t, &root->fs_info->trans_list, list) {
			if (t->transid == transid) {
				cur_trans = t;
J
Josef Bacik 已提交
459
				atomic_inc(&cur_trans->use_count);
460 461 462 463 464
				break;
			}
			if (t->transid > transid)
				break;
		}
J
Josef Bacik 已提交
465
		spin_unlock(&root->fs_info->trans_lock);
466 467
		ret = -EINVAL;
		if (!cur_trans)
J
Josef Bacik 已提交
468
			goto out;  /* bad transid */
469 470
	} else {
		/* find newest transaction that is committing | committed */
J
Josef Bacik 已提交
471
		spin_lock(&root->fs_info->trans_lock);
472 473 474 475
		list_for_each_entry_reverse(t, &root->fs_info->trans_list,
					    list) {
			if (t->in_commit) {
				if (t->commit_done)
476
					break;
477
				cur_trans = t;
J
Josef Bacik 已提交
478
				atomic_inc(&cur_trans->use_count);
479 480 481
				break;
			}
		}
J
Josef Bacik 已提交
482
		spin_unlock(&root->fs_info->trans_lock);
483
		if (!cur_trans)
J
Josef Bacik 已提交
484
			goto out;  /* nothing committing|committed */
485 486 487 488 489 490
	}

	wait_for_commit(root, cur_trans);

	put_transaction(cur_trans);
	ret = 0;
J
Josef Bacik 已提交
491
out:
492 493 494
	return ret;
}

C
Chris Mason 已提交
495 496
void btrfs_throttle(struct btrfs_root *root)
{
J
Josef Bacik 已提交
497
	if (!atomic_read(&root->fs_info->open_ioctl_trans))
498
		wait_current_trans(root);
C
Chris Mason 已提交
499 500
}

501 502 503 504
static int should_end_transaction(struct btrfs_trans_handle *trans,
				  struct btrfs_root *root)
{
	int ret;
505 506

	ret = btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
507 508 509 510 511 512 513 514
	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;
515
	int err;
516

J
Josef Bacik 已提交
517
	smp_mb();
518 519 520 521 522
	if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
		return 1;

	updates = trans->delayed_ref_updates;
	trans->delayed_ref_updates = 0;
523 524 525 526 527
	if (updates) {
		err = btrfs_run_delayed_refs(trans, root, updates);
		if (err) /* Error code will also eval true */
			return err;
	}
528 529 530 531

	return should_end_transaction(trans, root);
}

532
static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
533
			  struct btrfs_root *root, int throttle, int lock)
C
Chris Mason 已提交
534
{
535
	struct btrfs_transaction *cur_trans = trans->transaction;
536
	struct btrfs_fs_info *info = root->fs_info;
537
	int count = 0;
538
	int err = 0;
539

540 541 542 543 544
	if (--trans->use_count) {
		trans->block_rsv = trans->orig_rsv;
		return 0;
	}

545 546 547 548 549
	/*
	 * do the qgroup accounting as early as possible
	 */
	err = btrfs_delayed_refs_qgroup_accounting(trans, info);

550
	btrfs_trans_release_metadata(trans, root);
551
	trans->block_rsv = NULL;
552 553 554 555 556
	/*
	 * the same root has to be passed to start_transaction and
	 * end_transaction. Subvolume quota depends on this.
	 */
	WARN_ON(trans->root != root);
557 558 559 560 561 562

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

563 564 565
	if (!list_empty(&trans->new_bgs))
		btrfs_create_pending_block_groups(trans, root);

566
	while (count < 2) {
567 568 569 570 571 572 573 574 575 576
		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++;
577
	}
578 579
	btrfs_trans_release_metadata(trans, root);
	trans->block_rsv = NULL;
580

581 582 583
	if (!list_empty(&trans->new_bgs))
		btrfs_create_pending_block_groups(trans, root);

J
Josef Bacik 已提交
584 585
	if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
	    should_end_transaction(trans, root)) {
586
		trans->transaction->blocked = 1;
J
Josef Bacik 已提交
587 588
		smp_wmb();
	}
589

590
	if (lock && cur_trans->blocked && !cur_trans->in_commit) {
591 592 593 594 595 596 597
		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++;
598
			return btrfs_commit_transaction(trans, root);
599
		} else {
600
			wake_up_process(info->transaction_kthread);
601
		}
602 603
	}

604 605
	sb_end_intwrite(root->fs_info->sb);

606
	WARN_ON(cur_trans != info->running_transaction);
607 608
	WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
	atomic_dec(&cur_trans->num_writers);
609

610
	smp_mb();
C
Chris Mason 已提交
611 612 613
	if (waitqueue_active(&cur_trans->writer_wait))
		wake_up(&cur_trans->writer_wait);
	put_transaction(cur_trans);
J
Josef Bacik 已提交
614 615 616

	if (current->journal_info == trans)
		current->journal_info = NULL;
617

Y
Yan, Zheng 已提交
618 619 620
	if (throttle)
		btrfs_run_delayed_iputs(root);

621 622
	if (trans->aborted ||
	    root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
623
		err = -EIO;
624
	}
625
	assert_qgroups_uptodate(trans);
626

627 628 629
	memset(trans, 0, sizeof(*trans));
	kmem_cache_free(btrfs_trans_handle_cachep, trans);
	return err;
C
Chris Mason 已提交
630 631
}

632 633 634
int btrfs_end_transaction(struct btrfs_trans_handle *trans,
			  struct btrfs_root *root)
{
635 636 637 638 639 640
	int ret;

	ret = __btrfs_end_transaction(trans, root, 0, 1);
	if (ret)
		return ret;
	return 0;
641 642 643 644 645
}

int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
				   struct btrfs_root *root)
{
646 647 648 649 650 651
	int ret;

	ret = __btrfs_end_transaction(trans, root, 1, 1);
	if (ret)
		return ret;
	return 0;
652 653 654 655 656
}

int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans,
				 struct btrfs_root *root)
{
657 658 659 660 661 662 663 664 665 666 667 668
	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);
669 670
}

C
Chris Mason 已提交
671 672 673
/*
 * 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
674
 * those extents are sent to disk but does not wait on them
C
Chris Mason 已提交
675
 */
676
int btrfs_write_marked_extents(struct btrfs_root *root,
677
			       struct extent_io_tree *dirty_pages, int mark)
C
Chris Mason 已提交
678
{
679
	int err = 0;
680
	int werr = 0;
J
Josef Bacik 已提交
681
	struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
682
	u64 start = 0;
683
	u64 end;
684

J
Josef Bacik 已提交
685 686 687 688 689 690 691 692 693
	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;
694
	}
695 696 697 698 699 700 701 702 703 704 705 706
	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,
707
			      struct extent_io_tree *dirty_pages, int mark)
708 709 710
{
	int err = 0;
	int werr = 0;
J
Josef Bacik 已提交
711
	struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
712 713
	u64 start = 0;
	u64 end;
714

J
Josef Bacik 已提交
715 716 717 718 719 720 721 722
	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;
723
	}
724 725 726
	if (err)
		werr = err;
	return werr;
C
Chris Mason 已提交
727 728
}

729 730 731 732 733 734
/*
 * 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,
735
				struct extent_io_tree *dirty_pages, int mark)
736 737 738 739
{
	int ret;
	int ret2;

740 741
	ret = btrfs_write_marked_extents(root, dirty_pages, mark);
	ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
742 743 744 745 746 747

	if (ret)
		return ret;
	if (ret2)
		return ret2;
	return 0;
748 749
}

750 751 752 753 754 755 756 757 758
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,
759 760
					   &trans->transaction->dirty_pages,
					   EXTENT_DIRTY);
761 762
}

C
Chris Mason 已提交
763 764 765 766 767 768 769 770 771 772
/*
 * 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.
 */
773 774
static int update_cowonly_root(struct btrfs_trans_handle *trans,
			       struct btrfs_root *root)
C
Chris Mason 已提交
775 776
{
	int ret;
777
	u64 old_root_bytenr;
778
	u64 old_root_used;
779
	struct btrfs_root *tree_root = root->fs_info->tree_root;
C
Chris Mason 已提交
780

781
	old_root_used = btrfs_root_used(&root->root_item);
782
	btrfs_write_dirty_block_groups(trans, root);
783

C
Chris Mason 已提交
784
	while (1) {
785
		old_root_bytenr = btrfs_root_bytenr(&root->root_item);
786 787
		if (old_root_bytenr == root->node->start &&
		    old_root_used == btrfs_root_used(&root->root_item))
C
Chris Mason 已提交
788
			break;
789

790
		btrfs_set_root_node(&root->root_item, root->node);
C
Chris Mason 已提交
791
		ret = btrfs_update_root(trans, tree_root,
792 793
					&root->root_key,
					&root->root_item);
794 795
		if (ret)
			return ret;
796

797
		old_root_used = btrfs_root_used(&root->root_item);
798
		ret = btrfs_write_dirty_block_groups(trans, root);
799 800
		if (ret)
			return ret;
801
	}
802 803 804 805

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

806 807 808
	return 0;
}

C
Chris Mason 已提交
809 810
/*
 * update all the cowonly tree roots on disk
811 812 813 814
 *
 * 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
815
 */
816 817
static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
					 struct btrfs_root *root)
818 819 820
{
	struct btrfs_fs_info *fs_info = root->fs_info;
	struct list_head *next;
821
	struct extent_buffer *eb;
822
	int ret;
823

824
	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
825 826
	if (ret)
		return ret;
827

828
	eb = btrfs_lock_root_node(fs_info->tree_root);
829 830
	ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
			      0, &eb);
831 832
	btrfs_tree_unlock(eb);
	free_extent_buffer(eb);
833

834 835 836
	if (ret)
		return ret;

837
	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
838 839
	if (ret)
		return ret;
840

841 842 843
	ret = btrfs_run_dev_stats(trans, root->fs_info);
	BUG_ON(ret);

844 845 846 847 848 849 850
	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
851
	while (!list_empty(&fs_info->dirty_cowonly_roots)) {
852 853 854
		next = fs_info->dirty_cowonly_roots.next;
		list_del_init(next);
		root = list_entry(next, struct btrfs_root, dirty_list);
855

856 857 858
		ret = update_cowonly_root(trans, root);
		if (ret)
			return ret;
C
Chris Mason 已提交
859
	}
860 861 862 863 864

	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
865 866 867
	return 0;
}

C
Chris Mason 已提交
868 869 870 871 872
/*
 * 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
 */
873
int btrfs_add_dead_root(struct btrfs_root *root)
874
{
J
Josef Bacik 已提交
875
	spin_lock(&root->fs_info->trans_lock);
876
	list_add(&root->root_list, &root->fs_info->dead_roots);
J
Josef Bacik 已提交
877
	spin_unlock(&root->fs_info->trans_lock);
878 879 880
	return 0;
}

C
Chris Mason 已提交
881
/*
882
 * update all the cowonly tree roots on disk
C
Chris Mason 已提交
883
 */
884 885
static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
				    struct btrfs_root *root)
886 887
{
	struct btrfs_root *gang[8];
888
	struct btrfs_fs_info *fs_info = root->fs_info;
889 890
	int i;
	int ret;
891 892
	int err = 0;

J
Josef Bacik 已提交
893
	spin_lock(&fs_info->fs_roots_radix_lock);
C
Chris Mason 已提交
894
	while (1) {
895 896
		ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
						 (void **)gang, 0,
897 898 899 900 901 902
						 ARRAY_SIZE(gang),
						 BTRFS_ROOT_TRANS_TAG);
		if (ret == 0)
			break;
		for (i = 0; i < ret; i++) {
			root = gang[i];
903 904 905
			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
906
			spin_unlock(&fs_info->fs_roots_radix_lock);
Y
Yan Zheng 已提交
907

908
			btrfs_free_log(trans, root);
909
			btrfs_update_reloc_root(trans, root);
910
			btrfs_orphan_commit_root(trans, root);
911

912 913
			btrfs_save_ino_cache(root, trans);

914 915 916 917
			/* see comments in should_cow_block() */
			root->force_cow = 0;
			smp_wmb();

918
			if (root->commit_root != root->node) {
919
				mutex_lock(&root->fs_commit_mutex);
J
Josef Bacik 已提交
920
				switch_commit_root(root);
921 922 923
				btrfs_unpin_free_ino(root);
				mutex_unlock(&root->fs_commit_mutex);

924 925 926
				btrfs_set_root_node(&root->root_item,
						    root->node);
			}
927 928

			err = btrfs_update_root(trans, fs_info->tree_root,
929 930
						&root->root_key,
						&root->root_item);
J
Josef Bacik 已提交
931
			spin_lock(&fs_info->fs_roots_radix_lock);
932 933
			if (err)
				break;
934 935
		}
	}
J
Josef Bacik 已提交
936
	spin_unlock(&fs_info->fs_roots_radix_lock);
937
	return err;
938 939
}

C
Chris Mason 已提交
940 941 942 943
/*
 * defrag a given btree.  If cacheonly == 1, this won't read from the disk,
 * otherwise every leaf in the btree is read and defragged.
 */
944 945 946 947
int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
{
	struct btrfs_fs_info *info = root->fs_info;
	struct btrfs_trans_handle *trans;
948
	int ret;
949
	unsigned long nr;
950

951
	if (xchg(&root->defrag_running, 1))
952
		return 0;
953

954
	while (1) {
955 956 957 958
		trans = btrfs_start_transaction(root, 0);
		if (IS_ERR(trans))
			return PTR_ERR(trans);

959
		ret = btrfs_defrag_leaves(trans, root, cacheonly);
960

961
		nr = trans->blocks_used;
962
		btrfs_end_transaction(trans, root);
963
		btrfs_btree_balance_dirty(info->tree_root, nr);
964 965
		cond_resched();

966
		if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
967 968 969
			break;
	}
	root->defrag_running = 0;
970
	return ret;
971 972
}

C
Chris Mason 已提交
973 974 975 976
/*
 * new snapshots need to be created at a very specific time in the
 * transaction commit.  This does the actual creation
 */
977
static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
978 979 980 981
				   struct btrfs_fs_info *fs_info,
				   struct btrfs_pending_snapshot *pending)
{
	struct btrfs_key key;
982
	struct btrfs_root_item *new_root_item;
983 984
	struct btrfs_root *tree_root = fs_info->tree_root;
	struct btrfs_root *root = pending->root;
985
	struct btrfs_root *parent_root;
L
Liu Bo 已提交
986
	struct btrfs_block_rsv *rsv;
987
	struct inode *parent_inode;
988 989
	struct btrfs_path *path;
	struct btrfs_dir_item *dir_item;
990
	struct dentry *parent;
991
	struct dentry *dentry;
992
	struct extent_buffer *tmp;
993
	struct extent_buffer *old;
994
	struct timespec cur_time = CURRENT_TIME;
995
	int ret;
996
	u64 to_reserve = 0;
997
	u64 index = 0;
998
	u64 objectid;
L
Li Zefan 已提交
999
	u64 root_flags;
1000
	uuid_le new_uuid;
1001

1002 1003 1004 1005 1006 1007
	path = btrfs_alloc_path();
	if (!path) {
		ret = pending->error = -ENOMEM;
		goto path_alloc_fail;
	}

1008 1009
	new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
	if (!new_root_item) {
1010
		ret = pending->error = -ENOMEM;
1011
		goto root_item_alloc_fail;
1012
	}
1013

1014
	ret = btrfs_find_free_objectid(tree_root, &objectid);
1015 1016
	if (ret) {
		pending->error = ret;
1017
		goto no_free_objectid;
1018
	}
1019

1020
	btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
1021 1022

	if (to_reserve > 0) {
1023 1024
		ret = btrfs_block_rsv_add_noflush(root, &pending->block_rsv,
						  to_reserve);
1025 1026
		if (ret) {
			pending->error = ret;
1027
			goto no_free_objectid;
1028 1029 1030
		}
	}

A
Arne Jansen 已提交
1031 1032 1033 1034
	ret = btrfs_qgroup_inherit(trans, fs_info, root->root_key.objectid,
				   objectid, pending->inherit);
	if (ret) {
		pending->error = ret;
1035
		goto no_free_objectid;
A
Arne Jansen 已提交
1036 1037
	}

1038
	key.objectid = objectid;
1039 1040
	key.offset = (u64)-1;
	key.type = BTRFS_ROOT_ITEM_KEY;
1041

1042
	rsv = trans->block_rsv;
1043
	trans->block_rsv = &pending->block_rsv;
1044

1045
	dentry = pending->dentry;
1046 1047
	parent = dget_parent(dentry);
	parent_inode = parent->d_inode;
1048
	parent_root = BTRFS_I(parent_inode)->root;
C
Chris Mason 已提交
1049
	record_root_in_trans(trans, parent_root);
1050

1051 1052 1053
	/*
	 * insert the directory item
	 */
1054
	ret = btrfs_set_inode_index(parent_inode, &index);
1055
	BUG_ON(ret); /* -ENOMEM */
1056 1057 1058 1059 1060 1061 1062

	/* 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)) {
1063 1064
		pending->error = -EEXIST;
		goto fail;
1065 1066
	} else if (IS_ERR(dir_item)) {
		ret = PTR_ERR(dir_item);
1067
		goto abort_trans;
1068
	}
1069
	btrfs_release_path(path);
1070

1071 1072 1073 1074 1075 1076 1077
	/*
	 * 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);
1078 1079
	if (ret)	/* Transaction aborted */
		goto abort_trans;
1080

C
Chris Mason 已提交
1081
	record_root_in_trans(trans, root);
1082 1083
	btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
	memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1084
	btrfs_check_and_init_root_item(new_root_item);
1085

L
Li Zefan 已提交
1086 1087 1088 1089 1090 1091 1092
	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);

1093 1094 1095 1096 1097 1098 1099
	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);
1100
	new_root_item->otime.nsec = cpu_to_le32(cur_time.tv_nsec);
1101 1102 1103 1104 1105 1106
	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);

1107
	old = btrfs_lock_root_node(root);
1108
	ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
1109 1110 1111
	if (ret) {
		btrfs_tree_unlock(old);
		free_extent_buffer(old);
1112
		goto abort_trans;
1113
	}
1114

1115 1116
	btrfs_set_lock_blocking(old);

1117
	ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1118
	/* clean up in any case */
1119 1120
	btrfs_tree_unlock(old);
	free_extent_buffer(old);
1121
	if (ret)
1122
		goto abort_trans;
1123

1124 1125 1126 1127
	/* see comments in should_cow_block() */
	root->force_cow = 1;
	smp_wmb();

1128
	btrfs_set_root_node(new_root_item, tmp);
1129 1130 1131
	/* record when the snapshot was created in key.offset */
	key.offset = trans->transid;
	ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1132 1133
	btrfs_tree_unlock(tmp);
	free_extent_buffer(tmp);
1134
	if (ret)
1135
		goto abort_trans;
1136

1137 1138 1139 1140
	/*
	 * insert root back/forward references
	 */
	ret = btrfs_add_root_ref(trans, tree_root, objectid,
1141
				 parent_root->root_key.objectid,
L
Li Zefan 已提交
1142
				 btrfs_ino(parent_inode), index,
1143
				 dentry->d_name.name, dentry->d_name.len);
1144
	if (ret)
1145
		goto abort_trans;
1146

1147 1148
	key.offset = (u64)-1;
	pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1149 1150
	if (IS_ERR(pending->snap)) {
		ret = PTR_ERR(pending->snap);
1151
		goto abort_trans;
1152
	}
1153

1154 1155 1156
	ret = btrfs_reloc_post_snapshot(trans, pending);
	if (ret)
		goto abort_trans;
1157 1158 1159 1160

	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
	if (ret)
		goto abort_trans;
1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176

	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;
1177
fail:
1178
	dput(parent);
L
Liu Bo 已提交
1179
	trans->block_rsv = rsv;
1180 1181 1182
no_free_objectid:
	kfree(new_root_item);
root_item_alloc_fail:
1183 1184
	btrfs_free_path(path);
path_alloc_fail:
1185
	btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1);
1186 1187 1188 1189 1190
	return ret;

abort_trans:
	btrfs_abort_transaction(trans, root, ret);
	goto fail;
1191 1192
}

C
Chris Mason 已提交
1193 1194 1195
/*
 * create all the snapshots we've scheduled for creation
 */
1196 1197
static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
					     struct btrfs_fs_info *fs_info)
1198 1199 1200 1201
{
	struct btrfs_pending_snapshot *pending;
	struct list_head *head = &trans->transaction->pending_snapshots;

1202 1203
	list_for_each_entry(pending, head, list)
		create_pending_snapshot(trans, fs_info, pending);
1204 1205 1206
	return 0;
}

1207 1208 1209 1210 1211
static void update_super_roots(struct btrfs_root *root)
{
	struct btrfs_root_item *root_item;
	struct btrfs_super_block *super;

1212
	super = root->fs_info->super_copy;
1213 1214 1215 1216 1217 1218 1219 1220 1221 1222

	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;
1223
	if (btrfs_test_opt(root, SPACE_CACHE))
1224
		super->cache_generation = root_item->generation;
1225 1226
}

1227 1228 1229
int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
{
	int ret = 0;
J
Josef Bacik 已提交
1230
	spin_lock(&info->trans_lock);
1231 1232
	if (info->running_transaction)
		ret = info->running_transaction->in_commit;
J
Josef Bacik 已提交
1233
	spin_unlock(&info->trans_lock);
1234 1235 1236
	return ret;
}

1237 1238 1239
int btrfs_transaction_blocked(struct btrfs_fs_info *info)
{
	int ret = 0;
J
Josef Bacik 已提交
1240
	spin_lock(&info->trans_lock);
1241 1242
	if (info->running_transaction)
		ret = info->running_transaction->blocked;
J
Josef Bacik 已提交
1243
	spin_unlock(&info->trans_lock);
1244 1245 1246
	return ret;
}

S
Sage Weil 已提交
1247 1248 1249 1250 1251 1252 1253
/*
 * 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
1254
	wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit);
S
Sage Weil 已提交
1255 1256 1257 1258 1259 1260 1261 1262 1263
}

/*
 * 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
1264 1265
	wait_event(root->fs_info->transaction_wait,
		   trans->commit_done || (trans->in_commit && !trans->blocked));
S
Sage Weil 已提交
1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282
}

/*
 * 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);

1283 1284 1285 1286 1287 1288 1289 1290
	/*
	 * 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_);

1291 1292
	current->journal_info = ac->newtrans;

S
Sage Weil 已提交
1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304
	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
1305 1306
	if (!ac)
		return -ENOMEM;
S
Sage Weil 已提交
1307 1308 1309

	INIT_DELAYED_WORK(&ac->work, do_async_commit);
	ac->root = root;
1310
	ac->newtrans = btrfs_join_transaction(root);
1311 1312 1313 1314 1315
	if (IS_ERR(ac->newtrans)) {
		int err = PTR_ERR(ac->newtrans);
		kfree(ac);
		return err;
	}
S
Sage Weil 已提交
1316 1317 1318

	/* take transaction reference */
	cur_trans = trans->transaction;
1319
	atomic_inc(&cur_trans->use_count);
S
Sage Weil 已提交
1320 1321

	btrfs_end_transaction(trans, root);
1322 1323 1324 1325 1326 1327 1328 1329

	/*
	 * 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
1330 1331 1332 1333 1334 1335 1336 1337
	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);

1338 1339 1340 1341
	if (current->journal_info == trans)
		current->journal_info = NULL;

	put_transaction(cur_trans);
S
Sage Weil 已提交
1342 1343 1344
	return 0;
}

1345 1346

static void cleanup_transaction(struct btrfs_trans_handle *trans,
1347
				struct btrfs_root *root, int err)
1348 1349 1350 1351 1352
{
	struct btrfs_transaction *cur_trans = trans->transaction;

	WARN_ON(trans->use_count > 1);

1353 1354
	btrfs_abort_transaction(trans, root, err);

1355 1356
	spin_lock(&root->fs_info->trans_lock);
	list_del_init(&cur_trans->list);
1357 1358 1359 1360
	if (cur_trans == root->fs_info->running_transaction) {
		root->fs_info->running_transaction = NULL;
		root->fs_info->trans_no_join = 0;
	}
1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377
	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
1378 1379 1380 1381 1382 1383 1384
/*
 * 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
1385 1386 1387
int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
			     struct btrfs_root *root)
{
1388
	unsigned long joined = 0;
1389
	struct btrfs_transaction *cur_trans = trans->transaction;
C
Chris Mason 已提交
1390
	struct btrfs_transaction *prev_trans = NULL;
C
Chris Mason 已提交
1391
	DEFINE_WAIT(wait);
1392
	int ret = -EIO;
1393 1394
	int should_grow = 0;
	unsigned long now = get_seconds();
1395
	int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
C
Chris Mason 已提交
1396

1397 1398
	btrfs_run_ordered_operations(root, 0);

1399 1400 1401
	if (cur_trans->aborted)
		goto cleanup_transaction;

1402 1403 1404 1405
	/* 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);
1406 1407
	if (ret)
		goto cleanup_transaction;
1408

1409 1410 1411
	btrfs_trans_release_metadata(trans, root);
	trans->block_rsv = NULL;

1412
	cur_trans = trans->transaction;
1413

1414 1415 1416 1417
	/*
	 * set the flushing flag so procs in this transaction have to
	 * start sending their work down.
	 */
1418
	cur_trans->delayed_refs.flushing = 1;
1419

1420 1421 1422
	if (!list_empty(&trans->new_bgs))
		btrfs_create_pending_block_groups(trans, root);

1423
	ret = btrfs_run_delayed_refs(trans, root, 0);
1424 1425
	if (ret)
		goto cleanup_transaction;
1426

J
Josef Bacik 已提交
1427
	spin_lock(&cur_trans->commit_lock);
1428
	if (cur_trans->in_commit) {
J
Josef Bacik 已提交
1429
		spin_unlock(&cur_trans->commit_lock);
1430
		atomic_inc(&cur_trans->use_count);
1431
		ret = btrfs_end_transaction(trans, root);
C
Chris Mason 已提交
1432

1433
		wait_for_commit(root, cur_trans);
1434

C
Chris Mason 已提交
1435
		put_transaction(cur_trans);
1436

1437
		return ret;
C
Chris Mason 已提交
1438
	}
1439

C
Chris Mason 已提交
1440
	trans->transaction->in_commit = 1;
1441
	trans->transaction->blocked = 1;
J
Josef Bacik 已提交
1442
	spin_unlock(&cur_trans->commit_lock);
S
Sage Weil 已提交
1443 1444
	wake_up(&root->fs_info->transaction_blocked_wait);

J
Josef Bacik 已提交
1445
	spin_lock(&root->fs_info->trans_lock);
C
Chris Mason 已提交
1446 1447 1448 1449
	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) {
1450
			atomic_inc(&prev_trans->use_count);
J
Josef Bacik 已提交
1451
			spin_unlock(&root->fs_info->trans_lock);
C
Chris Mason 已提交
1452 1453 1454

			wait_for_commit(root, prev_trans);

1455
			put_transaction(prev_trans);
J
Josef Bacik 已提交
1456 1457
		} else {
			spin_unlock(&root->fs_info->trans_lock);
C
Chris Mason 已提交
1458
		}
J
Josef Bacik 已提交
1459 1460
	} else {
		spin_unlock(&root->fs_info->trans_lock);
C
Chris Mason 已提交
1461
	}
1462

1463 1464
	if (!btrfs_test_opt(root, SSD) &&
	    (now < cur_trans->start_time || now - cur_trans->start_time < 1))
1465 1466
		should_grow = 1;

1467
	do {
1468
		int snap_pending = 0;
J
Josef Bacik 已提交
1469

1470
		joined = cur_trans->num_joined;
1471 1472 1473
		if (!list_empty(&trans->transaction->pending_snapshots))
			snap_pending = 1;

C
Chris Mason 已提交
1474
		WARN_ON(cur_trans != trans->transaction);
1475

1476
		if (flush_on_commit || snap_pending) {
Y
Yan, Zheng 已提交
1477
			btrfs_start_delalloc_inodes(root, 1);
1478
			btrfs_wait_ordered_extents(root, 0, 1);
1479 1480
		}

1481
		ret = btrfs_run_delayed_items(trans, root);
1482 1483
		if (ret)
			goto cleanup_transaction;
1484

1485 1486 1487 1488 1489 1490 1491
		/*
		 * 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);

1492 1493 1494 1495 1496 1497 1498 1499 1500
		/*
		 * 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);

1501 1502 1503
		prepare_to_wait(&cur_trans->writer_wait, &wait,
				TASK_UNINTERRUPTIBLE);

1504
		if (atomic_read(&cur_trans->num_writers) > 1)
1505 1506 1507
			schedule_timeout(MAX_SCHEDULE_TIMEOUT);
		else if (should_grow)
			schedule_timeout(1);
1508 1509

		finish_wait(&cur_trans->writer_wait, &wait);
1510
	} while (atomic_read(&cur_trans->num_writers) > 1 ||
1511
		 (should_grow && cur_trans->num_joined != joined));
1512

1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523
	/*
	 * 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
1524 1525 1526 1527 1528 1529 1530
	/*
	 * 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);

1531 1532 1533 1534 1535 1536
	/*
	 * 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);
1537 1538 1539 1540
	if (ret) {
		mutex_unlock(&root->fs_info->reloc_mutex);
		goto cleanup_transaction;
	}
1541

1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552
	/*
	 * 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);
1553 1554 1555 1556
	if (ret) {
		mutex_unlock(&root->fs_info->reloc_mutex);
		goto cleanup_transaction;
	}
1557

1558
	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1559 1560 1561 1562
	if (ret) {
		mutex_unlock(&root->fs_info->reloc_mutex);
		goto cleanup_transaction;
	}
1563

1564 1565 1566 1567 1568 1569
	/*
	 * 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
1570
	WARN_ON(cur_trans != trans->transaction);
C
Chris Mason 已提交
1571

A
Arne Jansen 已提交
1572
	btrfs_scrub_pause(root);
1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587
	/* 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);

1588
	ret = commit_fs_roots(trans, root);
1589 1590
	if (ret) {
		mutex_unlock(&root->fs_info->tree_log_mutex);
1591
		mutex_unlock(&root->fs_info->reloc_mutex);
1592 1593
		goto cleanup_transaction;
	}
1594

1595
	/* commit_fs_roots gets rid of all the tree log roots, it is now
1596 1597 1598 1599
	 * safe to free the root of tree log roots
	 */
	btrfs_free_log_root_tree(trans, root->fs_info);

1600
	ret = commit_cowonly_roots(trans, root);
1601 1602
	if (ret) {
		mutex_unlock(&root->fs_info->tree_log_mutex);
1603
		mutex_unlock(&root->fs_info->reloc_mutex);
1604 1605
		goto cleanup_transaction;
	}
1606

1607 1608
	btrfs_prepare_extent_commit(trans, root);

C
Chris Mason 已提交
1609
	cur_trans = root->fs_info->running_transaction;
1610 1611 1612

	btrfs_set_root_node(&root->fs_info->tree_root->root_item,
			    root->fs_info->tree_root->node);
J
Josef Bacik 已提交
1613
	switch_commit_root(root->fs_info->tree_root);
1614 1615 1616

	btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
			    root->fs_info->chunk_root->node);
J
Josef Bacik 已提交
1617
	switch_commit_root(root->fs_info->chunk_root);
1618

1619
	assert_qgroups_uptodate(trans);
1620
	update_super_roots(root);
1621 1622

	if (!root->fs_info->log_root_recovering) {
1623 1624
		btrfs_set_super_log_root(root->fs_info->super_copy, 0);
		btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
1625 1626
	}

1627 1628
	memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
	       sizeof(*root->fs_info->super_copy));
C
Chris Mason 已提交
1629

1630
	trans->transaction->blocked = 0;
J
Josef Bacik 已提交
1631 1632 1633 1634
	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
1635
	mutex_unlock(&root->fs_info->reloc_mutex);
1636

1637
	wake_up(&root->fs_info->transaction_wait);
1638

C
Chris Mason 已提交
1639
	ret = btrfs_write_and_wait_transaction(trans, root);
1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651
	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;
	}
1652

1653 1654 1655 1656 1657 1658
	/*
	 * the super is written, we can safely allow the tree-loggers
	 * to go about their business
	 */
	mutex_unlock(&root->fs_info->tree_log_mutex);

1659
	btrfs_finish_extent_commit(trans, root);
1660

C
Chris Mason 已提交
1661
	cur_trans->commit_done = 1;
1662

1663
	root->fs_info->last_trans_committed = cur_trans->transid;
J
Josef Bacik 已提交
1664

C
Chris Mason 已提交
1665
	wake_up(&cur_trans->commit_wait);
1666

J
Josef Bacik 已提交
1667
	spin_lock(&root->fs_info->trans_lock);
1668
	list_del_init(&cur_trans->list);
J
Josef Bacik 已提交
1669 1670
	spin_unlock(&root->fs_info->trans_lock);

C
Chris Mason 已提交
1671
	put_transaction(cur_trans);
C
Chris Mason 已提交
1672
	put_transaction(cur_trans);
1673

1674 1675
	sb_end_intwrite(root->fs_info->sb);

1676 1677
	trace_btrfs_transaction_commit(root);

A
Arne Jansen 已提交
1678 1679
	btrfs_scrub_continue(root);

J
Josef Bacik 已提交
1680 1681 1682
	if (current->journal_info == trans)
		current->journal_info = NULL;

C
Chris Mason 已提交
1683
	kmem_cache_free(btrfs_trans_handle_cachep, trans);
Y
Yan, Zheng 已提交
1684 1685 1686 1687

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

C
Chris Mason 已提交
1688
	return ret;
1689 1690

cleanup_transaction:
1691 1692
	btrfs_trans_release_metadata(trans, root);
	trans->block_rsv = NULL;
1693 1694 1695 1696
	btrfs_printk(root->fs_info, "Skipping commit of aborted transaction.\n");
//	WARN_ON(1);
	if (current->journal_info == trans)
		current->journal_info = NULL;
1697
	cleanup_transaction(trans, root, ret);
1698 1699

	return ret;
C
Chris Mason 已提交
1700 1701
}

C
Chris Mason 已提交
1702 1703 1704
/*
 * interface function to delete all the snapshots we have scheduled for deletion
 */
1705 1706
int btrfs_clean_old_snapshots(struct btrfs_root *root)
{
1707 1708 1709
	LIST_HEAD(list);
	struct btrfs_fs_info *fs_info = root->fs_info;

J
Josef Bacik 已提交
1710
	spin_lock(&fs_info->trans_lock);
1711
	list_splice_init(&fs_info->dead_roots, &list);
J
Josef Bacik 已提交
1712
	spin_unlock(&fs_info->trans_lock);
1713

1714
	while (!list_empty(&list)) {
1715 1716
		int ret;

1717
		root = list_entry(list.next, struct btrfs_root, root_list);
1718 1719
		list_del(&root->root_list);

1720 1721
		btrfs_kill_all_delayed_nodes(root);

1722 1723
		if (btrfs_header_backref_rev(root->node) <
		    BTRFS_MIXED_BACKREF_REV)
1724
			ret = btrfs_drop_snapshot(root, NULL, 0, 0);
1725
		else
1726 1727
			ret =btrfs_drop_snapshot(root, NULL, 1, 0);
		BUG_ON(ret < 0);
1728 1729 1730
	}
	return 0;
}