transaction.c 47.0 KB
Newer Older
C
Chris Mason 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
/*
 * Copyright (C) 2007 Oracle.  All rights reserved.
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public
 * License v2 as published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * You should have received a copy of the GNU General Public
 * License along with this program; if not, write to the
 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
 * Boston, MA 021110-1307, USA.
 */

C
Chris Mason 已提交
19
#include <linux/fs.h>
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"
33
#include "dev-replace.h"
C
Chris Mason 已提交
34

35 36
#define BTRFS_ROOT_TRANS_TAG 0

37
void put_transaction(struct btrfs_transaction *transaction)
C
Chris Mason 已提交
38
{
39 40
	WARN_ON(atomic_read(&transaction->use_count) == 0);
	if (atomic_dec_and_test(&transaction->use_count)) {
J
Josef Bacik 已提交
41
		BUG_ON(!list_empty(&transaction->list));
42
		WARN_ON(transaction->delayed_refs.root.rb_node);
C
Chris Mason 已提交
43 44
		memset(transaction, 0, sizeof(*transaction));
		kmem_cache_free(btrfs_transaction_cachep, transaction);
C
Chris Mason 已提交
45
	}
C
Chris Mason 已提交
46 47
}

J
Josef Bacik 已提交
48 49 50 51 52 53
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
54 55 56
/*
 * either allocate a new transaction or hop into the existing one
 */
57
static noinline int join_transaction(struct btrfs_root *root, int type)
C
Chris Mason 已提交
58 59
{
	struct btrfs_transaction *cur_trans;
60
	struct btrfs_fs_info *fs_info = root->fs_info;
J
Josef Bacik 已提交
61

62
	spin_lock(&fs_info->trans_lock);
63
loop:
64
	/* The file system has been taken offline. No new transactions. */
65 66
	if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
		spin_unlock(&fs_info->trans_lock);
67 68 69
		return -EROFS;
	}

70
	if (fs_info->trans_no_join) {
71 72 73 74 75 76 77
		/* 
		 * If we are JOIN_NOLOCK we're already committing a current
		 * transaction, we just need a handle to deal with something
		 * when committing the transaction, such as inode cache and
		 * space cache. It is a special case.
		 */
		if (type != TRANS_JOIN_NOLOCK) {
78
			spin_unlock(&fs_info->trans_lock);
J
Josef Bacik 已提交
79 80 81 82
			return -EBUSY;
		}
	}

83
	cur_trans = fs_info->running_transaction;
J
Josef Bacik 已提交
84
	if (cur_trans) {
85
		if (cur_trans->aborted) {
86
			spin_unlock(&fs_info->trans_lock);
87
			return cur_trans->aborted;
88
		}
J
Josef Bacik 已提交
89
		atomic_inc(&cur_trans->use_count);
90
		atomic_inc(&cur_trans->num_writers);
91
		cur_trans->num_joined++;
92
		spin_unlock(&fs_info->trans_lock);
J
Josef Bacik 已提交
93
		return 0;
C
Chris Mason 已提交
94
	}
95
	spin_unlock(&fs_info->trans_lock);
J
Josef Bacik 已提交
96

97 98 99 100 101 102 103
	/*
	 * If we are ATTACH, we just want to catch the current transaction,
	 * and commit it. If there is no transaction, just return ENOENT.
	 */
	if (type == TRANS_ATTACH)
		return -ENOENT;

J
Josef Bacik 已提交
104 105 106
	cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS);
	if (!cur_trans)
		return -ENOMEM;
107

108 109
	spin_lock(&fs_info->trans_lock);
	if (fs_info->running_transaction) {
110 111 112 113
		/*
		 * someone started a transaction after we unlocked.  Make sure
		 * to redo the trans_no_join checks above
		 */
J
Josef Bacik 已提交
114
		kmem_cache_free(btrfs_transaction_cachep, cur_trans);
115
		cur_trans = fs_info->running_transaction;
116
		goto loop;
117 118
	} else if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
		spin_unlock(&fs_info->trans_lock);
119 120
		kmem_cache_free(btrfs_transaction_cachep, cur_trans);
		return -EROFS;
C
Chris Mason 已提交
121
	}
122

J
Josef Bacik 已提交
123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142
	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;
143 144 145 146 147 148

	/*
	 * although the tree mod log is per file system and not per transaction,
	 * the log must never go across transaction boundaries.
	 */
	smp_mb();
J
Julia Lawall 已提交
149 150
	if (!list_empty(&fs_info->tree_mod_seq_list))
		WARN(1, KERN_ERR "btrfs: tree_mod_seq_list not empty when "
151
			"creating a fresh transaction\n");
J
Julia Lawall 已提交
152 153
	if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log))
		WARN(1, KERN_ERR "btrfs: tree_mod_log rb tree not empty when "
154 155 156
			"creating a fresh transaction\n");
	atomic_set(&fs_info->tree_mod_seq, 0);

J
Josef Bacik 已提交
157 158 159 160
	spin_lock_init(&cur_trans->commit_lock);
	spin_lock_init(&cur_trans->delayed_refs.lock);

	INIT_LIST_HEAD(&cur_trans->pending_snapshots);
161
	list_add_tail(&cur_trans->list, &fs_info->trans_list);
J
Josef Bacik 已提交
162
	extent_io_tree_init(&cur_trans->dirty_pages,
163 164 165 166
			     fs_info->btree_inode->i_mapping);
	fs_info->generation++;
	cur_trans->transid = fs_info->generation;
	fs_info->running_transaction = cur_trans;
167
	cur_trans->aborted = 0;
168
	spin_unlock(&fs_info->trans_lock);
169

C
Chris Mason 已提交
170 171 172
	return 0;
}

C
Chris Mason 已提交
173
/*
C
Chris Mason 已提交
174 175 176 177
 * 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
178
 */
C
Chris Mason 已提交
179
static int record_root_in_trans(struct btrfs_trans_handle *trans,
J
Josef Bacik 已提交
180
			       struct btrfs_root *root)
181
{
182
	if (root->ref_cows && root->last_trans < trans->transid) {
183
		WARN_ON(root == root->fs_info->extent_root);
184 185
		WARN_ON(root->commit_root != root->node);

C
Chris Mason 已提交
186 187 188 189 190 191 192 193 194 195 196 197
		/*
		 * 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
198 199 200 201 202
		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;
		}
203 204 205
		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
206
		spin_unlock(&root->fs_info->fs_roots_radix_lock);
C
Chris Mason 已提交
207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227
		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
		 */
228
		btrfs_init_reloc_root(trans, root);
C
Chris Mason 已提交
229 230
		smp_wmb();
		root->in_trans_setup = 0;
231 232 233
	}
	return 0;
}
234

C
Chris Mason 已提交
235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257

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
258 259 260 261
/* 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
262
static void wait_current_trans(struct btrfs_root *root)
C
Chris Mason 已提交
263
{
264
	struct btrfs_transaction *cur_trans;
C
Chris Mason 已提交
265

J
Josef Bacik 已提交
266
	spin_lock(&root->fs_info->trans_lock);
267
	cur_trans = root->fs_info->running_transaction;
C
Chris Mason 已提交
268
	if (cur_trans && cur_trans->blocked) {
269
		atomic_inc(&cur_trans->use_count);
J
Josef Bacik 已提交
270
		spin_unlock(&root->fs_info->trans_lock);
L
Li Zefan 已提交
271 272 273

		wait_event(root->fs_info->transaction_wait,
			   !cur_trans->blocked);
274
		put_transaction(cur_trans);
J
Josef Bacik 已提交
275 276
	} else {
		spin_unlock(&root->fs_info->trans_lock);
277
	}
C
Chris Mason 已提交
278 279
}

280 281
static int may_wait_transaction(struct btrfs_root *root, int type)
{
J
Josef Bacik 已提交
282 283 284 285 286 287 288 289
	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))
290
		return 1;
J
Josef Bacik 已提交
291

292 293 294
	return 0;
}

M
Miao Xie 已提交
295 296 297
static struct btrfs_trans_handle *
start_transaction(struct btrfs_root *root, u64 num_items, int type,
		  enum btrfs_reserve_flush_enum flush)
C
Chris Mason 已提交
298
{
299 300
	struct btrfs_trans_handle *h;
	struct btrfs_transaction *cur_trans;
301
	u64 num_bytes = 0;
C
Chris Mason 已提交
302
	int ret;
303
	u64 qgroup_reserved = 0;
L
liubo 已提交
304 305 306

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

	if (current->journal_info) {
		WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK);
		h = current->journal_info;
		h->use_count++;
312
		WARN_ON(h->use_count > 2);
313 314 315 316
		h->orig_rsv = h->block_rsv;
		h->block_rsv = NULL;
		goto got_it;
	}
317 318 319 320 321 322

	/*
	 * 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) {
323 324 325 326 327 328 329 330
		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);
		}

331
		num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
M
Miao Xie 已提交
332 333 334
		ret = btrfs_block_rsv_add(root,
					  &root->fs_info->trans_block_rsv,
					  num_bytes, flush);
335 336 337
		if (ret)
			return ERR_PTR(ret);
	}
338 339 340 341
again:
	h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
	if (!h)
		return ERR_PTR(-ENOMEM);
C
Chris Mason 已提交
342

343 344 345 346 347 348
	/*
	 * If we are JOIN_NOLOCK we're already committing a transaction and
	 * waiting on this guy, so we don't need to do the sb_start_intwrite
	 * because we're already holding a ref.  We need this because we could
	 * have raced in and did an fsync() on a file which can kick a commit
	 * and then we deadlock with somebody doing a freeze.
349 350 351
	 *
	 * If we are ATTACH, it means we just want to catch the current
	 * transaction and commit it, so we needn't do sb_start_intwrite(). 
352
	 */
353
	if (type < TRANS_JOIN_NOLOCK)
354
		sb_start_intwrite(root->fs_info->sb);
355

356
	if (may_wait_transaction(root, type))
C
Chris Mason 已提交
357
		wait_current_trans(root);
358

J
Josef Bacik 已提交
359
	do {
360
		ret = join_transaction(root, type);
J
Josef Bacik 已提交
361 362 363 364
		if (ret == -EBUSY)
			wait_current_trans(root);
	} while (ret == -EBUSY);

T
Tsutomu Itoh 已提交
365
	if (ret < 0) {
366 367 368 369 370
		/* We must get the transaction if we are JOIN_NOLOCK. */
		BUG_ON(type == TRANS_JOIN_NOLOCK);

		if (type < TRANS_JOIN_NOLOCK)
			sb_end_intwrite(root->fs_info->sb);
371
		kmem_cache_free(btrfs_trans_handle_cachep, h);
T
Tsutomu Itoh 已提交
372 373
		return ERR_PTR(ret);
	}
374

375 376 377 378
	cur_trans = root->fs_info->running_transaction;

	h->transid = cur_trans->transid;
	h->transaction = cur_trans;
C
Chris Mason 已提交
379
	h->blocks_used = 0;
380
	h->bytes_reserved = 0;
381
	h->root = root;
382
	h->delayed_ref_updates = 0;
383
	h->use_count = 1;
384
	h->adding_csums = 0;
385
	h->block_rsv = NULL;
386
	h->orig_rsv = NULL;
387
	h->aborted = 0;
388
	h->qgroup_reserved = qgroup_reserved;
389
	h->delayed_ref_elem.seq = 0;
390
	h->type = type;
391
	INIT_LIST_HEAD(&h->qgroup_ref_list);
392
	INIT_LIST_HEAD(&h->new_bgs);
393

394 395 396 397 398 399
	smp_mb();
	if (cur_trans->blocked && may_wait_transaction(root, type)) {
		btrfs_commit_transaction(h, root);
		goto again;
	}

400
	if (num_bytes) {
J
Josef Bacik 已提交
401
		trace_btrfs_space_reservation(root->fs_info, "transaction",
402
					      h->transid, num_bytes, 1);
403 404
		h->block_rsv = &root->fs_info->trans_block_rsv;
		h->bytes_reserved = num_bytes;
405
	}
J
Josef Bacik 已提交
406

407
got_it:
J
Josef Bacik 已提交
408
	btrfs_record_root_in_trans(h, root);
409 410 411

	if (!current->journal_info && type != TRANS_USERSPACE)
		current->journal_info = h;
C
Chris Mason 已提交
412 413 414
	return h;
}

415
struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
416
						   int num_items)
417
{
M
Miao Xie 已提交
418 419
	return start_transaction(root, num_items, TRANS_START,
				 BTRFS_RESERVE_FLUSH_ALL);
420
}
421

M
Miao Xie 已提交
422
struct btrfs_trans_handle *btrfs_start_transaction_lflush(
423 424
					struct btrfs_root *root, int num_items)
{
M
Miao Xie 已提交
425 426
	return start_transaction(root, num_items, TRANS_START,
				 BTRFS_RESERVE_FLUSH_LIMIT);
427 428
}

429
struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
430
{
431
	return start_transaction(root, 0, TRANS_JOIN, 0);
432 433
}

434
struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
435
{
436
	return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0);
437 438
}

439
struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
440
{
441
	return start_transaction(root, 0, TRANS_USERSPACE, 0);
442 443
}

444
struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
445
{
446
	return start_transaction(root, 0, TRANS_ATTACH, 0);
447 448
}

C
Chris Mason 已提交
449
/* wait for a transaction commit to be fully complete */
450
static noinline void wait_for_commit(struct btrfs_root *root,
451 452
				    struct btrfs_transaction *commit)
{
L
Li Zefan 已提交
453
	wait_event(commit->commit_wait, commit->commit_done);
454 455
}

456 457 458 459 460 461 462 463
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
464
			goto out;
465 466

		/* find specified transaction */
J
Josef Bacik 已提交
467
		spin_lock(&root->fs_info->trans_lock);
468 469 470
		list_for_each_entry(t, &root->fs_info->trans_list, list) {
			if (t->transid == transid) {
				cur_trans = t;
J
Josef Bacik 已提交
471
				atomic_inc(&cur_trans->use_count);
472 473 474 475 476
				break;
			}
			if (t->transid > transid)
				break;
		}
J
Josef Bacik 已提交
477
		spin_unlock(&root->fs_info->trans_lock);
478 479
		ret = -EINVAL;
		if (!cur_trans)
J
Josef Bacik 已提交
480
			goto out;  /* bad transid */
481 482
	} else {
		/* find newest transaction that is committing | committed */
J
Josef Bacik 已提交
483
		spin_lock(&root->fs_info->trans_lock);
484 485 486 487
		list_for_each_entry_reverse(t, &root->fs_info->trans_list,
					    list) {
			if (t->in_commit) {
				if (t->commit_done)
488
					break;
489
				cur_trans = t;
J
Josef Bacik 已提交
490
				atomic_inc(&cur_trans->use_count);
491 492 493
				break;
			}
		}
J
Josef Bacik 已提交
494
		spin_unlock(&root->fs_info->trans_lock);
495
		if (!cur_trans)
J
Josef Bacik 已提交
496
			goto out;  /* nothing committing|committed */
497 498 499 500 501 502
	}

	wait_for_commit(root, cur_trans);

	put_transaction(cur_trans);
	ret = 0;
J
Josef Bacik 已提交
503
out:
504 505 506
	return ret;
}

C
Chris Mason 已提交
507 508
void btrfs_throttle(struct btrfs_root *root)
{
J
Josef Bacik 已提交
509
	if (!atomic_read(&root->fs_info->open_ioctl_trans))
510
		wait_current_trans(root);
C
Chris Mason 已提交
511 512
}

513 514 515 516
static int should_end_transaction(struct btrfs_trans_handle *trans,
				  struct btrfs_root *root)
{
	int ret;
517 518

	ret = btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
519 520 521 522 523 524 525 526
	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;
527
	int err;
528

J
Josef Bacik 已提交
529
	smp_mb();
530 531 532 533 534
	if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
		return 1;

	updates = trans->delayed_ref_updates;
	trans->delayed_ref_updates = 0;
535 536 537 538 539
	if (updates) {
		err = btrfs_run_delayed_refs(trans, root, updates);
		if (err) /* Error code will also eval true */
			return err;
	}
540 541 542 543

	return should_end_transaction(trans, root);
}

544
static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
545
			  struct btrfs_root *root, int throttle)
C
Chris Mason 已提交
546
{
547
	struct btrfs_transaction *cur_trans = trans->transaction;
548
	struct btrfs_fs_info *info = root->fs_info;
549
	int count = 0;
550
	int lock = (trans->type != TRANS_JOIN_NOLOCK);
551
	int err = 0;
552

553 554 555 556 557
	if (--trans->use_count) {
		trans->block_rsv = trans->orig_rsv;
		return 0;
	}

558 559 560 561 562
	/*
	 * do the qgroup accounting as early as possible
	 */
	err = btrfs_delayed_refs_qgroup_accounting(trans, info);

563
	btrfs_trans_release_metadata(trans, root);
564
	trans->block_rsv = NULL;
565 566 567 568 569
	/*
	 * the same root has to be passed to start_transaction and
	 * end_transaction. Subvolume quota depends on this.
	 */
	WARN_ON(trans->root != root);
570 571 572 573 574 575

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

576 577 578
	if (!list_empty(&trans->new_bgs))
		btrfs_create_pending_block_groups(trans, root);

579
	while (count < 2) {
580 581 582 583 584 585 586 587 588 589
		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++;
590
	}
591 592
	btrfs_trans_release_metadata(trans, root);
	trans->block_rsv = NULL;
593

594 595 596
	if (!list_empty(&trans->new_bgs))
		btrfs_create_pending_block_groups(trans, root);

J
Josef Bacik 已提交
597 598
	if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
	    should_end_transaction(trans, root)) {
599
		trans->transaction->blocked = 1;
J
Josef Bacik 已提交
600 601
		smp_wmb();
	}
602

603
	if (lock && cur_trans->blocked && !cur_trans->in_commit) {
604 605 606 607 608 609 610
		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++;
611
			return btrfs_commit_transaction(trans, root);
612
		} else {
613
			wake_up_process(info->transaction_kthread);
614
		}
615 616
	}

617
	if (trans->type < TRANS_JOIN_NOLOCK)
618
		sb_end_intwrite(root->fs_info->sb);
619

620
	WARN_ON(cur_trans != info->running_transaction);
621 622
	WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
	atomic_dec(&cur_trans->num_writers);
623

624
	smp_mb();
C
Chris Mason 已提交
625 626 627
	if (waitqueue_active(&cur_trans->writer_wait))
		wake_up(&cur_trans->writer_wait);
	put_transaction(cur_trans);
J
Josef Bacik 已提交
628 629 630

	if (current->journal_info == trans)
		current->journal_info = NULL;
631

Y
Yan, Zheng 已提交
632 633 634
	if (throttle)
		btrfs_run_delayed_iputs(root);

635 636
	if (trans->aborted ||
	    root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
637
		err = -EIO;
638
	}
639
	assert_qgroups_uptodate(trans);
640

641 642 643
	memset(trans, 0, sizeof(*trans));
	kmem_cache_free(btrfs_trans_handle_cachep, trans);
	return err;
C
Chris Mason 已提交
644 645
}

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

651
	ret = __btrfs_end_transaction(trans, root, 0);
652 653 654
	if (ret)
		return ret;
	return 0;
655 656 657 658 659
}

int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
				   struct btrfs_root *root)
{
660 661
	int ret;

662
	ret = __btrfs_end_transaction(trans, root, 1);
663 664 665 666 667 668 669 670
	if (ret)
		return ret;
	return 0;
}

int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans,
				struct btrfs_root *root)
{
671
	return __btrfs_end_transaction(trans, root, 1);
672 673
}

C
Chris Mason 已提交
674 675 676
/*
 * 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
677
 * those extents are sent to disk but does not wait on them
C
Chris Mason 已提交
678
 */
679
int btrfs_write_marked_extents(struct btrfs_root *root,
680
			       struct extent_io_tree *dirty_pages, int mark)
C
Chris Mason 已提交
681
{
682
	int err = 0;
683
	int werr = 0;
J
Josef Bacik 已提交
684
	struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
685
	struct extent_state *cached_state = NULL;
686
	u64 start = 0;
687
	u64 end;
688

J
Josef Bacik 已提交
689
	while (!find_first_extent_bit(dirty_pages, start, &start, &end,
690 691 692 693
				      mark, &cached_state)) {
		convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT,
				   mark, &cached_state, GFP_NOFS);
		cached_state = NULL;
J
Josef Bacik 已提交
694 695 696 697 698
		err = filemap_fdatawrite_range(mapping, start, end);
		if (err)
			werr = err;
		cond_resched();
		start = end + 1;
699
	}
700 701 702 703 704 705 706 707 708 709 710 711
	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,
712
			      struct extent_io_tree *dirty_pages, int mark)
713 714 715
{
	int err = 0;
	int werr = 0;
J
Josef Bacik 已提交
716
	struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
717
	struct extent_state *cached_state = NULL;
718 719
	u64 start = 0;
	u64 end;
720

J
Josef Bacik 已提交
721
	while (!find_first_extent_bit(dirty_pages, start, &start, &end,
722 723 724
				      EXTENT_NEED_WAIT, &cached_state)) {
		clear_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT,
				 0, 0, &cached_state, GFP_NOFS);
J
Josef Bacik 已提交
725 726 727 728 729
		err = filemap_fdatawait_range(mapping, start, end);
		if (err)
			werr = err;
		cond_resched();
		start = end + 1;
730
	}
731 732 733
	if (err)
		werr = err;
	return werr;
C
Chris Mason 已提交
734 735
}

736 737 738 739 740 741
/*
 * 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,
742
				struct extent_io_tree *dirty_pages, int mark)
743 744 745 746
{
	int ret;
	int ret2;

747 748
	ret = btrfs_write_marked_extents(root, dirty_pages, mark);
	ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
749 750 751 752 753 754

	if (ret)
		return ret;
	if (ret2)
		return ret2;
	return 0;
755 756
}

757 758 759 760 761 762 763 764 765
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,
766 767
					   &trans->transaction->dirty_pages,
					   EXTENT_DIRTY);
768 769
}

C
Chris Mason 已提交
770 771 772 773 774 775 776 777 778 779
/*
 * 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.
 */
780 781
static int update_cowonly_root(struct btrfs_trans_handle *trans,
			       struct btrfs_root *root)
C
Chris Mason 已提交
782 783
{
	int ret;
784
	u64 old_root_bytenr;
785
	u64 old_root_used;
786
	struct btrfs_root *tree_root = root->fs_info->tree_root;
C
Chris Mason 已提交
787

788
	old_root_used = btrfs_root_used(&root->root_item);
789
	btrfs_write_dirty_block_groups(trans, root);
790

C
Chris Mason 已提交
791
	while (1) {
792
		old_root_bytenr = btrfs_root_bytenr(&root->root_item);
793 794
		if (old_root_bytenr == root->node->start &&
		    old_root_used == btrfs_root_used(&root->root_item))
C
Chris Mason 已提交
795
			break;
796

797
		btrfs_set_root_node(&root->root_item, root->node);
C
Chris Mason 已提交
798
		ret = btrfs_update_root(trans, tree_root,
799 800
					&root->root_key,
					&root->root_item);
801 802
		if (ret)
			return ret;
803

804
		old_root_used = btrfs_root_used(&root->root_item);
805
		ret = btrfs_write_dirty_block_groups(trans, root);
806 807
		if (ret)
			return ret;
808
	}
809 810 811 812

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

813 814 815
	return 0;
}

C
Chris Mason 已提交
816 817
/*
 * update all the cowonly tree roots on disk
818 819 820 821
 *
 * 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
822
 */
823 824
static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
					 struct btrfs_root *root)
825 826 827
{
	struct btrfs_fs_info *fs_info = root->fs_info;
	struct list_head *next;
828
	struct extent_buffer *eb;
829
	int ret;
830

831
	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
832 833
	if (ret)
		return ret;
834

835
	eb = btrfs_lock_root_node(fs_info->tree_root);
836 837
	ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
			      0, &eb);
838 839
	btrfs_tree_unlock(eb);
	free_extent_buffer(eb);
840

841 842 843
	if (ret)
		return ret;

844
	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
845 846
	if (ret)
		return ret;
847

848
	ret = btrfs_run_dev_stats(trans, root->fs_info);
849 850 851
	WARN_ON(ret);
	ret = btrfs_run_dev_replace(trans, root->fs_info);
	WARN_ON(ret);
852

853 854 855 856 857 858 859
	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
860
	while (!list_empty(&fs_info->dirty_cowonly_roots)) {
861 862 863
		next = fs_info->dirty_cowonly_roots.next;
		list_del_init(next);
		root = list_entry(next, struct btrfs_root, dirty_list);
864

865 866 867
		ret = update_cowonly_root(trans, root);
		if (ret)
			return ret;
C
Chris Mason 已提交
868
	}
869 870 871 872 873

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

874 875
	btrfs_after_dev_replace_commit(fs_info);

C
Chris Mason 已提交
876 877 878
	return 0;
}

C
Chris Mason 已提交
879 880 881 882 883
/*
 * 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
 */
884
int btrfs_add_dead_root(struct btrfs_root *root)
885
{
J
Josef Bacik 已提交
886
	spin_lock(&root->fs_info->trans_lock);
887
	list_add(&root->root_list, &root->fs_info->dead_roots);
J
Josef Bacik 已提交
888
	spin_unlock(&root->fs_info->trans_lock);
889 890 891
	return 0;
}

C
Chris Mason 已提交
892
/*
893
 * update all the cowonly tree roots on disk
C
Chris Mason 已提交
894
 */
895 896
static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
				    struct btrfs_root *root)
897 898
{
	struct btrfs_root *gang[8];
899
	struct btrfs_fs_info *fs_info = root->fs_info;
900 901
	int i;
	int ret;
902 903
	int err = 0;

J
Josef Bacik 已提交
904
	spin_lock(&fs_info->fs_roots_radix_lock);
C
Chris Mason 已提交
905
	while (1) {
906 907
		ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
						 (void **)gang, 0,
908 909 910 911 912 913
						 ARRAY_SIZE(gang),
						 BTRFS_ROOT_TRANS_TAG);
		if (ret == 0)
			break;
		for (i = 0; i < ret; i++) {
			root = gang[i];
914 915 916
			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
917
			spin_unlock(&fs_info->fs_roots_radix_lock);
Y
Yan Zheng 已提交
918

919
			btrfs_free_log(trans, root);
920
			btrfs_update_reloc_root(trans, root);
921
			btrfs_orphan_commit_root(trans, root);
922

923 924
			btrfs_save_ino_cache(root, trans);

925 926 927 928
			/* see comments in should_cow_block() */
			root->force_cow = 0;
			smp_wmb();

929
			if (root->commit_root != root->node) {
930
				mutex_lock(&root->fs_commit_mutex);
J
Josef Bacik 已提交
931
				switch_commit_root(root);
932 933 934
				btrfs_unpin_free_ino(root);
				mutex_unlock(&root->fs_commit_mutex);

935 936 937
				btrfs_set_root_node(&root->root_item,
						    root->node);
			}
938 939

			err = btrfs_update_root(trans, fs_info->tree_root,
940 941
						&root->root_key,
						&root->root_item);
J
Josef Bacik 已提交
942
			spin_lock(&fs_info->fs_roots_radix_lock);
943 944
			if (err)
				break;
945 946
		}
	}
J
Josef Bacik 已提交
947
	spin_unlock(&fs_info->fs_roots_radix_lock);
948
	return err;
949 950
}

C
Chris Mason 已提交
951 952 953 954
/*
 * defrag a given btree.  If cacheonly == 1, this won't read from the disk,
 * otherwise every leaf in the btree is read and defragged.
 */
955 956 957 958
int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
{
	struct btrfs_fs_info *info = root->fs_info;
	struct btrfs_trans_handle *trans;
959
	int ret;
960

961
	if (xchg(&root->defrag_running, 1))
962
		return 0;
963

964
	while (1) {
965 966 967 968
		trans = btrfs_start_transaction(root, 0);
		if (IS_ERR(trans))
			return PTR_ERR(trans);

969
		ret = btrfs_defrag_leaves(trans, root, cacheonly);
970

971
		btrfs_end_transaction(trans, root);
972
		btrfs_btree_balance_dirty(info->tree_root);
973 974
		cond_resched();

975
		if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
976 977 978
			break;
	}
	root->defrag_running = 0;
979
	return ret;
980 981
}

C
Chris Mason 已提交
982 983 984 985
/*
 * new snapshots need to be created at a very specific time in the
 * transaction commit.  This does the actual creation
 */
986
static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
987 988 989 990
				   struct btrfs_fs_info *fs_info,
				   struct btrfs_pending_snapshot *pending)
{
	struct btrfs_key key;
991
	struct btrfs_root_item *new_root_item;
992 993
	struct btrfs_root *tree_root = fs_info->tree_root;
	struct btrfs_root *root = pending->root;
994
	struct btrfs_root *parent_root;
L
Liu Bo 已提交
995
	struct btrfs_block_rsv *rsv;
996
	struct inode *parent_inode;
997 998
	struct btrfs_path *path;
	struct btrfs_dir_item *dir_item;
999
	struct dentry *parent;
1000
	struct dentry *dentry;
1001
	struct extent_buffer *tmp;
1002
	struct extent_buffer *old;
1003
	struct timespec cur_time = CURRENT_TIME;
1004
	int ret;
1005
	u64 to_reserve = 0;
1006
	u64 index = 0;
1007
	u64 objectid;
L
Li Zefan 已提交
1008
	u64 root_flags;
1009
	uuid_le new_uuid;
1010

1011 1012 1013 1014 1015 1016
	path = btrfs_alloc_path();
	if (!path) {
		ret = pending->error = -ENOMEM;
		goto path_alloc_fail;
	}

1017 1018
	new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
	if (!new_root_item) {
1019
		ret = pending->error = -ENOMEM;
1020
		goto root_item_alloc_fail;
1021
	}
1022

1023
	ret = btrfs_find_free_objectid(tree_root, &objectid);
1024 1025
	if (ret) {
		pending->error = ret;
1026
		goto no_free_objectid;
1027
	}
1028

1029
	btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
1030 1031

	if (to_reserve > 0) {
M
Miao Xie 已提交
1032 1033 1034
		ret = btrfs_block_rsv_add(root, &pending->block_rsv,
					  to_reserve,
					  BTRFS_RESERVE_NO_FLUSH);
1035 1036
		if (ret) {
			pending->error = ret;
1037
			goto no_free_objectid;
1038 1039 1040
		}
	}

A
Arne Jansen 已提交
1041 1042 1043 1044
	ret = btrfs_qgroup_inherit(trans, fs_info, root->root_key.objectid,
				   objectid, pending->inherit);
	if (ret) {
		pending->error = ret;
1045
		goto no_free_objectid;
A
Arne Jansen 已提交
1046 1047
	}

1048
	key.objectid = objectid;
1049 1050
	key.offset = (u64)-1;
	key.type = BTRFS_ROOT_ITEM_KEY;
1051

1052
	rsv = trans->block_rsv;
1053
	trans->block_rsv = &pending->block_rsv;
1054

1055
	dentry = pending->dentry;
1056 1057
	parent = dget_parent(dentry);
	parent_inode = parent->d_inode;
1058
	parent_root = BTRFS_I(parent_inode)->root;
C
Chris Mason 已提交
1059
	record_root_in_trans(trans, parent_root);
1060

1061 1062 1063
	/*
	 * insert the directory item
	 */
1064
	ret = btrfs_set_inode_index(parent_inode, &index);
1065
	BUG_ON(ret); /* -ENOMEM */
1066 1067 1068 1069 1070 1071 1072

	/* 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)) {
1073 1074
		pending->error = -EEXIST;
		goto fail;
1075 1076
	} else if (IS_ERR(dir_item)) {
		ret = PTR_ERR(dir_item);
1077 1078
		btrfs_abort_transaction(trans, root, ret);
		goto fail;
1079
	}
1080
	btrfs_release_path(path);
1081

1082 1083 1084 1085 1086 1087 1088
	/*
	 * 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);
1089 1090 1091 1092
	if (ret) {	/* Transaction aborted */
		btrfs_abort_transaction(trans, root, ret);
		goto fail;
	}
1093

C
Chris Mason 已提交
1094
	record_root_in_trans(trans, root);
1095 1096
	btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
	memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1097
	btrfs_check_and_init_root_item(new_root_item);
1098

L
Li Zefan 已提交
1099 1100 1101 1102 1103 1104 1105
	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);

1106 1107 1108 1109 1110 1111 1112
	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);
1113
	new_root_item->otime.nsec = cpu_to_le32(cur_time.tv_nsec);
1114 1115 1116 1117 1118 1119
	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);

1120
	old = btrfs_lock_root_node(root);
1121
	ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
1122 1123 1124
	if (ret) {
		btrfs_tree_unlock(old);
		free_extent_buffer(old);
1125 1126
		btrfs_abort_transaction(trans, root, ret);
		goto fail;
1127
	}
1128

1129 1130
	btrfs_set_lock_blocking(old);

1131
	ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1132
	/* clean up in any case */
1133 1134
	btrfs_tree_unlock(old);
	free_extent_buffer(old);
1135 1136 1137 1138
	if (ret) {
		btrfs_abort_transaction(trans, root, ret);
		goto fail;
	}
1139

1140 1141 1142 1143
	/* see comments in should_cow_block() */
	root->force_cow = 1;
	smp_wmb();

1144
	btrfs_set_root_node(new_root_item, tmp);
1145 1146 1147
	/* record when the snapshot was created in key.offset */
	key.offset = trans->transid;
	ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1148 1149
	btrfs_tree_unlock(tmp);
	free_extent_buffer(tmp);
1150 1151 1152 1153
	if (ret) {
		btrfs_abort_transaction(trans, root, ret);
		goto fail;
	}
1154

1155 1156 1157 1158
	/*
	 * insert root back/forward references
	 */
	ret = btrfs_add_root_ref(trans, tree_root, objectid,
1159
				 parent_root->root_key.objectid,
L
Li Zefan 已提交
1160
				 btrfs_ino(parent_inode), index,
1161
				 dentry->d_name.name, dentry->d_name.len);
1162 1163 1164 1165
	if (ret) {
		btrfs_abort_transaction(trans, root, ret);
		goto fail;
	}
1166

1167 1168
	key.offset = (u64)-1;
	pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1169 1170
	if (IS_ERR(pending->snap)) {
		ret = PTR_ERR(pending->snap);
1171 1172
		btrfs_abort_transaction(trans, root, ret);
		goto fail;
1173
	}
1174

1175
	ret = btrfs_reloc_post_snapshot(trans, pending);
1176 1177 1178 1179
	if (ret) {
		btrfs_abort_transaction(trans, root, ret);
		goto fail;
	}
1180 1181

	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1182 1183 1184 1185
	if (ret) {
		btrfs_abort_transaction(trans, root, ret);
		goto fail;
	}
1186 1187 1188 1189 1190 1191 1192

	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);
1193 1194 1195 1196
	if (ret) {
		btrfs_abort_transaction(trans, root, ret);
		goto fail;
	}
1197 1198 1199 1200

	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;
1201
	ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode);
1202
	if (ret)
1203
		btrfs_abort_transaction(trans, root, ret);
1204
fail:
1205
	dput(parent);
L
Liu Bo 已提交
1206
	trans->block_rsv = rsv;
1207 1208 1209
no_free_objectid:
	kfree(new_root_item);
root_item_alloc_fail:
1210 1211
	btrfs_free_path(path);
path_alloc_fail:
1212
	btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1);
1213
	return ret;
1214 1215
}

C
Chris Mason 已提交
1216 1217 1218
/*
 * create all the snapshots we've scheduled for creation
 */
1219 1220
static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
					     struct btrfs_fs_info *fs_info)
1221 1222 1223 1224
{
	struct btrfs_pending_snapshot *pending;
	struct list_head *head = &trans->transaction->pending_snapshots;

1225 1226
	list_for_each_entry(pending, head, list)
		create_pending_snapshot(trans, fs_info, pending);
1227 1228 1229
	return 0;
}

1230 1231 1232 1233 1234
static void update_super_roots(struct btrfs_root *root)
{
	struct btrfs_root_item *root_item;
	struct btrfs_super_block *super;

1235
	super = root->fs_info->super_copy;
1236 1237 1238 1239 1240 1241 1242 1243 1244 1245

	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;
1246
	if (btrfs_test_opt(root, SPACE_CACHE))
1247
		super->cache_generation = root_item->generation;
1248 1249
}

1250 1251 1252
int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
{
	int ret = 0;
J
Josef Bacik 已提交
1253
	spin_lock(&info->trans_lock);
1254 1255
	if (info->running_transaction)
		ret = info->running_transaction->in_commit;
J
Josef Bacik 已提交
1256
	spin_unlock(&info->trans_lock);
1257 1258 1259
	return ret;
}

1260 1261 1262
int btrfs_transaction_blocked(struct btrfs_fs_info *info)
{
	int ret = 0;
J
Josef Bacik 已提交
1263
	spin_lock(&info->trans_lock);
1264 1265
	if (info->running_transaction)
		ret = info->running_transaction->blocked;
J
Josef Bacik 已提交
1266
	spin_unlock(&info->trans_lock);
1267 1268 1269
	return ret;
}

S
Sage Weil 已提交
1270 1271 1272 1273 1274 1275 1276
/*
 * 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
1277
	wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit);
S
Sage Weil 已提交
1278 1279 1280 1281 1282 1283 1284 1285 1286
}

/*
 * 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
1287 1288
	wait_event(root->fs_info->transaction_wait,
		   trans->commit_done || (trans->in_commit && !trans->blocked));
S
Sage Weil 已提交
1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305
}

/*
 * 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);

1306 1307 1308 1309
	/*
	 * We've got freeze protection passed with the transaction.
	 * Tell lockdep about it.
	 */
1310 1311 1312 1313
	if (ac->newtrans->type < TRANS_JOIN_NOLOCK)
		rwsem_acquire_read(
		     &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
		     0, 1, _THIS_IP_);
1314

1315 1316
	current->journal_info = ac->newtrans;

S
Sage Weil 已提交
1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328
	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
1329 1330
	if (!ac)
		return -ENOMEM;
S
Sage Weil 已提交
1331 1332 1333

	INIT_DELAYED_WORK(&ac->work, do_async_commit);
	ac->root = root;
1334
	ac->newtrans = btrfs_join_transaction(root);
1335 1336 1337 1338 1339
	if (IS_ERR(ac->newtrans)) {
		int err = PTR_ERR(ac->newtrans);
		kfree(ac);
		return err;
	}
S
Sage Weil 已提交
1340 1341 1342

	/* take transaction reference */
	cur_trans = trans->transaction;
1343
	atomic_inc(&cur_trans->use_count);
S
Sage Weil 已提交
1344 1345

	btrfs_end_transaction(trans, root);
1346 1347 1348 1349 1350

	/*
	 * Tell lockdep we've released the freeze rwsem, since the
	 * async commit thread will be the one to unlock it.
	 */
1351 1352 1353 1354
	if (trans->type < TRANS_JOIN_NOLOCK)
		rwsem_release(
			&root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
			1, _THIS_IP_);
1355

S
Sage Weil 已提交
1356 1357 1358 1359 1360 1361 1362 1363
	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);

1364 1365 1366 1367
	if (current->journal_info == trans)
		current->journal_info = NULL;

	put_transaction(cur_trans);
S
Sage Weil 已提交
1368 1369 1370
	return 0;
}

1371 1372

static void cleanup_transaction(struct btrfs_trans_handle *trans,
1373
				struct btrfs_root *root, int err)
1374 1375 1376 1377 1378
{
	struct btrfs_transaction *cur_trans = trans->transaction;

	WARN_ON(trans->use_count > 1);

1379 1380
	btrfs_abort_transaction(trans, root, err);

1381 1382
	spin_lock(&root->fs_info->trans_lock);
	list_del_init(&cur_trans->list);
1383 1384 1385 1386
	if (cur_trans == root->fs_info->running_transaction) {
		root->fs_info->running_transaction = NULL;
		root->fs_info->trans_no_join = 0;
	}
1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403
	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);
}

1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445
static int btrfs_flush_all_pending_stuffs(struct btrfs_trans_handle *trans,
					  struct btrfs_root *root)
{
	int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
	int snap_pending = 0;
	int ret;

	if (!flush_on_commit) {
		spin_lock(&root->fs_info->trans_lock);
		if (!list_empty(&trans->transaction->pending_snapshots))
			snap_pending = 1;
		spin_unlock(&root->fs_info->trans_lock);
	}

	if (flush_on_commit || snap_pending) {
		btrfs_start_delalloc_inodes(root, 1);
		btrfs_wait_ordered_extents(root, 1);
	}

	ret = btrfs_run_delayed_items(trans, root);
	if (ret)
		return ret;

	/*
	 * 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);

	/*
	 * 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);

	return 0;
}

S
Sage Weil 已提交
1446 1447 1448 1449 1450 1451 1452
/*
 * 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
1453 1454 1455
int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
			     struct btrfs_root *root)
{
1456
	unsigned long joined = 0;
1457
	struct btrfs_transaction *cur_trans = trans->transaction;
C
Chris Mason 已提交
1458
	struct btrfs_transaction *prev_trans = NULL;
C
Chris Mason 已提交
1459
	DEFINE_WAIT(wait);
1460
	int ret;
1461 1462
	int should_grow = 0;
	unsigned long now = get_seconds();
C
Chris Mason 已提交
1463

1464 1465 1466 1467 1468
	ret = btrfs_run_ordered_operations(root, 0);
	if (ret) {
		btrfs_abort_transaction(trans, root, ret);
		goto cleanup_transaction;
	}
1469

1470 1471
	if (cur_trans->aborted) {
		ret = cur_trans->aborted;
1472
		goto cleanup_transaction;
1473
	}
1474

1475 1476 1477 1478
	/* 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);
1479 1480
	if (ret)
		goto cleanup_transaction;
1481

1482 1483 1484
	btrfs_trans_release_metadata(trans, root);
	trans->block_rsv = NULL;

1485
	cur_trans = trans->transaction;
1486

1487 1488 1489 1490
	/*
	 * set the flushing flag so procs in this transaction have to
	 * start sending their work down.
	 */
1491
	cur_trans->delayed_refs.flushing = 1;
1492

1493 1494 1495
	if (!list_empty(&trans->new_bgs))
		btrfs_create_pending_block_groups(trans, root);

1496
	ret = btrfs_run_delayed_refs(trans, root, 0);
1497 1498
	if (ret)
		goto cleanup_transaction;
1499

J
Josef Bacik 已提交
1500
	spin_lock(&cur_trans->commit_lock);
1501
	if (cur_trans->in_commit) {
J
Josef Bacik 已提交
1502
		spin_unlock(&cur_trans->commit_lock);
1503
		atomic_inc(&cur_trans->use_count);
1504
		ret = btrfs_end_transaction(trans, root);
C
Chris Mason 已提交
1505

1506
		wait_for_commit(root, cur_trans);
1507

C
Chris Mason 已提交
1508
		put_transaction(cur_trans);
1509

1510
		return ret;
C
Chris Mason 已提交
1511
	}
1512

C
Chris Mason 已提交
1513
	trans->transaction->in_commit = 1;
1514
	trans->transaction->blocked = 1;
J
Josef Bacik 已提交
1515
	spin_unlock(&cur_trans->commit_lock);
S
Sage Weil 已提交
1516 1517
	wake_up(&root->fs_info->transaction_blocked_wait);

J
Josef Bacik 已提交
1518
	spin_lock(&root->fs_info->trans_lock);
C
Chris Mason 已提交
1519 1520 1521 1522
	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) {
1523
			atomic_inc(&prev_trans->use_count);
J
Josef Bacik 已提交
1524
			spin_unlock(&root->fs_info->trans_lock);
C
Chris Mason 已提交
1525 1526 1527

			wait_for_commit(root, prev_trans);

1528
			put_transaction(prev_trans);
J
Josef Bacik 已提交
1529 1530
		} else {
			spin_unlock(&root->fs_info->trans_lock);
C
Chris Mason 已提交
1531
		}
J
Josef Bacik 已提交
1532 1533
	} else {
		spin_unlock(&root->fs_info->trans_lock);
C
Chris Mason 已提交
1534
	}
1535

1536 1537
	if (!btrfs_test_opt(root, SSD) &&
	    (now < cur_trans->start_time || now - cur_trans->start_time < 1))
1538 1539
		should_grow = 1;

1540 1541
	do {
		joined = cur_trans->num_joined;
1542

C
Chris Mason 已提交
1543
		WARN_ON(cur_trans != trans->transaction);
1544

1545
		ret = btrfs_flush_all_pending_stuffs(trans, root);
1546 1547
		if (ret)
			goto cleanup_transaction;
1548

1549 1550 1551
		prepare_to_wait(&cur_trans->writer_wait, &wait,
				TASK_UNINTERRUPTIBLE);

1552
		if (atomic_read(&cur_trans->num_writers) > 1)
1553 1554 1555
			schedule_timeout(MAX_SCHEDULE_TIMEOUT);
		else if (should_grow)
			schedule_timeout(1);
1556 1557

		finish_wait(&cur_trans->writer_wait, &wait);
1558
	} while (atomic_read(&cur_trans->num_writers) > 1 ||
1559
		 (should_grow && cur_trans->num_joined != joined));
1560

1561 1562 1563 1564
	ret = btrfs_flush_all_pending_stuffs(trans, root);
	if (ret)
		goto cleanup_transaction;

1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575
	/*
	 * 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
1576 1577 1578 1579 1580 1581 1582
	/*
	 * 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);

1583 1584 1585 1586 1587 1588
	/*
	 * 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);
1589 1590 1591 1592
	if (ret) {
		mutex_unlock(&root->fs_info->reloc_mutex);
		goto cleanup_transaction;
	}
1593

1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604
	/*
	 * 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);
1605 1606 1607 1608
	if (ret) {
		mutex_unlock(&root->fs_info->reloc_mutex);
		goto cleanup_transaction;
	}
1609

1610
	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1611 1612 1613 1614
	if (ret) {
		mutex_unlock(&root->fs_info->reloc_mutex);
		goto cleanup_transaction;
	}
1615

1616 1617 1618 1619 1620 1621
	/*
	 * 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
1622
	WARN_ON(cur_trans != trans->transaction);
C
Chris Mason 已提交
1623

A
Arne Jansen 已提交
1624
	btrfs_scrub_pause(root);
1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639
	/* 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);

1640
	ret = commit_fs_roots(trans, root);
1641 1642
	if (ret) {
		mutex_unlock(&root->fs_info->tree_log_mutex);
1643
		mutex_unlock(&root->fs_info->reloc_mutex);
1644 1645
		goto cleanup_transaction;
	}
1646

1647
	/* commit_fs_roots gets rid of all the tree log roots, it is now
1648 1649 1650 1651
	 * safe to free the root of tree log roots
	 */
	btrfs_free_log_root_tree(trans, root->fs_info);

1652
	ret = commit_cowonly_roots(trans, root);
1653 1654
	if (ret) {
		mutex_unlock(&root->fs_info->tree_log_mutex);
1655
		mutex_unlock(&root->fs_info->reloc_mutex);
1656 1657
		goto cleanup_transaction;
	}
1658

1659 1660
	btrfs_prepare_extent_commit(trans, root);

C
Chris Mason 已提交
1661
	cur_trans = root->fs_info->running_transaction;
1662 1663 1664

	btrfs_set_root_node(&root->fs_info->tree_root->root_item,
			    root->fs_info->tree_root->node);
J
Josef Bacik 已提交
1665
	switch_commit_root(root->fs_info->tree_root);
1666 1667 1668

	btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
			    root->fs_info->chunk_root->node);
J
Josef Bacik 已提交
1669
	switch_commit_root(root->fs_info->chunk_root);
1670

1671
	assert_qgroups_uptodate(trans);
1672
	update_super_roots(root);
1673 1674

	if (!root->fs_info->log_root_recovering) {
1675 1676
		btrfs_set_super_log_root(root->fs_info->super_copy, 0);
		btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
1677 1678
	}

1679 1680
	memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
	       sizeof(*root->fs_info->super_copy));
C
Chris Mason 已提交
1681

1682
	trans->transaction->blocked = 0;
J
Josef Bacik 已提交
1683 1684 1685 1686
	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
1687
	mutex_unlock(&root->fs_info->reloc_mutex);
1688

1689
	wake_up(&root->fs_info->transaction_wait);
1690

C
Chris Mason 已提交
1691
	ret = btrfs_write_and_wait_transaction(trans, root);
1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703
	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;
	}
1704

1705 1706 1707 1708 1709 1710
	/*
	 * the super is written, we can safely allow the tree-loggers
	 * to go about their business
	 */
	mutex_unlock(&root->fs_info->tree_log_mutex);

1711
	btrfs_finish_extent_commit(trans, root);
1712

C
Chris Mason 已提交
1713
	cur_trans->commit_done = 1;
1714

1715
	root->fs_info->last_trans_committed = cur_trans->transid;
J
Josef Bacik 已提交
1716

C
Chris Mason 已提交
1717
	wake_up(&cur_trans->commit_wait);
1718

J
Josef Bacik 已提交
1719
	spin_lock(&root->fs_info->trans_lock);
1720
	list_del_init(&cur_trans->list);
J
Josef Bacik 已提交
1721 1722
	spin_unlock(&root->fs_info->trans_lock);

C
Chris Mason 已提交
1723
	put_transaction(cur_trans);
C
Chris Mason 已提交
1724
	put_transaction(cur_trans);
1725

1726 1727
	if (trans->type < TRANS_JOIN_NOLOCK)
		sb_end_intwrite(root->fs_info->sb);
1728

1729 1730
	trace_btrfs_transaction_commit(root);

A
Arne Jansen 已提交
1731 1732
	btrfs_scrub_continue(root);

J
Josef Bacik 已提交
1733 1734 1735
	if (current->journal_info == trans)
		current->journal_info = NULL;

C
Chris Mason 已提交
1736
	kmem_cache_free(btrfs_trans_handle_cachep, trans);
Y
Yan, Zheng 已提交
1737 1738 1739 1740

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

C
Chris Mason 已提交
1741
	return ret;
1742 1743

cleanup_transaction:
1744 1745
	btrfs_trans_release_metadata(trans, root);
	trans->block_rsv = NULL;
1746 1747 1748 1749
	btrfs_printk(root->fs_info, "Skipping commit of aborted transaction.\n");
//	WARN_ON(1);
	if (current->journal_info == trans)
		current->journal_info = NULL;
1750
	cleanup_transaction(trans, root, ret);
1751 1752

	return ret;
C
Chris Mason 已提交
1753 1754
}

C
Chris Mason 已提交
1755 1756 1757
/*
 * interface function to delete all the snapshots we have scheduled for deletion
 */
1758 1759
int btrfs_clean_old_snapshots(struct btrfs_root *root)
{
1760 1761 1762
	LIST_HEAD(list);
	struct btrfs_fs_info *fs_info = root->fs_info;

J
Josef Bacik 已提交
1763
	spin_lock(&fs_info->trans_lock);
1764
	list_splice_init(&fs_info->dead_roots, &list);
J
Josef Bacik 已提交
1765
	spin_unlock(&fs_info->trans_lock);
1766

1767
	while (!list_empty(&list)) {
1768 1769
		int ret;

1770
		root = list_entry(list.next, struct btrfs_root, root_list);
1771 1772
		list_del(&root->root_list);

1773 1774
		btrfs_kill_all_delayed_nodes(root);

1775 1776
		if (btrfs_header_backref_rev(root->node) <
		    BTRFS_MIXED_BACKREF_REV)
1777
			ret = btrfs_drop_snapshot(root, NULL, 0, 0);
1778
		else
1779 1780
			ret =btrfs_drop_snapshot(root, NULL, 1, 0);
		BUG_ON(ret < 0);
1781 1782 1783
	}
	return 0;
}