transaction.c 66.1 KB
Newer Older
Chris Mason's avatar
Chris Mason committed
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.
 */

Chris Mason's avatar
Chris Mason committed
19
#include <linux/fs.h>
20
#include <linux/slab.h>
Chris Mason's avatar
Chris Mason committed
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>
Chris Mason's avatar
Chris Mason committed
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"
34
#include "qgroup.h"
Chris Mason's avatar
Chris Mason committed
35

36 37
#define BTRFS_ROOT_TRANS_TAG 0

38
static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = {
39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
	[TRANS_STATE_RUNNING]		= 0U,
	[TRANS_STATE_BLOCKED]		= (__TRANS_USERSPACE |
					   __TRANS_START),
	[TRANS_STATE_COMMIT_START]	= (__TRANS_USERSPACE |
					   __TRANS_START |
					   __TRANS_ATTACH),
	[TRANS_STATE_COMMIT_DOING]	= (__TRANS_USERSPACE |
					   __TRANS_START |
					   __TRANS_ATTACH |
					   __TRANS_JOIN),
	[TRANS_STATE_UNBLOCKED]		= (__TRANS_USERSPACE |
					   __TRANS_START |
					   __TRANS_ATTACH |
					   __TRANS_JOIN |
					   __TRANS_JOIN_NOLOCK),
	[TRANS_STATE_COMPLETED]		= (__TRANS_USERSPACE |
					   __TRANS_START |
					   __TRANS_ATTACH |
					   __TRANS_JOIN |
					   __TRANS_JOIN_NOLOCK),
};

61
void btrfs_put_transaction(struct btrfs_transaction *transaction)
Chris Mason's avatar
Chris Mason committed
62
{
63 64
	WARN_ON(refcount_read(&transaction->use_count) == 0);
	if (refcount_dec_and_test(&transaction->use_count)) {
Josef Bacik's avatar
Josef Bacik committed
65
		BUG_ON(!list_empty(&transaction->list));
Liu Bo's avatar
Liu Bo committed
66
		WARN_ON(!RB_EMPTY_ROOT(&transaction->delayed_refs.href_root));
67
		if (transaction->delayed_refs.pending_csums)
68 69 70
			btrfs_err(transaction->fs_info,
				  "pending csums is %llu",
				  transaction->delayed_refs.pending_csums);
71 72 73 74 75 76 77 78
		while (!list_empty(&transaction->pending_chunks)) {
			struct extent_map *em;

			em = list_first_entry(&transaction->pending_chunks,
					      struct extent_map, list);
			list_del_init(&em->list);
			free_extent_map(em);
		}
79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95
		/*
		 * If any block groups are found in ->deleted_bgs then it's
		 * because the transaction was aborted and a commit did not
		 * happen (things failed before writing the new superblock
		 * and calling btrfs_finish_extent_commit()), so we can not
		 * discard the physical locations of the block groups.
		 */
		while (!list_empty(&transaction->deleted_bgs)) {
			struct btrfs_block_group_cache *cache;

			cache = list_first_entry(&transaction->deleted_bgs,
						 struct btrfs_block_group_cache,
						 bg_list);
			list_del_init(&cache->bg_list);
			btrfs_put_block_group_trimming(cache);
			btrfs_put_block_group(cache);
		}
96
		kfree(transaction);
97
	}
Chris Mason's avatar
Chris Mason committed
98 99
}

100 101 102
static void clear_btree_io_tree(struct extent_io_tree *tree)
{
	spin_lock(&tree->lock);
103 104 105 106 107 108
	/*
	 * Do a single barrier for the waitqueue_active check here, the state
	 * of the waitqueue should not change once clear_btree_io_tree is
	 * called.
	 */
	smp_mb();
109 110 111 112 113 114 115 116 117 118 119 120 121 122
	while (!RB_EMPTY_ROOT(&tree->state)) {
		struct rb_node *node;
		struct extent_state *state;

		node = rb_first(&tree->state);
		state = rb_entry(node, struct extent_state, rb_node);
		rb_erase(&state->rb_node, &tree->state);
		RB_CLEAR_NODE(&state->rb_node);
		/*
		 * btree io trees aren't supposed to have tasks waiting for
		 * changes in the flags of extent states ever.
		 */
		ASSERT(!waitqueue_active(&state->wq));
		free_extent_state(state);
123 124

		cond_resched_lock(&tree->lock);
125 126 127 128
	}
	spin_unlock(&tree->lock);
}

129
static noinline void switch_commit_roots(struct btrfs_transaction *trans)
130
{
131
	struct btrfs_fs_info *fs_info = trans->fs_info;
132 133 134 135 136 137 138 139 140 141
	struct btrfs_root *root, *tmp;

	down_write(&fs_info->commit_root_sem);
	list_for_each_entry_safe(root, tmp, &trans->switch_commits,
				 dirty_list) {
		list_del_init(&root->dirty_list);
		free_extent_buffer(root->commit_root);
		root->commit_root = btrfs_root_node(root);
		if (is_fstree(root->objectid))
			btrfs_unpin_free_ino(root);
142
		clear_btree_io_tree(&root->dirty_log_pages);
143
	}
144 145 146 147 148 149 150 151 152 153 154 155

	/* We can free old roots now. */
	spin_lock(&trans->dropped_roots_lock);
	while (!list_empty(&trans->dropped_roots)) {
		root = list_first_entry(&trans->dropped_roots,
					struct btrfs_root, root_list);
		list_del_init(&root->root_list);
		spin_unlock(&trans->dropped_roots_lock);
		btrfs_drop_and_free_fs_root(fs_info, root);
		spin_lock(&trans->dropped_roots_lock);
	}
	spin_unlock(&trans->dropped_roots_lock);
156
	up_write(&fs_info->commit_root_sem);
157 158
}

159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181
static inline void extwriter_counter_inc(struct btrfs_transaction *trans,
					 unsigned int type)
{
	if (type & TRANS_EXTWRITERS)
		atomic_inc(&trans->num_extwriters);
}

static inline void extwriter_counter_dec(struct btrfs_transaction *trans,
					 unsigned int type)
{
	if (type & TRANS_EXTWRITERS)
		atomic_dec(&trans->num_extwriters);
}

static inline void extwriter_counter_init(struct btrfs_transaction *trans,
					  unsigned int type)
{
	atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0));
}

static inline int extwriter_counter_read(struct btrfs_transaction *trans)
{
	return atomic_read(&trans->num_extwriters);
182 183
}

184 185 186
/*
 * either allocate a new transaction or hop into the existing one
 */
187 188
static noinline int join_transaction(struct btrfs_fs_info *fs_info,
				     unsigned int type)
Chris Mason's avatar
Chris Mason committed
189 190
{
	struct btrfs_transaction *cur_trans;
Josef Bacik's avatar
Josef Bacik committed
191

192
	spin_lock(&fs_info->trans_lock);
193
loop:
194
	/* The file system has been taken offline. No new transactions. */
195
	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
196
		spin_unlock(&fs_info->trans_lock);
197 198 199
		return -EROFS;
	}

200
	cur_trans = fs_info->running_transaction;
Josef Bacik's avatar
Josef Bacik committed
201
	if (cur_trans) {
202
		if (cur_trans->aborted) {
203
			spin_unlock(&fs_info->trans_lock);
204
			return cur_trans->aborted;
205
		}
206
		if (btrfs_blocked_trans_types[cur_trans->state] & type) {
207 208 209
			spin_unlock(&fs_info->trans_lock);
			return -EBUSY;
		}
210
		refcount_inc(&cur_trans->use_count);
211
		atomic_inc(&cur_trans->num_writers);
212
		extwriter_counter_inc(cur_trans, type);
213
		spin_unlock(&fs_info->trans_lock);
Josef Bacik's avatar
Josef Bacik committed
214
		return 0;
Chris Mason's avatar
Chris Mason committed
215
	}
216
	spin_unlock(&fs_info->trans_lock);
Josef Bacik's avatar
Josef Bacik committed
217

218 219 220 221 222 223 224
	/*
	 * 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;

225 226 227 228 229 230
	/*
	 * JOIN_NOLOCK only happens during the transaction commit, so
	 * it is impossible that ->running_transaction is NULL
	 */
	BUG_ON(type == TRANS_JOIN_NOLOCK);

231
	cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS);
Josef Bacik's avatar
Josef Bacik committed
232 233
	if (!cur_trans)
		return -ENOMEM;
234

235 236
	spin_lock(&fs_info->trans_lock);
	if (fs_info->running_transaction) {
237 238
		/*
		 * someone started a transaction after we unlocked.  Make sure
239
		 * to redo the checks above
240
		 */
241
		kfree(cur_trans);
242
		goto loop;
243
	} else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
244
		spin_unlock(&fs_info->trans_lock);
245
		kfree(cur_trans);
246
		return -EROFS;
Chris Mason's avatar
Chris Mason committed
247
	}
248

249
	cur_trans->fs_info = fs_info;
Josef Bacik's avatar
Josef Bacik committed
250
	atomic_set(&cur_trans->num_writers, 1);
251
	extwriter_counter_init(cur_trans, type);
Josef Bacik's avatar
Josef Bacik committed
252 253
	init_waitqueue_head(&cur_trans->writer_wait);
	init_waitqueue_head(&cur_trans->commit_wait);
254
	init_waitqueue_head(&cur_trans->pending_wait);
255
	cur_trans->state = TRANS_STATE_RUNNING;
Josef Bacik's avatar
Josef Bacik committed
256 257 258 259
	/*
	 * One for this trans handle, one so it will live on until we
	 * commit the transaction.
	 */
260
	refcount_set(&cur_trans->use_count, 2);
261
	atomic_set(&cur_trans->pending_ordered, 0);
262
	cur_trans->flags = 0;
Josef Bacik's avatar
Josef Bacik committed
263 264
	cur_trans->start_time = get_seconds();

265 266
	memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs));

Liu Bo's avatar
Liu Bo committed
267
	cur_trans->delayed_refs.href_root = RB_ROOT;
268
	cur_trans->delayed_refs.dirty_extent_root = RB_ROOT;
269
	atomic_set(&cur_trans->delayed_refs.num_entries, 0);
270 271 272 273 274 275

	/*
	 * although the tree mod log is per file system and not per transaction,
	 * the log must never go across transaction boundaries.
	 */
	smp_mb();
Julia Lawall's avatar
Julia Lawall committed
276
	if (!list_empty(&fs_info->tree_mod_seq_list))
277
		WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n");
Julia Lawall's avatar
Julia Lawall committed
278
	if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log))
279
		WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n");
280
	atomic64_set(&fs_info->tree_mod_seq, 0);
281

Josef Bacik's avatar
Josef Bacik committed
282 283 284
	spin_lock_init(&cur_trans->delayed_refs.lock);

	INIT_LIST_HEAD(&cur_trans->pending_snapshots);
285
	INIT_LIST_HEAD(&cur_trans->pending_chunks);
286
	INIT_LIST_HEAD(&cur_trans->switch_commits);
287
	INIT_LIST_HEAD(&cur_trans->dirty_bgs);
288
	INIT_LIST_HEAD(&cur_trans->io_bgs);
289
	INIT_LIST_HEAD(&cur_trans->dropped_roots);
290
	mutex_init(&cur_trans->cache_write_mutex);
291
	cur_trans->num_dirty_bgs = 0;
292
	spin_lock_init(&cur_trans->dirty_bgs_lock);
293
	INIT_LIST_HEAD(&cur_trans->deleted_bgs);
294
	spin_lock_init(&cur_trans->dropped_roots_lock);
295
	list_add_tail(&cur_trans->list, &fs_info->trans_list);
Josef Bacik's avatar
Josef Bacik committed
296
	extent_io_tree_init(&cur_trans->dirty_pages,
297
			     fs_info->btree_inode);
298 299 300
	fs_info->generation++;
	cur_trans->transid = fs_info->generation;
	fs_info->running_transaction = cur_trans;
301
	cur_trans->aborted = 0;
302
	spin_unlock(&fs_info->trans_lock);
303

Chris Mason's avatar
Chris Mason committed
304 305 306
	return 0;
}

307
/*
308 309 310 311
 * 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
312
 */
Chris Mason's avatar
Chris Mason committed
313
static int record_root_in_trans(struct btrfs_trans_handle *trans,
314 315
			       struct btrfs_root *root,
			       int force)
316
{
317 318
	struct btrfs_fs_info *fs_info = root->fs_info;

319 320
	if ((test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
	    root->last_trans < trans->transid) || force) {
321
		WARN_ON(root == fs_info->extent_root);
322 323
		WARN_ON(root->commit_root != root->node);

Chris Mason's avatar
Chris Mason committed
324
		/*
325
		 * see below for IN_TRANS_SETUP usage rules
Chris Mason's avatar
Chris Mason committed
326 327 328
		 * we have the reloc mutex held now, so there
		 * is only one writer in this function
		 */
329
		set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
Chris Mason's avatar
Chris Mason committed
330

331
		/* make sure readers find IN_TRANS_SETUP before
Chris Mason's avatar
Chris Mason committed
332 333 334 335
		 * they find our root->last_trans update
		 */
		smp_wmb();

336
		spin_lock(&fs_info->fs_roots_radix_lock);
337
		if (root->last_trans == trans->transid && !force) {
338
			spin_unlock(&fs_info->fs_roots_radix_lock);
Josef Bacik's avatar
Josef Bacik committed
339 340
			return 0;
		}
341 342 343 344
		radix_tree_tag_set(&fs_info->fs_roots_radix,
				   (unsigned long)root->root_key.objectid,
				   BTRFS_ROOT_TRANS_TAG);
		spin_unlock(&fs_info->fs_roots_radix_lock);
Chris Mason's avatar
Chris Mason committed
345 346 347 348 349 350 351 352 353 354 355 356 357
		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
358
		 * with root IN_TRANS_SETUP.  When this is 1, we're still
Chris Mason's avatar
Chris Mason committed
359 360 361 362 363 364 365
		 * 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
		 */
366
		btrfs_init_reloc_root(trans, root);
367
		smp_mb__before_atomic();
368
		clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
369 370 371
	}
	return 0;
}
372

Chris Mason's avatar
Chris Mason committed
373

374 375 376
void btrfs_add_dropped_root(struct btrfs_trans_handle *trans,
			    struct btrfs_root *root)
{
377
	struct btrfs_fs_info *fs_info = root->fs_info;
378 379 380 381 382 383 384 385
	struct btrfs_transaction *cur_trans = trans->transaction;

	/* Add ourselves to the transaction dropped list */
	spin_lock(&cur_trans->dropped_roots_lock);
	list_add_tail(&root->root_list, &cur_trans->dropped_roots);
	spin_unlock(&cur_trans->dropped_roots_lock);

	/* Make sure we don't try to update the root at commit time */
386 387
	spin_lock(&fs_info->fs_roots_radix_lock);
	radix_tree_tag_clear(&fs_info->fs_roots_radix,
388 389
			     (unsigned long)root->root_key.objectid,
			     BTRFS_ROOT_TRANS_TAG);
390
	spin_unlock(&fs_info->fs_roots_radix_lock);
391 392
}

Chris Mason's avatar
Chris Mason committed
393 394 395
int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
			       struct btrfs_root *root)
{
396 397
	struct btrfs_fs_info *fs_info = root->fs_info;

398
	if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
Chris Mason's avatar
Chris Mason committed
399 400 401
		return 0;

	/*
402
	 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
Chris Mason's avatar
Chris Mason committed
403 404 405 406
	 * and barriers
	 */
	smp_rmb();
	if (root->last_trans == trans->transid &&
407
	    !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state))
Chris Mason's avatar
Chris Mason committed
408 409
		return 0;

410
	mutex_lock(&fs_info->reloc_mutex);
411
	record_root_in_trans(trans, root, 0);
412
	mutex_unlock(&fs_info->reloc_mutex);
Chris Mason's avatar
Chris Mason committed
413 414 415 416

	return 0;
}

417 418 419
static inline int is_transaction_blocked(struct btrfs_transaction *trans)
{
	return (trans->state >= TRANS_STATE_BLOCKED &&
420 421
		trans->state < TRANS_STATE_UNBLOCKED &&
		!trans->aborted);
422 423
}

424 425 426 427
/* 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.
 */
428
static void wait_current_trans(struct btrfs_fs_info *fs_info)
Chris Mason's avatar
Chris Mason committed
429
{
430
	struct btrfs_transaction *cur_trans;
Chris Mason's avatar
Chris Mason committed
431

432 433
	spin_lock(&fs_info->trans_lock);
	cur_trans = fs_info->running_transaction;
434
	if (cur_trans && is_transaction_blocked(cur_trans)) {
435
		refcount_inc(&cur_trans->use_count);
436
		spin_unlock(&fs_info->trans_lock);
Li Zefan's avatar
Li Zefan committed
437

438
		wait_event(fs_info->transaction_wait,
439 440
			   cur_trans->state >= TRANS_STATE_UNBLOCKED ||
			   cur_trans->aborted);
441
		btrfs_put_transaction(cur_trans);
Josef Bacik's avatar
Josef Bacik committed
442
	} else {
443
		spin_unlock(&fs_info->trans_lock);
444
	}
Chris Mason's avatar
Chris Mason committed
445 446
}

447
static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type)
448
{
449
	if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
Josef Bacik's avatar
Josef Bacik committed
450 451 452 453 454 455
		return 0;

	if (type == TRANS_USERSPACE)
		return 1;

	if (type == TRANS_START &&
456
	    !atomic_read(&fs_info->open_ioctl_trans))
457
		return 1;
Josef Bacik's avatar
Josef Bacik committed
458

459 460 461
	return 0;
}

462 463
static inline bool need_reserve_reloc_root(struct btrfs_root *root)
{
464 465 466
	struct btrfs_fs_info *fs_info = root->fs_info;

	if (!fs_info->reloc_ctl ||
467
	    !test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
468 469 470 471 472 473 474
	    root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
	    root->reloc_root)
		return false;

	return true;
}

475
static struct btrfs_trans_handle *
476
start_transaction(struct btrfs_root *root, unsigned int num_items,
477 478
		  unsigned int type, enum btrfs_reserve_flush_enum flush,
		  bool enforce_qgroups)
Chris Mason's avatar
Chris Mason committed
479
{
480 481
	struct btrfs_fs_info *fs_info = root->fs_info;

482 483
	struct btrfs_trans_handle *h;
	struct btrfs_transaction *cur_trans;
484
	u64 num_bytes = 0;
485
	u64 qgroup_reserved = 0;
486 487
	bool reloc_reserved = false;
	int ret;
488

489
	/* Send isn't supposed to start transactions. */
490
	ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB);
491

492
	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
493
		return ERR_PTR(-EROFS);
494

495
	if (current->journal_info) {
496
		WARN_ON(type & TRANS_EXTWRITERS);
497
		h = current->journal_info;
498 499
		refcount_inc(&h->use_count);
		WARN_ON(refcount_read(&h->use_count) > 2);
500 501 502 503
		h->orig_rsv = h->block_rsv;
		h->block_rsv = NULL;
		goto got_it;
	}
504 505 506 507 508

	/*
	 * Do the reservation before we join the transaction so we can do all
	 * the appropriate flushing if need be.
	 */
509
	if (num_items && root != fs_info->chunk_root) {
510
		qgroup_reserved = num_items * fs_info->nodesize;
511 512
		ret = btrfs_qgroup_reserve_meta(root, qgroup_reserved,
						enforce_qgroups);
513 514
		if (ret)
			return ERR_PTR(ret);
515

516
		num_bytes = btrfs_calc_trans_metadata_size(fs_info, num_items);
517 518 519
		/*
		 * Do the reservation for the relocation root creation
		 */
520
		if (need_reserve_reloc_root(root)) {
521
			num_bytes += fs_info->nodesize;
522 523 524
			reloc_reserved = true;
		}

525
		ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
526
					  num_bytes, flush);
527
		if (ret)
528
			goto reserve_fail;
529
	}
530
again:
531
	h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS);
532 533 534 535
	if (!h) {
		ret = -ENOMEM;
		goto alloc_fail;
	}
Chris Mason's avatar
Chris Mason committed
536

537 538 539 540 541 542
	/*
	 * 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.
543 544 545
	 *
	 * 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(). 
546
	 */
547
	if (type & __TRANS_FREEZABLE)
548
		sb_start_intwrite(fs_info->sb);
549

550 551
	if (may_wait_transaction(fs_info, type))
		wait_current_trans(fs_info);
552

Josef Bacik's avatar
Josef Bacik committed
553
	do {
554
		ret = join_transaction(fs_info, type);
555
		if (ret == -EBUSY) {
556
			wait_current_trans(fs_info);
557 558 559
			if (unlikely(type == TRANS_ATTACH))
				ret = -ENOENT;
		}
Josef Bacik's avatar
Josef Bacik committed
560 561
	} while (ret == -EBUSY);

562
	if (ret < 0)
563
		goto join_fail;
564

565
	cur_trans = fs_info->running_transaction;
566 567 568

	h->transid = cur_trans->transid;
	h->transaction = cur_trans;
569
	h->root = root;
570
	refcount_set(&h->use_count, 1);
571
	h->fs_info = root->fs_info;
572

573
	h->type = type;
574
	h->can_flush_pending_bgs = true;
575
	INIT_LIST_HEAD(&h->new_bgs);
576

577
	smp_mb();
578
	if (cur_trans->state >= TRANS_STATE_BLOCKED &&
579
	    may_wait_transaction(fs_info, type)) {
580
		current->journal_info = h;
581
		btrfs_commit_transaction(h);
582 583 584
		goto again;
	}

585
	if (num_bytes) {
586
		trace_btrfs_space_reservation(fs_info, "transaction",
587
					      h->transid, num_bytes, 1);
588
		h->block_rsv = &fs_info->trans_block_rsv;
589
		h->bytes_reserved = num_bytes;
590
		h->reloc_reserved = reloc_reserved;
591
	}
Josef Bacik's avatar
Josef Bacik committed
592

593
got_it:
Josef Bacik's avatar
Josef Bacik committed
594
	btrfs_record_root_in_trans(h, root);
595 596 597

	if (!current->journal_info && type != TRANS_USERSPACE)
		current->journal_info = h;
Chris Mason's avatar
Chris Mason committed
598
	return h;
599 600

join_fail:
601
	if (type & __TRANS_FREEZABLE)
602
		sb_end_intwrite(fs_info->sb);
603 604 605
	kmem_cache_free(btrfs_trans_handle_cachep, h);
alloc_fail:
	if (num_bytes)
606
		btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv,
607 608
					num_bytes);
reserve_fail:
609
	btrfs_qgroup_free_meta(root, qgroup_reserved);
610
	return ERR_PTR(ret);
Chris Mason's avatar
Chris Mason committed
611 612
}

613
struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
614
						   unsigned int num_items)
615
{
616
	return start_transaction(root, num_items, TRANS_START,
617
				 BTRFS_RESERVE_FLUSH_ALL, true);
618
}
619

620 621 622 623 624
struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv(
					struct btrfs_root *root,
					unsigned int num_items,
					int min_factor)
{
625
	struct btrfs_fs_info *fs_info = root->fs_info;
626 627 628 629
	struct btrfs_trans_handle *trans;
	u64 num_bytes;
	int ret;

630 631 632 633 634 635 636 637
	/*
	 * We have two callers: unlink and block group removal.  The
	 * former should succeed even if we will temporarily exceed
	 * quota and the latter operates on the extent root so
	 * qgroup enforcement is ignored anyway.
	 */
	trans = start_transaction(root, num_items, TRANS_START,
				  BTRFS_RESERVE_FLUSH_ALL, false);
638 639 640 641 642 643 644
	if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
		return trans;

	trans = btrfs_start_transaction(root, 0);
	if (IS_ERR(trans))
		return trans;

645 646 647
	num_bytes = btrfs_calc_trans_metadata_size(fs_info, num_items);
	ret = btrfs_cond_migrate_bytes(fs_info, &fs_info->trans_block_rsv,
				       num_bytes, min_factor);
648
	if (ret) {
649
		btrfs_end_transaction(trans);
650 651 652
		return ERR_PTR(ret);
	}

653
	trans->block_rsv = &fs_info->trans_block_rsv;
654
	trans->bytes_reserved = num_bytes;
655
	trace_btrfs_space_reservation(fs_info, "transaction",
656
				      trans->transid, num_bytes, 1);
657 658 659

	return trans;
}
660

661
struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
662
{
663 664
	return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH,
				 true);
665 666
}

667
struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
668
{
669
	return start_transaction(root, 0, TRANS_JOIN_NOLOCK,
670
				 BTRFS_RESERVE_NO_FLUSH, true);
671 672
}

673
struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
674
{
675
	return start_transaction(root, 0, TRANS_USERSPACE,
676
				 BTRFS_RESERVE_NO_FLUSH, true);
677 678
}

679 680 681 682 683 684 685 686 687 688 689 690 691
/*
 * btrfs_attach_transaction() - catch the running transaction
 *
 * It is used when we want to commit the current the transaction, but
 * don't want to start a new one.
 *
 * Note: If this function return -ENOENT, it just means there is no
 * running transaction. But it is possible that the inactive transaction
 * is still in the memory, not fully on disk. If you hope there is no
 * inactive transaction in the fs when -ENOENT is returned, you should
 * invoke
 *     btrfs_attach_transaction_barrier()
 */
692
struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
693
{
694
	return start_transaction(root, 0, TRANS_ATTACH,
695
				 BTRFS_RESERVE_NO_FLUSH, true);
696 697
}

698
/*
699
 * btrfs_attach_transaction_barrier() - catch the running transaction
700 701 702 703 704 705 706 707 708 709
 *
 * It is similar to the above function, the differentia is this one
 * will wait for all the inactive transactions until they fully
 * complete.
 */
struct btrfs_trans_handle *
btrfs_attach_transaction_barrier(struct btrfs_root *root)
{
	struct btrfs_trans_handle *trans;

710
	trans = start_transaction(root, 0, TRANS_ATTACH,
711
				  BTRFS_RESERVE_NO_FLUSH, true);
712
	if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT)
713
		btrfs_wait_for_commit(root->fs_info, 0);
714 715 716 717

	return trans;
}

718
/* wait for a transaction commit to be fully complete */
719
static noinline void wait_for_commit(struct btrfs_transaction *commit)
720
{
721
	wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED);
722 723
}

724
int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid)
725 726
{
	struct btrfs_transaction *cur_trans = NULL, *t;
727
	int ret = 0;
728 729

	if (transid) {
730
		if (transid <= fs_info->last_trans_committed)
Josef Bacik's avatar
Josef Bacik committed
731
			goto out;
732 733

		/* find specified transaction */
734 735
		spin_lock(&fs_info->trans_lock);
		list_for_each_entry(t, &fs_info->trans_list, list) {
736 737
			if (t->transid == transid) {
				cur_trans = t;
738
				refcount_inc(&cur_trans->use_count);
739
				ret = 0;
740 741
				break;
			}
742 743
			if (t->transid > transid) {
				ret = 0;
744
				break;
745
			}
746
		}
747
		spin_unlock(&fs_info->trans_lock);
748 749 750 751 752 753

		/*
		 * The specified transaction doesn't exist, or we
		 * raced with btrfs_commit_transaction
		 */
		if (!cur_trans) {
754
			if (transid > fs_info->last_trans_committed)
755
				ret = -EINVAL;
756
			goto out;
757
		}
758 759
	} else {
		/* find newest transaction that is committing | committed */
760 761
		spin_lock(&fs_info->trans_lock);
		list_for_each_entry_reverse(t, &fs_info->trans_list,
762
					    list) {
763 764
			if (t->state >= TRANS_STATE_COMMIT_START) {
				if (t->state == TRANS_STATE_COMPLETED)
765
					break;
766
				cur_trans = t;
767
				refcount_inc(&cur_trans->use_count);
768 769 770
				break;
			}
		}
771
		spin_unlock(&fs_info->trans_lock);
772
		if (!cur_trans)
Josef Bacik's avatar
Josef Bacik committed
773
			goto out;  /* nothing committing|committed */
774 775
	}

776
	wait_for_commit(cur_trans);
777
	btrfs_put_transaction(cur_trans);
Josef Bacik's avatar
Josef Bacik committed
778
out:
779 780 781
	return ret;
}

782
void btrfs_throttle(struct btrfs_fs_info *fs_info)
Chris Mason's avatar
Chris Mason committed
783
{
784
	if (!atomic_read(&fs_info->open_ioctl_trans))
785
		wait_current_trans(fs_info);
Chris Mason's avatar
Chris Mason committed
786 787
}

788
static int should_end_transaction(struct btrfs_trans_handle *trans)
789
{
790
	struct btrfs_fs_info *fs_info = trans->fs_info;
791

792
	if (btrfs_check_space_for_delayed_refs(trans, fs_info))
793
		return 1;
794

795
	return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5);
796 797
}

798
int btrfs_should_end_transaction(struct btrfs_trans_handle *trans)
799 800
{
	struct btrfs_transaction *cur_trans = trans->transaction;
801
	struct btrfs_fs_info *fs_info = trans->fs_info;
802
	int updates;
803
	int err;
804

Josef Bacik's avatar
Josef Bacik committed
805
	smp_mb();
806 807
	if (cur_trans->state >= TRANS_STATE_BLOCKED ||
	    cur_trans->delayed_refs.flushing)
808 809 810 811
		return 1;

	updates = trans->delayed_ref_updates;
	trans->delayed_ref_updates = 0;
812
	if (updates) {
813
		err = btrfs_run_delayed_refs(trans, fs_info, updates * 2);
814 815 816
		if (err) /* Error code will also eval true */
			return err;
	}
817

818
	return should_end_transaction(trans);
819 820
}

821 822
static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans)

823
{
824 825
	struct btrfs_fs_info *fs_info = trans->fs_info;

826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841
	if (!trans->block_rsv) {
		ASSERT(!trans->bytes_reserved);
		return;
	}

	if (!trans->bytes_reserved)
		return;

	ASSERT(trans->block_rsv == &fs_info->trans_block_rsv);
	trace_btrfs_space_reservation(fs_info, "transaction",
				      trans->transid, trans->bytes_reserved, 0);
	btrfs_block_rsv_release(fs_info, trans->block_rsv,
				trans->bytes_reserved);
	trans->bytes_reserved = 0;
}

842
static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
843
				   int throttle)
Chris Mason's avatar
Chris Mason committed
844
{
845
	struct btrfs_fs_info *info = trans->fs_info;
846
	struct btrfs_transaction *cur_trans = trans->transaction;
847
	u64 transid = trans->transid;
848
	unsigned long cur = trans->delayed_ref_updates;
849
	int lock = (trans->type != TRANS_JOIN_NOLOCK);
850
	int err = 0;
Chris Mason's avatar
Chris Mason committed
851
	int must_run_delayed_refs = 0;
852

853 854
	if (refcount_read(&trans->use_count) > 1) {
		refcount_dec(&trans->use_count);
855 856 857 858
		trans->block_rsv = trans->orig_rsv;
		return 0;
	}

859
	btrfs_trans_release_metadata(trans);
860
	trans->block_rsv = NULL;
861

862
	if (!list_empty(&trans->new_bgs))
863
		btrfs_create_pending_block_groups(trans);
864

865
	trans->delayed_ref_updates = 0;
Chris Mason's avatar
Chris Mason committed
866 867
	if (!trans->sync) {
		must_run_delayed_refs =
868
			btrfs_should_throttle_delayed_refs(trans, info);
869
		cur = max_t(unsigned long, cur, 32);
Chris Mason's avatar
Chris Mason committed
870 871 872 873 874 875 876 877

		/*
		 * don't make the caller wait if they are from a NOLOCK
		 * or ATTACH transaction, it will deadlock with commit
		 */
		if (must_run_delayed_refs == 1 &&
		    (trans->type & (__TRANS_JOIN_NOLOCK | __TRANS_ATTACH)))
			must_run_delayed_refs = 2;
878
	}
879

880
	btrfs_trans_release_metadata(trans);
881
	trans->block_rsv = NULL;
882

883
	if (!list_empty(&trans->new_bgs))
884
		btrfs_create_pending_block_groups(trans);
885

886 887
	btrfs_trans_release_chunk_metadata(trans);

888
	if (lock && !atomic_read(&info->open_ioctl_trans) &&
889
	    should_end_transaction(trans) &&
890
	    READ_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) {
891 892 893 894
		spin_lock(&info->trans_lock);
		if (cur_trans->state == TRANS_STATE_RUNNING)
			cur_trans->state = TRANS_STATE_BLOCKED;
		spin_unlock(&info->trans_lock);
Josef Bacik's avatar
Josef Bacik committed
895
	}
896

897
	if (lock && READ_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) {
898
		if (throttle)
899
			return btrfs_commit_transaction(trans);
900
		else
901 902 903
			wake_up_process(info->transaction_kthread);
	}

904
	if (trans->type & __TRANS_FREEZABLE)
905
		sb_end_intwrite(info->sb);
906

907
	WARN_ON(cur_trans != info->running_transaction);
908 909
	WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
	atomic_dec(&cur_trans->num_writers);
910
	extwriter_counter_dec(cur_trans, trans->type);
911

912 913 914
	/*
	 * Make sure counter is updated before we wake up waiters.
	 */
915
	smp_mb();
Chris Mason's avatar
Chris Mason committed
916 917
	if (waitqueue_active(&cur_trans->writer_wait))
		wake_up(&cur_trans->writer_wait);
918
	btrfs_put_transaction(cur_trans);
Josef Bacik's avatar
Josef Bacik committed
919 920 921

	if (current->journal_info == trans)
		current->journal_info = NULL;
922

Yan, Zheng's avatar
Yan, Zheng committed
923
	if (throttle)
924
		btrfs_run_delayed_iputs(info);
Yan, Zheng's avatar
Yan, Zheng committed
925

926
	if (trans->aborted ||
927
	    test_bit(BTRFS_FS_STATE_ERROR, &info->fs_state)) {
928
		wake_up_process(info->transaction_kthread);
929
		err = -EIO;
930
	}
931

932
	kmem_cache_free(btrfs_trans_handle_cachep, trans);
Chris Mason's avatar
Chris Mason committed
933
	if (must_run_delayed_refs) {
934
		btrfs_async_run_delayed_refs(info, cur, transid,
Chris Mason's avatar
Chris Mason committed
935 936
					     must_run_delayed_refs == 1);
	}
937
	return err;
Chris Mason's avatar
Chris Mason committed
938 939
}

940
int btrfs_end_transaction(struct btrfs_trans_handle *trans)
941
{
942
	return __btrfs_end_transaction(trans, 0);
943 944
}

945
int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans)
946
{
947
	return __btrfs_end_transaction(trans, 1);
948 949
}

950 951 952
/*
 * 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
953
 * those extents are sent to disk but does not wait on them
954
 */
955
int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info,
956
			       struct extent_io_tree *dirty_pages, int mark)
Chris Mason's avatar
Chris Mason committed
957
{
958
	int err = 0;
959
	int werr = 0;
960
	struct address_space *mapping = fs_info->btree_inode->i_mapping;
961
	struct extent_state *cached_state = NULL;
962
	u64 start = 0;
963
	u64 end;
964

965
	atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers);
966
	while (!find_first_extent_bit(dirty_pages, start, &start, &end,
967
				      mark, &cached_state)) {
968 969 970 971
		bool wait_writeback = false;

		err = convert_extent_bit(dirty_pages, start, end,
					 EXTENT_NEED_WAIT,
972
					 mark, &cached_state);
973 974 975 976 977
		/*
		 * convert_extent_bit can return -ENOMEM, which is most of the
		 * time a temporary error. So when it happens, ignore the error
		 * and wait for writeback of this range to finish - because we
		 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
978 979 980 981 982
		 * to __btrfs_wait_marked_extents() would not know that
		 * writeback for this range started and therefore wouldn't
		 * wait for it to finish - we don't want to commit a
		 * superblock that points to btree nodes/leafs for which
		 * writeback hasn't finished yet (and without errors).
983 984 985 986 987 988 989 990 991
		 * We cleanup any entries left in the io tree when committing
		 * the transaction (through clear_btree_io_tree()).
		 */
		if (err == -ENOMEM) {
			err = 0;
			wait_writeback = true;
		}
		if (!err)
			err = filemap_fdatawrite_range(mapping, start, end);
992 993
		if (err)
			werr = err;
994 995
		else if (wait_writeback)
			werr = filemap_fdatawait_range(mapping, start, end);
996
		free_extent_state(cached_state);
997
		cached_state = NULL;
998 999
		cond_resched();
		start = end + 1;
1000
	}
1001
	atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers);
1002 1003 1004 1005 1006 1007 1008 1009 1010
	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
 */
1011 1012
static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info,
				       struct extent_io_tree *dirty_pages)
1013 1014 1015
{
	int err = 0;
	int werr = 0;
1016
	struct address_space *mapping = fs_info->btree_inode->i_mapping;
1017
	struct extent_state *cached_state = NULL;
1018 1019
	u64 start = 0;
	u64 end;
1020

1021
	while (!find_first_extent_bit(dirty_pages, start, &start, &end,
1022
				      EXTENT_NEED_WAIT, &cached_state)) {
1023 1024 1025 1026 1027 1028 1029 1030 1031
		/*
		 * Ignore -ENOMEM errors returned by clear_extent_bit().
		 * When committing the transaction, we'll remove any entries
		 * left in the io tree. For a log commit, we don't remove them
		 * after committing the log because the tree can be accessed
		 * concurrently - we do it only at transaction commit time when
		 * it's safe to do it (through clear_btree_io_tree()).
		 */
		err = clear_extent_bit(dirty_pages, start, end,
1032
				       EXTENT_NEED_WAIT, 0, 0, &cached_state);
1033 1034 1035 1036
		if (err == -ENOMEM)
			err = 0;
		if (!err)
			err = filemap_fdatawait_range(mapping, start, end);
1037 1038
		if (err)
			werr = err;
1039 1040
		free_extent_state(cached_state);
		cached_state = NULL;
1041 1042
		cond_resched();
		start = end + 1;
1043
	}
1044 1045
	if (err)
		werr = err;
1046 1047
	return werr;
}
1048

1049 1050 1051 1052 1053
int btrfs_wait_extents(struct btrfs_fs_info *fs_info,
		       struct extent_io_tree *dirty_pages)
{
	bool errors = false;
	int err;
1054

1055 1056 1057 1058 1059 1060 1061 1062
	err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
	if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags))
		errors = true;

	if (errors && !err)
		err = -EIO;
	return err;
}
1063

1064 1065 1066 1067 1068 1069
int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark)
{
	struct btrfs_fs_info *fs_info = log_root->fs_info;
	struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages;
	bool errors = false;
	int err;
1070

1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084
	ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);

	err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
	if ((mark & EXTENT_DIRTY) &&
	    test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags))
		errors = true;

	if ((mark & EXTENT_NEW) &&
	    test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags))
		errors = true;

	if (errors && !err)
		err = -EIO;
	return err;
Chris Mason's avatar
Chris Mason committed
1085 1086
}

1087
/*
1088 1089 1090 1091 1092
 * When btree blocks are allocated the corresponding extents are marked dirty.
 * This function ensures such extents are persisted on disk for transaction or
 * log commit.
 *
 * @trans: transaction whose dirty pages we'd like to write
1093
 */
1094 1095
static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
					    struct btrfs_fs_info *fs_info)
1096 1097 1098
{
	int ret;
	int ret2;
1099
	struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages;
1100
	struct blk_plug plug;
1101

1102
	blk_start_plug(&plug);
1103
	ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY);
1104
	blk_finish_plug(&plug);
1105
	ret2 = btrfs_wait_extents(fs_info, dirty_pages);
1106

1107 1108
	clear_btree_io_tree(&trans->transaction->dirty_pages);

1109 1110
	if (ret)
		return ret;
1111
	else if (ret2)
1112
		return ret2;
1113 1114
	else
		return 0;
1115 1116
}

1117 1118 1119 1120 1121 1122 1123 1124 1125 1126
/*
 * 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.
 */
1127 1128
static int update_cowonly_root(struct btrfs_trans_handle *trans,
			       struct btrfs_root *root)
Chris Mason's avatar
Chris Mason committed
1129 1130
{
	int ret;
1131
	u64 old_root_bytenr;
1132
	u64 old_root_used;
1133 1134
	struct btrfs_fs_info *fs_info = root->fs_info;
	struct btrfs_root *tree_root = fs_info->tree_root;
Chris Mason's avatar
Chris Mason committed
1135

1136
	old_root_used = btrfs_root_used(&root->root_item);
1137

1138
	while (1) {
1139
		old_root_bytenr = btrfs_root_bytenr(&root->root_item);
1140
		if (old_root_bytenr == root->node->start &&
1141
		    old_root_used == btrfs_root_used(&root->root_item))
Chris Mason's avatar
Chris Mason committed
1142
			break;
1143

1144
		btrfs_set_root_node(&root->root_item, root->node);
Chris Mason's avatar
Chris Mason committed
1145
		ret = btrfs_update_root(trans, tree_root,
1146 1147
					&root->root_key,
					&root->root_item);
1148 1149
		if (ret)
			return ret;
1150

1151
		old_root_used = btrfs_root_used(&root->root_item);
1152
	}
1153

1154 1155 1156
	return 0;
}

1157 1158
/*
 * update all the cowonly tree roots on disk
1159 1160 1161 1162
 *
 * 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.
1163
 */
1164
static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans)
1165
{
1166
	struct btrfs_fs_info *fs_info = trans->fs_info;
1167
	struct list_head *dirty_bgs = &trans->transaction->dirty_bgs;
1168
	struct list_head *io_bgs = &trans->transaction->io_bgs;
1169
	struct list_head *next;
1170
	struct extent_buffer *eb;
1171
	int ret;
1172 1173

	eb = btrfs_lock_root_node(fs_info->tree_root);
1174 1175
	ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
			      0, &eb);
1176 1177
	btrfs_tree_unlock(eb);
	free_extent_buffer(eb);
1178

1179 1180 1181
	if (ret)
		return ret;

1182
	ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1);
1183 1184
	if (ret)
		return ret;
1185

1186
	ret = btrfs_run_dev_stats(trans, fs_info);
1187 1188
	if (ret)
		return ret;
1189
	ret = btrfs_run_dev_replace(trans, fs_info);
1190 1191
	if (ret)
		return ret;
1192
	ret = btrfs_run_qgroups(trans, fs_info);
1193 1194
	if (ret)
		return ret;
1195

1196
	ret = btrfs_setup_space_cache(trans, fs_info);
1197 1198 1199
	if (ret)
		return ret;

1200
	/* run_qgroups might have added some more refs */
1201
	ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1);
1202 1203
	if (ret)
		return ret;
1204
again:
1205
	while (!list_empty(&fs_info->dirty_cowonly_roots)) {
1206
		struct btrfs_root *root;
1207 1208 1209
		next = fs_info->dirty_cowonly_roots.next;
		list_del_init(next);
		root = list_entry(next, struct btrfs_root, dirty_list);
1210
		clear_bit(BTRFS_ROOT_DIRTY, &root->state);
1211

1212 1213 1214
		if (root != fs_info->extent_root)
			list_add_tail(&root->dirty_list,
				      &trans->transaction->switch_commits);
1215 1216 1217
		ret = update_cowonly_root(trans, root);
		if (ret)
			return ret;
1218
		ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1);
1219 1220
		if (ret)
			return ret;
Chris Mason's avatar
Chris Mason committed
1221
	}
1222

1223
	while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) {
1224
		ret = btrfs_write_dirty_block_groups(trans, fs_info);
1225 1226
		if (ret)
			return ret;
1227
		ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1);
1228 1229 1230 1231 1232 1233 1234
		if (ret)
			return ret;
	}

	if (!list_empty(&fs_info->dirty_cowonly_roots))
		goto again;

1235 1236
	list_add_tail(&fs_info->extent_root->dirty_list,
		      &trans->transaction->switch_commits);
1237 1238
	btrfs_after_dev_replace_commit(fs_info);

Chris Mason's avatar
Chris Mason committed
1239 1240 1241
	return 0;
}

1242 1243 1244 1245 1246
/*
 * 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
 */
1247
void btrfs_add_dead_root(struct btrfs_root *root)
1248
{
1249 1250 1251
	struct btrfs_fs_info *fs_info = root->fs_info;

	spin_lock(&fs_info->trans_lock);
1252
	if (list_empty(&root->root_list))
1253 1254
		list_add_tail(&root->root_list, &fs_info->dead_roots);
	spin_unlock(&fs_info->trans_lock);
1255 1256
}

1257
/*
1258
 * update all the cowonly tree roots on disk
1259
 */
1260
static noinline int commit_fs_roots(struct btrfs_trans_handle *trans)
1261
{
1262
	struct btrfs_fs_info *fs_info = trans->fs_info;
1263 1264 1265
	struct btrfs_root *gang[8];
	int i;
	int ret;
1266 1267
	int err = 0;

Josef Bacik's avatar
Josef Bacik committed
1268
	spin_lock(&fs_info->fs_roots_radix_lock);
1269
	while (1) {
1270 1271
		ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
						 (void **)gang, 0,
1272 1273 1274 1275 1276
						 ARRAY_SIZE(gang),
						 BTRFS_ROOT_TRANS_TAG);
		if (ret == 0)
			break;
		for (i = 0; i < ret; i++) {
1277
			struct btrfs_root *root = gang[i];
1278 1279 1280
			radix_tree_tag_clear(&fs_info->fs_roots_radix,
					(unsigned long)root->root_key.objectid,
					BTRFS_ROOT_TRANS_TAG);
Josef Bacik's avatar
Josef Bacik committed
1281
			spin_unlock(&fs_info->fs_roots_radix_lock);
1282

1283
			btrfs_free_log(trans, root);
1284
			btrfs_update_reloc_root(trans, root);
1285
			btrfs_orphan_commit_root(trans, root);
1286

1287 1288
			btrfs_save_ino_cache(root, trans);

1289
			/* see comments in should_cow_block() */
1290
			clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1291
			smp_mb__after_atomic();
1292

1293
			if (root->commit_root != root->node) {
1294 1295
				list_add_tail(&root->dirty_list,
					&trans->transaction->switch_commits);
1296 1297 1298
				btrfs_set_root_node(&root->root_item,
						    root->node);
			}
1299 1300

			err = btrfs_update_root(trans, fs_info->tree_root,
1301 1302
						&root->root_key,
						&root->root_item);
Josef Bacik's avatar
Josef Bacik committed
1303
			spin_lock(&fs_info->fs_roots_radix_lock);
1304 1305
			if (err)
				break;
1306
			btrfs_qgroup_free_meta_all(root);
1307 1308
		}
	}
Josef Bacik's avatar
Josef Bacik committed
1309
	spin_unlock(&fs_info->fs_roots_radix_lock);
1310
	return err;
1311 1312
}

1313
/*
1314 1315
 * defrag a given btree.
 * Every leaf in the btree is read and defragged.
1316
 */
1317
int btrfs_defrag_root(struct btrfs_root *root)
1318 1319 1320
{
	struct btrfs_fs_info *info = root->fs_info;
	struct btrfs_trans_handle *trans;
1321
	int ret;
1322

1323
	if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state))
1324
		return 0;
1325

1326
	while (1) {
1327 1328 1329 1330
		trans = btrfs_start_transaction(root, 0);
		if (IS_ERR(trans))
			return PTR_ERR(trans);

1331
		ret = btrfs_defrag_leaves(trans, root);
1332

1333
		btrfs_end_transaction(trans);
1334
		btrfs_btree_balance_dirty(info);
1335 1336
		cond_resched();

1337
		if (btrfs_fs_closing(info) || ret != -EAGAIN)
1338
			break;
1339

1340 1341
		if (btrfs_defrag_cancelled(info)) {
			btrfs_debug(info, "defrag_root cancelled");
1342 1343 1344
			ret = -EAGAIN;
			break;
		}
1345
	}
1346
	clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state);
1347
	return ret;
1348 1349
}

1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370
/*
 * Do all special snapshot related qgroup dirty hack.
 *
 * Will do all needed qgroup inherit and dirty hack like switch commit
 * roots inside one transaction and write all btree into disk, to make
 * qgroup works.
 */
static int qgroup_account_snapshot(struct btrfs_trans_handle *trans,
				   struct btrfs_root *src,
				   struct btrfs_root *parent,
				   struct btrfs_qgroup_inherit *inherit,
				   u64 dst_objectid)
{
	struct btrfs_fs_info *fs_info = src->fs_info;
	int ret;

	/*
	 * Save some performance in the case that qgroups are not
	 * enabled. If this check races with the ioctl, rescan will
	 * kick in anyway.
	 */
1371
	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1372 1373 1374 1375 1376 1377 1378 1379
		return 0;

	/*
	 * We are going to commit transaction, see btrfs_commit_transaction()
	 * comment for reason locking tree_log_mutex
	 */
	mutex_lock(&fs_info->tree_log_mutex);

1380
	ret = commit_fs_roots(trans);
1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405
	if (ret)
		goto out;
	ret = btrfs_qgroup_account_extents(trans, fs_info);
	if (ret < 0)
		goto out;

	/* Now qgroup are all updated, we can inherit it to new qgroups */
	ret = btrfs_qgroup_inherit(trans, fs_info,
				   src->root_key.objectid, dst_objectid,
				   inherit);
	if (ret < 0)
		goto out;

	/*
	 * Now we do a simplified commit transaction, which will:
	 * 1) commit all subvolume and extent tree
	 *    To ensure all subvolume and extent tree have a valid
	 *    commit_root to accounting later insert_dir_item()
	 * 2) write all btree blocks onto disk
	 *    This is to make sure later btree modification will be cowed
	 *    Or commit_root can be populated and cause wrong qgroup numbers
	 * In this simplified commit, we don't really care about other trees
	 * like chunk and root tree, as they won't affect qgroup.
	 * And we don't write super to avoid half committed status.
	 */
1406
	ret = commit_cowonly_roots(trans);
1407 1408
	if (ret)
		goto out;
1409
	switch_commit_roots(trans->transaction);
1410
	ret = btrfs_write_and_wait_transaction(trans, fs_info);
1411
	if (ret)
1412
		btrfs_handle_fs_error(fs_info, ret,
1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428
			"Error while writing out transaction for qgroup");

out:
	mutex_unlock(&fs_info->tree_log_mutex);

	/*
	 * Force parent root to be updated, as we recorded it before so its
	 * last_trans == cur_transid.
	 * Or it won't be committed again onto disk after later
	 * insert_dir_item()
	 */
	if (!ret)
		record_root_in_trans(trans, parent, 1);
	return ret;
}

1429 1430
/*
 * new snapshots need to be created at a very specific time in the
1431 1432 1433 1434 1435 1436
 * transaction commit.  This does the actual creation.
 *
 * Note:
 * If the error which may affect the commitment of the current transaction
 * happens, we should return the error number. If the error which just affect
 * the creation of the pending snapshots, just return 0.
1437
 */
1438
static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1439 1440
				   struct btrfs_pending_snapshot *pending)
{
1441 1442

	struct btrfs_fs_info *fs_info = trans->fs_info;
1443
	struct btrfs_key key;
1444
	struct btrfs_root_item *new_root_item;
1445 1446
	struct btrfs_root *tree_root = fs_info->tree_root;
	struct btrfs_root *root = pending->root;
1447
	struct btrfs_root *parent_root;
1448
	struct btrfs_block_rsv *rsv;
1449
	struct inode *parent_inode;
1450 1451
	struct btrfs_path *path;
	struct btrfs_dir_item *dir_item;
1452
	struct dentry *dentry;
1453
	struct extent_buffer *tmp;
1454
	struct extent_buffer *old;
1455
	struct timespec cur_time;
1456
	int ret = 0;
1457
	u64 to_reserve = 0;
1458
	u64 index = 0;
1459
	u64 objectid;
1460
	u64 root_flags;
1461
	uuid_le new_uuid;
1462

1463 1464
	ASSERT(pending->path);
	path = pending->path;
1465

1466 1467
	ASSERT(pending->root_item);
	new_root_item = pending->root_item;
1468

1469 1470
	pending->error = btrfs_find_free_objectid(tree_root, &objectid);
	if (pending->error)
1471
		goto no_free_objectid;
1472

1473 1474 1475 1476 1477 1478
	/*
	 * Make qgroup to skip current new snapshot's qgroupid, as it is
	 * accounted by later btrfs_qgroup_inherit().
	 */
	btrfs_set_skip_qgroup(trans, objectid);

1479
	btrfs_reloc_pre_snapshot(pending, &to_reserve);
1480 1481

	if (to_reserve > 0) {
1482 1483 1484 1485 1486
		pending->error = btrfs_block_rsv_add(root,
						     &pending->block_rsv,
						     to_reserve,
						     BTRFS_RESERVE_NO_FLUSH);
		if (pending->error)
1487
			goto clear_skip_qgroup;
1488 1489
	}

1490
	key.objectid = objectid;
1491 1492
	key.offset = (u64)-1;
	key.type = BTRFS_ROOT_ITEM_KEY;
1493

1494
	rsv = trans->block_rsv;
1495
	trans->block_rsv = &pending->block_rsv;
1496
	trans->bytes_reserved = trans->block_rsv->reserved;
1497
	trace_btrfs_space_reservation(fs_info, "transaction",
1498 1499
				      trans->transid,
				      trans->bytes_reserved, 1);
1500
	dentry = pending->dentry;
1501
	parent_inode = pending->dir;
1502
	parent_root = BTRFS_I(parent_inode)->root;
1503
	record_root_in_trans(trans, parent_root, 0);
1504

1505
	cur_time = current_time(parent_inode);
1506

1507 1508 1509
	/*
	 * insert the directory item
	 */
1510
	ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index);
1511
	BUG_ON(ret); /* -ENOMEM */
1512 1513 1514

	/* check if there is a file/dir which has the same name. */
	dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1515
					 btrfs_ino(BTRFS_I(parent_inode)),
1516 1517 1518
					 dentry->d_name.name,
					 dentry->d_name.len, 0);
	if (dir_item != NULL && !IS_ERR(dir_item)) {
1519
		pending->error = -EEXIST;
1520
		goto dir_item_existed;
1521 1522
	} else if (IS_ERR(dir_item)) {
		ret = PTR_ERR(dir_item);
1523
		btrfs_abort_transaction(trans, ret);
1524
		goto fail;
1525
	}
1526
	btrfs_release_path(path);
1527

1528 1529 1530 1531 1532 1533
	/*
	 * pull in the delayed directory update
	 * and the delayed inode item
	 * otherwise we corrupt the FS during
	 * snapshot
	 */
1534
	ret = btrfs_run_delayed_items(trans);
1535
	if (ret) {	/* Transaction aborted */
1536
		btrfs_abort_transaction(trans, ret);
1537 1538
		goto fail;
	}
1539

1540
	record_root_in_trans(trans, root, 0);
1541 1542
	btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
	memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1543
	btrfs_check_and_init_root_item(new_root_item);
1544

1545 1546 1547 1548 1549 1550 1551
	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);

1552 1553 1554 1555 1556 1557
	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);
1558 1559 1560 1561 1562 1563 1564 1565
	if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
		memset(new_root_item->received_uuid, 0,
		       sizeof(new_root_item->received_uuid));
		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);
	}
1566 1567
	btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
	btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
1568 1569
	btrfs_set_root_otransid(new_root_item, trans->transid);

1570
	old = btrfs_lock_root_node(root);
1571
	ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
1572 1573 1574
	if (ret) {
		btrfs_tree_unlock(old);
		free_extent_buffer(old);
1575
		btrfs_abort_transaction(trans, ret);
1576
		goto fail;
1577
	}
1578

1579 1580
	btrfs_set_lock_blocking(old);

1581
	ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1582
	/* clean up in any case */
1583 1584
	btrfs_tree_unlock(old);
	free_extent_buffer(old);
1585
	if (ret) {
1586
		btrfs_abort_transaction(trans, ret);
1587 1588
		goto fail;
	}
1589
	/* see comments in should_cow_block() */
1590
	set_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1591 1592
	smp_wmb();

1593
	btrfs_set_root_node(new_root_item, tmp);
1594 1595 1596
	/* record when the snapshot was created in key.offset */
	key.offset = trans->transid;
	ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1597 1598
	btrfs_tree_unlock(tmp);
	free_extent_buffer(tmp);
1599
	if (ret) {
1600
		btrfs_abort_transaction(trans, ret);
1601 1602
		goto fail;
	}
1603

1604 1605 1606
	/*
	 * insert root back/forward references
	 */
1607
	ret = btrfs_add_root_ref(trans, fs_info, objectid,
1608
				 parent_root->root_key.objectid,
1609
				 btrfs_ino(BTRFS_I(parent_inode)), index,
1610
				 dentry->d_name.name, dentry->d_name.len);
1611
	if (ret) {
1612
		btrfs_abort_transaction(trans, ret);
1613 1614
		goto fail;
	}
1615

1616
	key.offset = (u64)-1;
1617
	pending->snap = btrfs_read_fs_root_no_name(fs_info, &key);
1618 1619
	if (IS_ERR(pending->snap)) {
		ret = PTR_ERR(pending->snap);
1620
		btrfs_abort_transaction(trans, ret);
1621
		goto fail;
1622
	}
1623

1624
	ret = btrfs_reloc_post_snapshot(trans, pending);
1625
	if (ret) {
1626
		btrfs_abort_transaction(trans, ret);
1627 1628
		goto fail;
	}
1629

1630
	ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1);
1631
	if (ret) {
1632
		btrfs_abort_transaction(trans, ret);
1633 1634
		goto fail;
	}
1635

1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646
	/*
	 * Do special qgroup accounting for snapshot, as we do some qgroup
	 * snapshot hack to do fast snapshot.
	 * To co-operate with that hack, we do hack again.
	 * Or snapshot will be greatly slowed down by a subtree qgroup rescan
	 */
	ret = qgroup_account_snapshot(trans, root, parent_root,
				      pending->inherit, objectid);
	if (ret < 0)
		goto fail;

1647 1648
	ret = btrfs_insert_dir_item(trans, parent_root,
				    dentry->d_name.name, dentry->d_name.len,
1649
				    BTRFS_I(parent_inode), &key,
1650 1651
				    BTRFS_FT_DIR, index);
	/* We have check then name at the beginning, so it is impossible. */
1652
	BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
1653
	if (ret) {
1654
		btrfs_abort_transaction(trans, ret);
1655 1656
		goto fail;
	}
1657

1658
	btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size +
1659
					 dentry->d_name.len * 2);
1660
	parent_inode->i_mtime = parent_inode->i_ctime =
1661
		current_time(parent_inode);
1662
	ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode);
1663
	if (ret) {
1664
		btrfs_abort_transaction(trans, ret);
1665 1666
		goto fail;
	}
1667
	ret = btrfs_uuid_tree_add(trans, fs_info, new_uuid.b,
1668 1669
				  BTRFS_UUID_KEY_SUBVOL, objectid);
	if (ret) {
1670
		btrfs_abort_transaction(trans, ret);
1671 1672 1673
		goto fail;
	}
	if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1674
		ret = btrfs_uuid_tree_add(trans, fs_info,
1675 1676 1677 1678
					  new_root_item->received_uuid,
					  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
					  objectid);
		if (ret && ret != -EEXIST) {
1679
			btrfs_abort_transaction(trans, ret);
1680 1681 1682
			goto fail;
		}
	}
1683

1684
	ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1);
1685
	if (ret) {
1686
		btrfs_abort_transaction(trans, ret);
1687 1688 1689
		goto fail;
	}

1690
fail:
1691 1692
	pending->error = ret;
dir_item_existed:
1693
	trans->block_rsv = rsv;
1694
	trans->bytes_reserved = 0;
1695 1696
clear_skip_qgroup:
	btrfs_clear_skip_qgroup(trans);
1697 1698
no_free_objectid:
	kfree(new_root_item);
1699
	pending->root_item = NULL;
1700
	btrfs_free_path(path);
1701 1702
	pending->path = NULL;

1703
	return ret;
1704 1705
}

1706 1707 1708
/*
 * create all the snapshots we've scheduled for creation
 */
1709
static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans)
1710
{
1711
	struct btrfs_pending_snapshot *pending, *next;
1712
	struct list_head *head = &trans->transaction->pending_snapshots;
1713
	int ret = 0;
1714

1715 1716
	list_for_each_entry_safe(pending, next, head, list) {
		list_del(&pending->list);
1717
		ret = create_pending_snapshot(trans, pending);
1718 1719 1720 1721
		if (ret)
			break;
	}
	return ret;
1722 1723
}

1724
static void update_super_roots(struct btrfs_fs_info *fs_info)
1725 1726 1727 1728
{
	struct btrfs_root_item *root_item;
	struct btrfs_super_block *super;

1729
	super = fs_info->super_copy;
1730

1731
	root_item = &fs_info->chunk_root->root_item;
1732 1733 1734
	super->chunk_root = root_item->bytenr;
	super->chunk_root_generation = root_item->generation;
	super->chunk_root_level = root_item->level;
1735

1736
	root_item = &fs_info->tree_root->root_item;
1737 1738 1739
	super->root = root_item->bytenr;
	super->generation = root_item->generation;
	super->root_level = root_item->level;
1740
	if (btrfs_test_opt(fs_info, SPACE_CACHE))
1741
		super->cache_generation = root_item->generation;
1742
	if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags))
1743
		super->uuid_tree_generation = root_item->generation;
1744 1745
}

1746 1747
int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
{
1748
	struct btrfs_transaction *trans;
1749
	int ret = 0;
1750

Josef Bacik's avatar
Josef Bacik committed
1751
	spin_lock(&info->trans_lock);
1752 1753 1754
	trans = info->running_transaction;
	if (trans)
		ret = (trans->state >= TRANS_STATE_COMMIT_START);
Josef Bacik's avatar
Josef Bacik committed
1755
	spin_unlock(&info->trans_lock);
1756 1757 1758
	return ret;
}

1759 1760
int btrfs_transaction_blocked(struct btrfs_fs_info *info)
{
1761
	struct btrfs_transaction *trans;
1762
	int ret = 0;
1763

Josef Bacik's avatar
Josef Bacik committed
1764
	spin_lock(&info->trans_lock);
1765 1766 1767
	trans = info->running_transaction;
	if (trans)
		ret = is_transaction_blocked(trans);
Josef Bacik's avatar
Josef Bacik committed
1768
	spin_unlock(&info->trans_lock);
1769 1770 1771
	return ret;
}

Sage Weil's avatar
Sage Weil committed
1772 1773 1774 1775
/*
 * wait for the current transaction commit to start and block subsequent
 * transaction joins
 */
1776
static void wait_current_trans_commit_start(struct btrfs_fs_info *fs_info,
Sage Weil's avatar
Sage Weil committed
1777 1778
					    struct btrfs_transaction *trans)
{
1779 1780
	wait_event(fs_info->transaction_blocked_wait,
		   trans->state >= TRANS_STATE_COMMIT_START || trans->aborted);
Sage Weil's avatar
Sage Weil committed
1781 1782 1783 1784 1785 1786
}

/*
 * wait for the current transaction to start and then become unblocked.
 * caller holds ref.
 */
1787 1788 1789
static void wait_current_trans_commit_start_and_unblock(
					struct btrfs_fs_info *fs_info,
					struct btrfs_transaction *trans)
Sage Weil's avatar
Sage Weil committed
1790
{
1791 1792
	wait_event(fs_info->transaction_wait,
		   trans->state >= TRANS_STATE_UNBLOCKED || trans->aborted);
Sage Weil's avatar
Sage Weil committed
1793 1794 1795 1796 1797 1798 1799 1800
}

/*
 * 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;
1801
	struct work_struct work;
Sage Weil's avatar
Sage Weil committed
1802 1803 1804 1805 1806
};

static void do_async_commit(struct work_struct *work)
{
	struct btrfs_async_commit *ac =
1807
		container_of(work, struct btrfs_async_commit, work);
Sage Weil's avatar
Sage Weil committed
1808

1809 1810 1811 1812
	/*
	 * We've got freeze protection passed with the transaction.
	 * Tell lockdep about it.
	 */
1813
	if (ac->newtrans->type & __TRANS_FREEZABLE)
1814
		__sb_writers_acquired(ac->newtrans->fs_info->sb, SB_FREEZE_FS);
1815

1816 1817
	current->journal_info = ac->newtrans;

1818
	btrfs_commit_transaction(ac->newtrans);
Sage Weil's avatar
Sage Weil committed
1819 1820 1821 1822 1823 1824
	kfree(ac);
}

int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
				   int wait_for_unblock)
{
1825
	struct btrfs_fs_info *fs_info = trans->fs_info;
Sage Weil's avatar
Sage Weil committed
1826 1827 1828 1829
	struct btrfs_async_commit *ac;
	struct btrfs_transaction *cur_trans;

	ac = kmalloc(sizeof(*ac), GFP_NOFS);
Tsutomu Itoh's avatar
Tsutomu Itoh committed
1830 1831
	if (!ac)
		return -ENOMEM;
Sage Weil's avatar
Sage Weil committed
1832

1833
	INIT_WORK(&ac->work, do_async_commit);
1834
	ac->newtrans = btrfs_join_transaction(trans->root);
1835 1836 1837 1838 1839
	if (IS_ERR(ac->newtrans)) {
		int err = PTR_ERR(ac->newtrans);
		kfree(ac);
		return err;
	}
Sage Weil's avatar
Sage Weil committed
1840 1841 1842

	/* take transaction reference */
	cur_trans = trans->transaction;
1843
	refcount_inc(&cur_trans->use_count);
Sage Weil's avatar
Sage Weil committed
1844

1845
	btrfs_end_transaction(trans);
1846 1847 1848 1849 1850

	/*
	 * Tell lockdep we've released the freeze rwsem, since the
	 * async commit thread will be the one to unlock it.
	 */
1851
	if (ac->newtrans->type & __TRANS_FREEZABLE)
1852
		__sb_writers_release(fs_info->sb, SB_FREEZE_FS);
1853

1854
	schedule_work(&ac->work);
Sage Weil's avatar
Sage Weil committed
1855 1856 1857

	/* wait for transaction to start and unblock */
	if (wait_for_unblock)
1858
		wait_current_trans_commit_start_and_unblock(fs_info, cur_trans);
Sage Weil's avatar
Sage Weil committed
1859
	else
1860
		wait_current_trans_commit_start(fs_info, cur_trans);
Sage Weil's avatar
Sage Weil committed
1861

1862 1863 1864
	if (current->journal_info == trans)
		current->journal_info = NULL;

1865
	btrfs_put_transaction(cur_trans);
Sage Weil's avatar
Sage Weil committed
1866 1867 1868
	return 0;
}

1869

1870
static void cleanup_transaction(struct btrfs_trans_handle *trans, int err)
1871
{
1872
	struct btrfs_fs_info *fs_info = trans->fs_info;
1873
	struct btrfs_transaction *cur_trans = trans->transaction;
1874
	DEFINE_WAIT(wait);
1875

1876
	WARN_ON(refcount_read(&trans->use_count) > 1);
1877

1878
	btrfs_abort_transaction(trans, err);
1879

1880
	spin_lock(&fs_info->trans_lock);
1881

1882 1883 1884 1885 1886 1887
	/*
	 * If the transaction is removed from the list, it means this
	 * transaction has been committed successfully, so it is impossible
	 * to call the cleanup function.
	 */
	BUG_ON(list_empty(&cur_trans->list));
1888

1889
	list_del_init(&cur_trans->list);
1890
	if (cur_trans == fs_info->running_transaction) {
1891
		cur_trans->state = TRANS_STATE_COMMIT_DOING;
1892
		spin_unlock(&fs_info->trans_lock);
1893 1894 1895
		wait_event(cur_trans->writer_wait,
			   atomic_read(&cur_trans->num_writers) == 1);

1896
		spin_lock(&fs_info->trans_lock);
1897
	}
1898
	spin_unlock(&fs_info->trans_lock);
1899

1900
	btrfs_cleanup_one_transaction(trans->transaction, fs_info);
1901

1902 1903 1904 1905
	spin_lock(&fs_info->trans_lock);
	if (cur_trans == fs_info->running_transaction)
		fs_info->running_transaction = NULL;
	spin_unlock(&fs_info->trans_lock);
1906

1907
	if (trans->type & __TRANS_FREEZABLE)
1908
		sb_end_intwrite(fs_info->sb);
1909 1910
	btrfs_put_transaction(cur_trans);
	btrfs_put_transaction(cur_trans);
1911

1912
	trace_btrfs_transaction_commit(trans->root);
1913 1914 1915

	if (current->journal_info == trans)
		current->journal_info = NULL;
1916
	btrfs_scrub_cancel(fs_info);
1917 1918 1919 1920

	kmem_cache_free(btrfs_trans_handle_cachep, trans);
}

1921 1922
static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
{
1923 1924 1925 1926 1927 1928 1929 1930 1931
	/*
	 * We use writeback_inodes_sb here because if we used
	 * btrfs_start_delalloc_roots we would deadlock with fs freeze.
	 * Currently are holding the fs freeze lock, if we do an async flush
	 * we'll do btrfs_join_transaction() and deadlock because we need to
	 * wait for the fs freeze lock.  Using the direct flushing we benefit
	 * from already being in a transaction and our join_transaction doesn't
	 * have to re-take the fs freeze lock.
	 */
1932
	if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
1933
		writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC);
1934 1935 1936 1937 1938
	return 0;
}

static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
{
1939
	if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
1940
		btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
1941 1942
}

1943
static inline void
1944
btrfs_wait_pending_ordered(struct btrfs_transaction *cur_trans)
1945
{
1946 1947
	wait_event(cur_trans->pending_wait,
		   atomic_read(&cur_trans->pending_ordered) == 0);
1948 1949
}

1950
int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
Chris Mason's avatar
Chris Mason committed
1951
{
1952
	struct btrfs_fs_info *fs_info = trans->fs_info;
1953
	struct btrfs_transaction *cur_trans = trans->transaction;
Chris Mason's avatar
Chris Mason committed
1954
	struct btrfs_transaction *prev_trans = NULL;
1955
	int ret;
Chris Mason's avatar
Chris Mason committed
1956

1957
	/* Stop the commit early if ->aborted is set */
1958
	if (unlikely(READ_ONCE(cur_trans->aborted))) {
1959
		ret = cur_trans->aborted;
1960
		btrfs_end_transaction(trans);
1961
		return ret;
1962
	}
1963

1964 1965 1966
	/* make a pass through all the delayed refs we have so far
	 * any runnings procs may add more while we are here
	 */
1967
	ret = btrfs_run_delayed_refs(trans, fs_info, 0);
1968
	if (ret) {
1969
		btrfs_end_transaction(trans);
1970 1971
		return ret;
	}
1972

1973
	btrfs_trans_release_metadata(trans);
1974 1975
	trans->block_rsv = NULL;

1976
	cur_trans = trans->transaction;
1977

1978 1979 1980 1981
	/*
	 * set the flushing flag so procs in this transaction have to
	 * start sending their work down.
	 */
1982
	cur_trans->delayed_refs.flushing = 1;
1983
	smp_wmb();
1984

1985
	if (!list_empty(&trans->new_bgs))
1986
		btrfs_create_pending_block_groups(trans);
1987

1988
	ret = btrfs_run_delayed_refs(trans, fs_info, 0);
1989
	if (ret) {
1990
		btrfs_end_transaction(trans);
1991 1992
		return ret;
	}
1993

1994
	if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) {
1995 1996 1997 1998 1999 2000 2001 2002
		int run_it = 0;

		/* this mutex is also taken before trying to set
		 * block groups readonly.  We need to make sure
		 * that nobody has set a block group readonly
		 * after a extents from that block group have been
		 * allocated for cache files.  btrfs_set_block_group_ro
		 * will wait for the transaction to commit if it
2003
		 * finds BTRFS_TRANS_DIRTY_BG_RUN set.
2004
		 *
2005 2006
		 * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure
		 * only one process starts all the block group IO.  It wouldn't
2007 2008 2009
		 * hurt to have more than one go through, but there's no
		 * real advantage to it either.
		 */
2010
		mutex_lock(&fs_info->ro_block_group_mutex);
2011 2012
		if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN,
				      &cur_trans->flags))
2013
			run_it = 1;
2014
		mutex_unlock(&fs_info->ro_block_group_mutex);
2015 2016

		if (run_it)
2017
			ret = btrfs_start_dirty_block_groups(trans);
2018 2019
	}
	if (ret) {
2020
		btrfs_end_transaction(trans);
2021 2022 2023
		return ret;
	}

2024
	spin_lock(&fs_info->trans_lock);
2025
	if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
2026
		spin_unlock(&fs_info->trans_lock);
2027
		refcount_inc(&cur_trans->use_count);
2028
		ret = btrfs_end_transaction(trans);
Chris Mason's avatar
Chris Mason committed
2029

2030
		wait_for_commit(cur_trans);
2031

2032 2033 2034
		if (unlikely(cur_trans->aborted))
			ret = cur_trans->aborted;

2035
		btrfs_put_transaction(cur_trans);
2036

2037
		return ret;
Chris Mason's avatar
Chris Mason committed
2038
	}
2039

2040
	cur_trans->state = TRANS_STATE_COMMIT_START;
2041
	wake_up(&fs_info->transaction_blocked_wait);
Sage Weil's avatar
Sage Weil committed
2042

2043
	if (cur_trans->list.prev != &fs_info->trans_list) {
Chris Mason's avatar
Chris Mason committed
2044 2045
		prev_trans = list_entry(cur_trans->list.prev,
					struct btrfs_transaction, list);
2046
		if (prev_trans->state != TRANS_STATE_COMPLETED) {
2047
			refcount_inc(&prev_trans->use_count);
2048
			spin_unlock(&fs_info->trans_lock);
Chris Mason's avatar
Chris Mason committed
2049

2050
			wait_for_commit(prev_trans);
2051
			ret = prev_trans->aborted;
Chris Mason's avatar
Chris Mason committed
2052

2053
			btrfs_put_transaction(prev_trans);
2054 2055
			if (ret)
				goto cleanup_transaction;
Josef Bacik's avatar
Josef Bacik committed
2056
		} else {
2057
			spin_unlock(&fs_info->trans_lock);
Chris Mason's avatar
Chris Mason committed
2058
		}
Josef Bacik's avatar
Josef Bacik committed
2059
	} else {
2060
		spin_unlock(&fs_info->trans_lock);
Chris Mason's avatar
Chris Mason committed
2061
	}
2062

2063 2064
	extwriter_counter_dec(cur_trans, trans->type);

2065
	ret = btrfs_start_delalloc_flush(fs_info);
2066 2067 2068
	if (ret)
		goto cleanup_transaction;

2069
	ret = btrfs_run_delayed_items(trans);
2070 2071
	if (ret)
		goto cleanup_transaction;
2072

2073 2074
	wait_event(cur_trans->writer_wait,
		   extwriter_counter_read(cur_trans) == 0);
2075

2076
	/* some pending stuffs might be added after the previous flush. */
2077
	ret = btrfs_run_delayed_items(trans);
2078 2079 2080
	if (ret)
		goto cleanup_transaction;

2081
	btrfs_wait_delalloc_flush(fs_info);
2082

2083
	btrfs_wait_pending_ordered(cur_trans);
2084

2085
	btrfs_scrub_pause(fs_info);
2086 2087 2088
	/*
	 * 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
2089
	 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
2090
	 */
2091
	spin_lock(&fs_info->trans_lock);
2092
	cur_trans->state = TRANS_STATE_COMMIT_DOING;
2093
	spin_unlock(&fs_info->trans_lock);
2094 2095 2096
	wait_event(cur_trans->writer_wait,
		   atomic_read(&cur_trans->num_writers) == 1);

2097
	/* ->aborted might be set after the previous check, so check it */
2098
	if (unlikely(READ_ONCE(cur_trans->aborted))) {
2099
		ret = cur_trans->aborted;
2100
		goto scrub_continue;
2101
	}
Chris Mason's avatar
Chris Mason committed
2102 2103 2104 2105 2106
	/*
	 * the reloc mutex makes sure that we stop
	 * the balancing code from coming in and moving
	 * extents around in the middle of the commit
	 */
2107
	mutex_lock(&fs_info->reloc_mutex);
Chris Mason's avatar
Chris Mason committed
2108

2109 2110 2111 2112 2113
	/*
	 * 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.
	 */
2114
	ret = create_pending_snapshots(trans);
2115
	if (ret) {
2116
		mutex_unlock(&fs_info->reloc_mutex);
2117
		goto scrub_continue;
2118
	}
2119

2120 2121 2122 2123 2124 2125 2126 2127 2128 2129
	/*
	 * 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.
	 */
2130
	ret = btrfs_run_delayed_items(trans);
2131
	if (ret) {
2132
		mutex_unlock(&fs_info->reloc_mutex);
2133
		goto scrub_continue;
2134
	}
2135

2136
	ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1);
2137
	if (ret) {
2138
		mutex_unlock(&fs_info->reloc_mutex);
2139
		goto scrub_continue;
2140
	}
2141

2142 2143 2144 2145
	/*
	 * make sure none of the code above managed to slip in a
	 * delayed item
	 */
2146
	btrfs_assert_delayed_root_empty(fs_info);
2147

2148
	WARN_ON(cur_trans != trans->transaction);
2149

2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162
	/* 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.
	 */
2163
	mutex_lock(&fs_info->tree_log_mutex);
2164

2165
	ret = commit_fs_roots(trans);
2166
	if (ret) {
2167 2168
		mutex_unlock(&fs_info->tree_log_mutex);
		mutex_unlock(&fs_info->reloc_mutex);
2169
		goto scrub_continue;
2170
	}
2171

2172
	/*
2173 2174
	 * Since the transaction is done, we can apply the pending changes
	 * before the next transaction.
2175
	 */
2176
	btrfs_apply_pending_changes(fs_info);
2177

2178
	/* commit_fs_roots gets rid of all the tree log roots, it is now
2179 2180
	 * safe to free the root of tree log roots
	 */
2181
	btrfs_free_log_root_tree(trans, fs_info);
2182

2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193
	/*
	 * commit_fs_roots() can call btrfs_save_ino_cache(), which generates
	 * new delayed refs. Must handle them or qgroup can be wrong.
	 */
	ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1);
	if (ret) {
		mutex_unlock(&fs_info->tree_log_mutex);
		mutex_unlock(&fs_info->reloc_mutex);
		goto scrub_continue;
	}

2194 2195 2196 2197
	/*
	 * Since fs roots are all committed, we can get a quite accurate
	 * new_roots. So let's do quota accounting.
	 */
2198
	ret = btrfs_qgroup_account_extents(trans, fs_info);
2199
	if (ret < 0) {
2200 2201
		mutex_unlock(&fs_info->tree_log_mutex);
		mutex_unlock(&fs_info->reloc_mutex);
2202 2203 2204
		goto scrub_continue;
	}

2205
	ret = commit_cowonly_roots(trans);
2206
	if (ret) {
2207 2208
		mutex_unlock(&fs_info->tree_log_mutex);
		mutex_unlock(&fs_info->reloc_mutex);
2209
		goto scrub_continue;
2210
	}
2211

2212 2213 2214 2215
	/*
	 * The tasks which save the space cache and inode cache may also
	 * update ->aborted, check it.
	 */
2216
	if (unlikely(READ_ONCE(cur_trans->aborted))) {
2217
		ret = cur_trans->aborted;
2218 2219
		mutex_unlock(&fs_info->tree_log_mutex);
		mutex_unlock(&fs_info->reloc_mutex);
2220
		goto scrub_continue;
2221 2222
	}

2223
	btrfs_prepare_extent_commit(fs_info);
2224

2225
	cur_trans = fs_info->running_transaction;
2226

2227 2228 2229
	btrfs_set_root_node(&fs_info->tree_root->root_item,
			    fs_info->tree_root->node);
	list_add_tail(&fs_info->tree_root->dirty_list,
2230
		      &cur_trans->switch_commits);
2231

2232 2233 2234
	btrfs_set_root_node(&fs_info->chunk_root->root_item,
			    fs_info->chunk_root->node);
	list_add_tail(&fs_info->chunk_root->dirty_list,
2235 2236
		      &cur_trans->switch_commits);

2237
	switch_commit_roots(cur_trans);
2238

2239
	ASSERT(list_empty(&cur_trans->dirty_bgs));
2240
	ASSERT(list_empty(&cur_trans->io_bgs));
2241
	update_super_roots(fs_info);
2242

2243 2244 2245 2246
	btrfs_set_super_log_root(fs_info->super_copy, 0);
	btrfs_set_super_log_root_level(fs_info->super_copy, 0);
	memcpy(fs_info->super_for_commit, fs_info->super_copy,
	       sizeof(*fs_info->super_copy));
Chris Mason's avatar
Chris Mason committed
2247

2248
	btrfs_update_commit_device_size(fs_info);
2249
	btrfs_update_commit_device_bytes_used(cur_trans);
2250

2251 2252
	clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
	clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
2253

2254 2255
	btrfs_trans_release_chunk_metadata(trans);

2256
	spin_lock(&fs_info->trans_lock);
2257
	cur_trans->state = TRANS_STATE_UNBLOCKED;
2258 2259 2260
	fs_info->running_transaction = NULL;
	spin_unlock(&fs_info->trans_lock);
	mutex_unlock(&fs_info->reloc_mutex);
2261

2262
	wake_up(&fs_info->transaction_wait);
2263

2264
	ret = btrfs_write_and_wait_transaction(trans, fs_info);
2265
	if (ret) {
2266 2267 2268
		btrfs_handle_fs_error(fs_info, ret,
				      "Error while writing out transaction");
		mutex_unlock(&fs_info->tree_log_mutex);
2269
		goto scrub_continue;
2270 2271
	}

2272
	ret = write_all_supers(fs_info, 0);
2273 2274 2275 2276
	/*
	 * the super is written, we can safely allow the tree-loggers
	 * to go about their business
	 */
2277
	mutex_unlock(&fs_info->tree_log_mutex);
2278 2279
	if (ret)
		goto scrub_continue;
2280

2281
	btrfs_finish_extent_commit(trans, fs_info);
2282

2283
	if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags))
2284
		btrfs_clear_space_info_full(fs_info);
Zhao Lei's avatar
Zhao Lei committed
2285

2286
	fs_info->last_trans_committed = cur_trans->transid;
2287 2288 2289 2290 2291
	/*
	 * We needn't acquire the lock here because there is no other task
	 * which can change it.
	 */
	cur_trans->state = TRANS_STATE_COMPLETED;
2292
	wake_up(&cur_trans->commit_wait);
2293

2294
	spin_lock(&fs_info->trans_lock);
2295
	list_del_init(&cur_trans->list);
2296
	spin_unlock(&fs_info->trans_lock);
Josef Bacik's avatar
Josef Bacik committed
2297

2298 2299
	btrfs_put_transaction(cur_trans);
	btrfs_put_transaction(cur_trans);
2300

2301
	if (trans->type & __TRANS_FREEZABLE)
2302
		sb_end_intwrite(fs_info->sb);
2303

2304
	trace_btrfs_transaction_commit(trans->root);
2305

2306
	btrfs_scrub_continue(fs_info);
Arne Jansen's avatar
Arne Jansen committed
2307

Josef Bacik's avatar
Josef Bacik committed
2308 2309 2310
	if (current->journal_info == trans)
		current->journal_info = NULL;

2311
	kmem_cache_free(btrfs_trans_handle_cachep, trans);
Yan, Zheng's avatar
Yan, Zheng committed
2312

2313 2314 2315 2316
	/*
	 * If fs has been frozen, we can not handle delayed iputs, otherwise
	 * it'll result in deadlock about SB_FREEZE_FS.
	 */
2317
	if (current != fs_info->transaction_kthread &&
2318 2319
	    current != fs_info->cleaner_kthread &&
	    !test_bit(BTRFS_FS_FROZEN, &fs_info->flags))
2320
		btrfs_run_delayed_iputs(fs_info);
Yan, Zheng's avatar
Yan, Zheng committed
2321

Chris Mason's avatar
Chris Mason committed
2322
	return ret;
2323

2324
scrub_continue:
2325
	btrfs_scrub_continue(fs_info);
2326
cleanup_transaction:
2327
	btrfs_trans_release_metadata(trans);
2328
	btrfs_trans_release_chunk_metadata(trans);
2329
	trans->block_rsv = NULL;
2330
	btrfs_warn(fs_info, "Skipping commit of aborted transaction.");
2331 2332
	if (current->journal_info == trans)
		current->journal_info = NULL;
2333
	cleanup_transaction(trans, ret);
2334 2335

	return ret;
Chris Mason's avatar
Chris Mason committed
2336 2337
}

2338
/*
2339 2340 2341 2342 2343 2344 2345 2346
 * return < 0 if error
 * 0 if there are no more dead_roots at the time of call
 * 1 there are more to be processed, call me again
 *
 * The return value indicates there are certainly more snapshots to delete, but
 * if there comes a new one during processing, it may return 0. We don't mind,
 * because btrfs_commit_super will poke cleaner thread and it will process it a
 * few seconds later.
2347
 */
2348
int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root)
2349
{
2350
	int ret;
2351 2352
	struct btrfs_fs_info *fs_info = root->fs_info;

Josef Bacik's avatar
Josef Bacik committed
2353
	spin_lock(&fs_info->trans_lock);
2354 2355 2356 2357 2358 2359
	if (list_empty(&fs_info->dead_roots)) {
		spin_unlock(&fs_info->trans_lock);
		return 0;
	}
	root = list_first_entry(&fs_info->dead_roots,
			struct btrfs_root, root_list);
2360
	list_del_init(&root->root_list);
Josef Bacik's avatar
Josef Bacik committed
2361
	spin_unlock(&fs_info->trans_lock);
2362

2363
	btrfs_debug(fs_info, "cleaner removing %llu", root->objectid);
2364

2365
	btrfs_kill_all_delayed_nodes(root);
2366

2367 2368 2369 2370 2371
	if (btrfs_header_backref_rev(root->node) <
			BTRFS_MIXED_BACKREF_REV)
		ret = btrfs_drop_snapshot(root, NULL, 0, 0);
	else
		ret = btrfs_drop_snapshot(root, NULL, 1, 0);
2372

2373
	return (ret < 0) ? 0 : 1;
2374
}
2375 2376 2377 2378 2379 2380

void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info)
{
	unsigned long prev;
	unsigned long bit;

2381
	prev = xchg(&fs_info->pending_changes, 0);
2382 2383 2384
	if (!prev)
		return;

2385 2386 2387 2388 2389 2390 2391 2392 2393 2394
	bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE;
	if (prev & bit)
		btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE);
	prev &= ~bit;

	bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE;
	if (prev & bit)
		btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE);
	prev &= ~bit;

2395 2396 2397 2398 2399
	bit = 1 << BTRFS_PENDING_COMMIT;
	if (prev & bit)
		btrfs_debug(fs_info, "pending commit done");
	prev &= ~bit;

2400 2401 2402 2403
	if (prev)
		btrfs_warn(fs_info,
			"unknown pending changes left 0x%lx, ignoring", prev);
}