Commit 5039eddc authored by Josef Bacik's avatar Josef Bacik Committed by Chris Mason
Browse files

Btrfs: make fsync latency less sucky 



Looking into some performance related issues with large amounts of metadata
revealed that we can have some pretty huge swings in fsync() performance.  If we
have a lot of delayed refs backed up (as you will tend to do with lots of
metadata) fsync() will wander off and try to run some of those delayed refs
which can result in reading from disk and such.  Since the actual act of fsync()
doesn't create any delayed refs there is no need to make it throttle on delayed
ref stuff, that will be handled by other people.  With this patch we get much
smoother fsync performance with large amounts of metadata.  Thanks,

Signed-off-by: default avatarJosef Bacik <jbacik@fb.com>
Signed-off-by: default avatarChris Mason <clm@fb.com>
parent 63541927

fs/btrfs/file.c

+12 −0
Original line number Diff line number Diff line
@@ -1928,12 +1928,24 @@ int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync) 
	if (file->private_data)

		btrfs_ioctl_trans_end(file);



	/*

	 * We use start here because we will need to wait on the IO to complete

	 * in btrfs_sync_log, which could require joining a transaction (for

	 * example checking cross references in the nocow path).  If we use join

	 * here we could get into a situation where we're waiting on IO to

	 * happen that is blocked on a transaction trying to commit.  With start

	 * we inc the extwriter counter, so we wait for all extwriters to exit

	 * before we start blocking join'ers.  This comment is to keep somebody

	 * from thinking they are super smart and changing this to

	 * btrfs_join_transaction *cough*Josef*cough*.

	 */

	trans = btrfs_start_transaction(root, 0);

	if (IS_ERR(trans)) {

		ret = PTR_ERR(trans);

		mutex_unlock(&inode->i_mutex);

		goto out;

	}

	trans->sync = true;



	ret = btrfs_log_dentry_safe(trans, root, dentry);

	if (ret < 0) {

fs/btrfs/transaction.c

+2 −1
Original line number Diff line number Diff line
@@ -474,6 +474,7 @@ again: 
	h->type = type;

	h->allocating_chunk = false;

	h->reloc_reserved = false;

	h->sync = false;

	INIT_LIST_HEAD(&h->qgroup_ref_list);

	INIT_LIST_HEAD(&h->new_bgs);
@@ -713,7 +714,7 @@ static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, 
		btrfs_create_pending_block_groups(trans, root);



	trans->delayed_ref_updates = 0;

	if (btrfs_should_throttle_delayed_refs(trans, root)) {

	if (!trans->sync && btrfs_should_throttle_delayed_refs(trans, root)) {

		cur = max_t(unsigned long, cur, 1);

		trans->delayed_ref_updates = 0;

		btrfs_run_delayed_refs(trans, root, cur);

fs/btrfs/transaction.h

+1 −0
Original line number Diff line number Diff line
@@ -93,6 +93,7 @@ struct btrfs_trans_handle { 
	short adding_csums;

	bool allocating_chunk;

	bool reloc_reserved;

	bool sync;

	unsigned int type;

	/*

	 * this root is only needed to validate that the root passed to

CRA Git | Maintained and supported by SUSTech CRA and CCSE