Commit 2931cdcb authored by Darrick J. Wong's avatar Darrick J. Wong Committed by Linus Torvalds
Browse files

ocfs2: improve fsync efficiency and fix deadlock between aio_write and sync_file 



Currently, ocfs2_sync_file grabs i_mutex and forces the current journal
transaction to complete.  This isn't terribly efficient, since sync_file
really only needs to wait for the last transaction involving that inode
to complete, and this doesn't require i_mutex.

Therefore, implement the necessary bits to track the newest tid
associated with an inode, and teach sync_file to wait for that instead
of waiting for everything in the journal to commit.  Furthermore, only
issue the flush request to the drive if jbd2 hasn't already done so.

This also eliminates the deadlock between ocfs2_file_aio_write() and
ocfs2_sync_file().  aio_write takes i_mutex then calls
ocfs2_aiodio_wait() to wait for unaligned dio writes to finish.
However, if that dio completion involves calling fsync, then we can get
into trouble when some ocfs2_sync_file tries to take i_mutex.

Signed-off-by: default avatarDarrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: default avatarMark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: default avatarAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: default avatarLinus Torvalds <torvalds@linux-foundation.org>
parent a75fe48c

fs/ocfs2/alloc.c

+1 −0
Original line number Diff line number Diff line
@@ -6932,6 +6932,7 @@ int ocfs2_convert_inline_data_to_extents(struct inode *inode, 
	di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);

	spin_unlock(&oi->ip_lock);



	ocfs2_update_inode_fsync_trans(handle, inode, 1);

	ocfs2_dinode_new_extent_list(inode, di);



	ocfs2_journal_dirty(handle, di_bh);

fs/ocfs2/aops.c

+1 −0
Original line number Diff line number Diff line
@@ -2039,6 +2039,7 @@ out_write_size: 
	inode->i_mtime = inode->i_ctime = CURRENT_TIME;

	di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);

	di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);

	ocfs2_update_inode_fsync_trans(handle, inode, 1);

	ocfs2_journal_dirty(handle, wc->w_di_bh);



	ocfs2_commit_trans(osb, handle);

fs/ocfs2/dir.c

+4 −0
Original line number Diff line number Diff line
@@ -2957,6 +2957,7 @@ static int ocfs2_expand_inline_dir(struct inode *dir, struct buffer_head *di_bh, 
		ocfs2_init_dir_trailer(dir, dirdata_bh, i);

	}



	ocfs2_update_inode_fsync_trans(handle, dir, 1);

	ocfs2_journal_dirty(handle, dirdata_bh);



	if (ocfs2_supports_indexed_dirs(osb) && !dx_inline) {
@@ -3338,6 +3339,7 @@ do_extend: 
	} else {

		de->rec_len = cpu_to_le16(sb->s_blocksize);

	}

	ocfs2_update_inode_fsync_trans(handle, dir, 1);

	ocfs2_journal_dirty(handle, new_bh);



	dir_i_size += dir->i_sb->s_blocksize;
@@ -3896,6 +3898,7 @@ out_commit: 
		dquot_free_space_nodirty(dir,

				ocfs2_clusters_to_bytes(dir->i_sb, 1));



	ocfs2_update_inode_fsync_trans(handle, dir, 1);

	ocfs2_commit_trans(osb, handle);



out:
@@ -4134,6 +4137,7 @@ static int ocfs2_expand_inline_dx_root(struct inode *dir, 
		mlog_errno(ret);

	did_quota = 0;



	ocfs2_update_inode_fsync_trans(handle, dir, 1);

	ocfs2_journal_dirty(handle, dx_root_bh);



out_commit:

fs/ocfs2/file.c

+15 −21
Original line number Diff line number Diff line
@@ -175,9 +175,13 @@ static int ocfs2_sync_file(struct file *file, loff_t start, loff_t end, 
			   int datasync)

{

	int err = 0;

	journal_t *journal;

	struct inode *inode = file->f_mapping->host;

	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);

	struct ocfs2_inode_info *oi = OCFS2_I(inode);

	journal_t *journal = osb->journal->j_journal;

	int ret;

	tid_t commit_tid;

	bool needs_barrier = false;



	trace_ocfs2_sync_file(inode, file, file->f_path.dentry,

			      OCFS2_I(inode)->ip_blkno,
@@ -192,29 +196,19 @@ static int ocfs2_sync_file(struct file *file, loff_t start, loff_t end, 
	if (err)

		return err;



	/*

	 * Probably don't need the i_mutex at all in here, just putting it here

	 * to be consistent with how fsync used to be called, someone more

	 * familiar with the fs could possibly remove it.

	 */

	mutex_lock(&inode->i_mutex);

	if (datasync && !(inode->i_state & I_DIRTY_DATASYNC)) {

		/*

		 * We still have to flush drive's caches to get data to the

		 * platter

		 */

		if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)

			blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);

		goto bail;

	commit_tid = datasync ? oi->i_datasync_tid : oi->i_sync_tid;

	if (journal->j_flags & JBD2_BARRIER &&

	    !jbd2_trans_will_send_data_barrier(journal, commit_tid))

		needs_barrier = true;

	err = jbd2_complete_transaction(journal, commit_tid);

	if (needs_barrier) {

		ret = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);

		if (!err)

			err = ret;

	}



	journal = osb->journal->j_journal;

	err = jbd2_journal_force_commit(journal);



bail:

	if (err)

		mlog_errno(err);

	mutex_unlock(&inode->i_mutex);



	return (err < 0) ? -EIO : 0;

}
@@ -650,7 +644,7 @@ restarted_transaction: 
			mlog_errno(status);

		goto leave;

	}



	ocfs2_update_inode_fsync_trans(handle, inode, 1);

	ocfs2_journal_dirty(handle, bh);



	spin_lock(&OCFS2_I(inode)->ip_lock);

fs/ocfs2/inode.c

+28 −0
Original line number Diff line number Diff line
@@ -130,6 +130,7 @@ struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, unsigned flags, 
	struct inode *inode = NULL;

	struct super_block *sb = osb->sb;

	struct ocfs2_find_inode_args args;

	journal_t *journal = OCFS2_SB(sb)->journal->j_journal;



	trace_ocfs2_iget_begin((unsigned long long)blkno, flags,

			       sysfile_type);
@@ -169,6 +170,32 @@ struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, unsigned flags, 
		goto bail;

	}



	/*

	 * Set transaction id's of transactions that have to be committed

	 * to finish f[data]sync. We set them to currently running transaction

	 * as we cannot be sure that the inode or some of its metadata isn't

	 * part of the transaction - the inode could have been reclaimed and

	 * now it is reread from disk.

	 */

	if (journal) {

		transaction_t *transaction;

		tid_t tid;

		struct ocfs2_inode_info *oi = OCFS2_I(inode);



		read_lock(&journal->j_state_lock);

		if (journal->j_running_transaction)

			transaction = journal->j_running_transaction;

		else

			transaction = journal->j_committing_transaction;

		if (transaction)

			tid = transaction->t_tid;

		else

			tid = journal->j_commit_sequence;

		read_unlock(&journal->j_state_lock);

		oi->i_sync_tid = tid;

		oi->i_datasync_tid = tid;

	}



bail:

	if (!IS_ERR(inode)) {

		trace_ocfs2_iget_end(inode,
@@ -1260,6 +1287,7 @@ int ocfs2_mark_inode_dirty(handle_t *handle, 
	fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);



	ocfs2_journal_dirty(handle, bh);

	ocfs2_update_inode_fsync_trans(handle, inode, 1);

leave:

	return status;

}

CRA Git | Maintained and supported by SUSTech CRA and CCSE