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: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

1 files changed, 15 insertions, 21 deletions

diff --git a/fs/ocfs2/file.c b/fs/ocfs2/file.c
index 1673438789f..bd94d26b0b2 100644
--- a/fs/ocfs2/file.c
+++ b/fs/ocfs2/file.c
@@ -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);