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, 28 insertions, 0 deletions

diff --git a/fs/ocfs2/inode.c b/fs/ocfs2/inode.c
index f29a90fde619..28ab8a9e88a1 100644
--- a/fs/ocfs2/inode.c
+++ b/fs/ocfs2/inode.c
@@ -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;
 }