Replace UnixEndPoint with UnixDomainSocketEndPoint

[mguinness]

While looking into https://stackoverflow.com/questions/71742985 I noticed that OpenUnixSocketAsync() uses the UnixEndPoint class that was added with PR #119 back in 2016.

MySqlConnector/src/MySqlConnector/Core/ServerSession.cs

Lines 1139 to 1186 in e5cc811

	private async Task<bool> OpenUnixSocketAsync(ConnectionSettings cs, IOBehavior ioBehavior, CancellationToken cancellationToken)
	{
	m_logArguments[1] = cs.UnixSocket;
	Log.Trace("Session{0} connecting to UNIX Socket '{1}'", m_logArguments);
	var socket = new Socket(AddressFamily.Unix, SocketType.Stream, ProtocolType.IP);
	var unixEp = new UnixEndPoint(cs.UnixSocket!);
	try
	{
	using (cancellationToken.Register(() => socket.Dispose()))
	{
	try
	{
	if (ioBehavior == IOBehavior.Asynchronous)
	{
	await Task.Factory.FromAsync(socket.BeginConnect, socket.EndConnect, unixEp, null).ConfigureAwait(false);
	}
	else
	{
	socket.Connect(unixEp);
	}
	}
	catch (ObjectDisposedException) when (cancellationToken.IsCancellationRequested)
	{
	Log.Info("Session{0} connect timeout expired connecting to UNIX Socket '{1}'", m_logArguments);
	throw new MySqlException(MySqlErrorCode.UnableToConnectToHost, "Connect Timeout expired.");
	}
	}
	}
	catch (SocketException)
	{
	socket.Dispose();
	}
	if (socket.Connected)
	{
	m_socket = socket;
	m_stream = new NetworkStream(socket);
	m_activityTags.Add(ActivitySourceHelper.NetTransportTagName, ActivitySourceHelper.NetTransportUnixValue);
	m_activityTags.Add(ActivitySourceHelper.NetPeerNameTagName, cs.UnixSocket);
	lock (m_lock)
	m_state = State.Connected;
	return true;
	}
	return false;
	}

Since .NET Standard 2.1 the UnixDomainSocketEndPoint class is available which should provide the same functionality as UnixEndPoint that was copied from Mono.

It probably doesn't offer any benefits other than code maintainability, but if the code implementing Unix Domain Sockets can easily switch to the new class that would be worthwhile. The Postgres driver also made this switch in npgsql/npgsql#3314,

The question as to why the uid option is needed in the connection string for the unix socket authentication plugin to work is another issue which I asked the OP to submit.

[bgrainger]

The question as to why the uid option is needed in the connection string for the unix socket authentication plugin to work is another issue which I asked the OP to submit.

MySqlConnector always sends a handshake response packet (when connecting via any protocol, including Unix sockets), which contains a "username" field. In the SO poster's case, it will contain the empty string. The MariaDB documentation makes it sound as if the client doesn't need to provide a user name:

The unix_socket authentication plugin works by calling the getsockopt system call with the SO_PEERCRED socket option, which allows it to retrieve the uid of the process that is connected to the socket. It is then able to get the user name associated with that uid. Once it has the user name, it will authenticate the connecting user as the MariaDB account that has the same user name.

Maybe client libraries are supposed to implicitly set the user name to whoami when connecting via Unix sockets, and this is just undocumented expected behaviour?

[mguinness]

That's really not clear to me either, some examples show the command line client with the --user option and others don't.

The MySQL plugin documentation shows how to map a DB user to a different OS user which necessitates providing the user.

Maybe the answer is when connection protocol is set to unix and user is empty then just replace it with Environment.UserName?

Would you be able to do a packet capture on mysql client without user to see what is sent in the handshake response packet?

[bgrainger]

https://dev.mysql.com/doc/refman/8.0/en/user-names.html

On Unix, most MySQL clients by default try to log in using the current Unix user name as the MySQL user name, but that is for convenience only.

https://dev.mysql.com/doc/c-api/8.0/en/mysql-real-connect.html

The user argument contains the user's MySQL login ID. If user is NULL or the empty string "", the current user is assumed. Under Unix, this is the current login name.

Sounds like the official mysql client (and probably any others that use the mysql_real_connect function in their implementation) would default to Environment.UserName if not specified.

Maybe the answer is when connection protocol is set to unix and user is empty then just replace it with Environment.UserName?

I filed a suggestion for this improvement: #1161.

[bgrainger]

Implemented in 2.1.9.