I've noticed how many people found other pages of this blog trying to find more information about Unix sockets, and so I thought it's about time we shed some light on this seeming mysterious, but really simple concept.
What is a Unix socket?
A Unix socket (the technically correct name for it is Unix domain socket, UDS) is a way of inter-process communication (IPC) in Unix. Like almost everything in Unix, a socket is a file. It's a special file, to be precise. Unix processes which want to communicate between each other use special set of functions to access the special file of a Unix socket, and easily exchange data in both directions.
In very simple terms, a Unix socket is nothing but a byte steam – a data transfer between processes running locally or on networked Unix systems.
Examples of Unix sockets
Most well know examples of Unix sockets are probably those servicing the graphics system on your Unix box: X11 server socket and some optional sockets of programs managing it, like GDM (Gnome Desktop Manager).
bash-2.05b$ ls -al /tmp/.gdm_socket srw-rw-rw- 1 root root 0 Oct 25 17:49 /tmp/.gdm_socket bash-2.05b$ file /tmp/.gdm_socket /tmp/.gdm_socket: socket
The message logging daemon, syslogd, uses /dev/log on most systems to accept new messages to be logged in log files. Here's how this file looks in RedHat:
bash-2.05b$ ls -al /dev/log srw-rw-rw- 1 root root 0 Oct 25 17:49 /dev/log bash-2.05b$ file /dev/log /dev/log: socket
Types of sockets in Unix
Here is another thing which can be quite confusing about sockets in Unix – the classification.
On the highest level, there are two types of sockets: Unix domain sockets for IPC (AF_UNIX) and Unix network sockets using Internet family of protocols (AF_INET), most commonly referred to as Unix Internet sockets.
These two types essentially define a set of communication protocols supported by a socket. Unix domain sockets are for IPC (interprocess communication) only, which means they can only be used for processes running locally on the same Unix system and communicating to each other. The Internet sockets support protocols which allow you to connect processes between different Unix systems: IP protocol for the network communication, and TCP/UDP protocols for the transport.
Connection-oriented and connectionless sockets
Both Unix domain sockets and Unix Internet sockets can use reliable (guaranteed) and unreliable (best effort) approaches for establishing and maintaining connections. With Unix domain sockets it doesn't really matter, as all the communication is local to your Unix system, but with Internet sockets it's a different story.
Connection-oriented sockets are called stream sockets (SOCK_STREAM), and are the reliable and guaranteed way to communicate. For Internet sockets, the TCP protocol is used to ensure that your data is confirmed to be delivered to the destination point, and your data packets will be received in the same order they were sent out.
The reason such sockets are called stream sockets is because your Unix will create and maintain a stream – an active connection between the source and the destination points, using TCP to manage this connection. All packets are automatically acknowledged and synchronization is maintained to ensure the ordered way of sending and receiving packets.
Connectionless sockets are called datagram sockets (SOCK_DGRAM) and they use UDP for their communication, which means that the packets of data which you send may or may not be received on the other end. Because there is no control over the transfer (no acknowledgments of the delivery, no ordered packet arrangements), there is no need to maintain such a connection. Hence the name – connectionless. You simply send a message out, and it's then a problem of a higher level protocol to ensure the data is transferred successfully.
That's it – should be enough for a brief Unix sockets introduction. Do ask your questions in the comments to this post, and I'll be sure to answer them in future posts on Unix sockets – there's still plenty to show and explain!