Chapter 46. Connectivity
TCP/IP -- IP Addresses and Ports
TCP/IP networking is a part of the Open Systems Interconnection (OSI) Model. Much like you can string together lots of little single-purpose Unix tools to do complex tasks, the OSI Model is made up of specific single-purpose layers that work together. Each layer builds on the layers below. Layers 1 and 2 are concerned with hardware; physical standards such as required voltages and low-level protocols like Ethernet reside there. Layers 3 and 4 are networking layers, which this article introduces. Layers 5 through 7 are application layers, where networking interfaces such as BSD sockets and applications such as web browsers, telnet clients, and diagnostic tools live.
For most Unixes, the fundamentals of networking (once you get past the network device drivers) are the Layer 3 Internet Protocol (IP) and a Layer 4 protocol on top of it, either the Transport Control Protocol (TCP), the User Datagram Protocol (UDP), or the IP Control Message Protocol (ICMP). These four protocols are so commonly treated as one unit that you'll often see them referred to together as TCP/IP.
46.1.1. Internet Protocol (IP)
IP's job is to get small chunks of data, called packets, from one machine to another. It is a "best effort" protocol; that is, it makes its best effort to deliver each packet to the right host, and if it can't, it simply drops the packet on the floor. It may seem like losing bits of your data would be a bad thing, but it turns out that this feature is part of what allows the Internet to route traffic around problems; higher-level protocols and applications notice that packets are being dropped and resend them, sometimes through better routes.
IP identifies machines through IP addresses. Every machine that wants to communicate with another machine via TCP/IP must have a unique IP address, unless it's using Network Address Translation (NAT) (Section 46.1). When you dial up your ISP with a modem, your ISP assigns you a dynamic IP address, good for that modem session. When you have a dedicated broadband connection, often your ISP will assign you a small block of static IP addresses to use as you like. Each ISP is in turn assigned large blocks of IP addresses for them to dole out to their users, and traffic on the Internet travels from ISP to ISP based on the addresses they hold.
The current standard version of IP is Version 4 (IPv4), which uses 32-bit addresses. With the explosion of the Internet, addresses are being used up at quite an impressive rate; remember that normally every single machine connected to the Internet needs its own IP address. Version 6 (IPv6) is, at the time of this writing, a proposed standard that uses 128-bit addresses. For the purposes of this book, we'll gloss over the differences, since they mostly don't matter at this level. Our examples will use IPv4 addresses, since that's what you're most likely to be dealing with for a little while yet.
46.1.2. Layer 4 Protocols: TCP, UDP, and ICMP
TCP's job is to provide ordered and guaranteed delivery. Ordered delivery means that the application at the other end of the TCP connection reads data in the same order as it was sent. Guaranteed delivery means that TCP keeps track of which packets arrived at the other end and resends packets that were dropped. Together, these two characteristics provide a network communication mechanism that acts very much like a Unix pipe from an application's point of view; you simply write bytes in one end and they come out the other. Many common network applications sit on top of TCP and use these services, including telnet, HTTP servers and web browsers, SSH (Section 46.6), and email (Section 46.8).
UDP provides application access to the basic delivery mechanism of IP and adds port addressing (see below). Some applications don't need guaranteed delivery and want the lower overhead of UDP, or want the low-level control of network error recovery UDP can provide, or need to be able to do certain kinds of broadcast. Services like DNS (Section 46.9) and DHCP (Section 46.10) use UDP rather than TCP, as do many Internet games.
Both TCP and UDP provide addressing of their own above and beyond IP addresses; these addresses are called ports . Generally, simply getting a packet to a machine isn't quite enough; if you want two programs to communicate, they need a rendezvous point; there can be a lot of programs running on the destination machine, and TCP and UDP need to know to which program packets should be delivered. An IP address and a port provide the means for two programs to hook up and start talking. Every communication needs a port on each machine; one side "listens" on a well-known port and the other side "connects" to that port, generating a random port of its own.
Ports are represented by an integer number. Ports below 1024 are usually accessible only by programs running as root and are thus mostly reserved for system services and the like. /etc/services (Section 46.2) lists most of the well-known ports used by the main system services.
Finally, ICMP provides diagnostic and traffic control messages. ICMP is primarily used by applications such as ping and traceroute (Section 46.4) to diagnose problems, check network status, and the like. Routers can also use ICMP to control traffic.
Copyright © 2003 O'Reilly & Associates. All rights reserved.