The Linux Programming Interface

1198 Chapter 59

Internet domain datagram sockets are implemented on top of UDP. UDP sockets

are similar to their UNIX domain counterparts, but note the following differences:

z UNIX domain datagram sockets are reliable, but UDP sockets are not—data-

grams may be lost, duplicated, or arrive in a different order from that in which

they were sent.

z Sending on a UNIX domain datagram socket will block if the queue of data for

the receiving socket is full. By contrast, with UDP, if the incoming datagram

would overflow the receiver’s queue, then the datagram is silently dropped.

59.2 Network Byte Order

IP addresses and port numbers are integer values. One problem we encounter

when passing these values across a network is that different hardware architectures

store the bytes of a multibyte integer in different orders. As shown in Figure 59-1,

architectures that store integers with the most significant byte first (i.e., at the lowest

memory address) are termed big endian; those that store the least significant byte

first are termed little endian. (The terms derive from Jonathan Swift’s 1726 satirical

novel Gulliver’s Travels, in which the terms refer to opposing political factions who

open their boiled eggs at opposite ends.) The most notable example of a little-

endian architecture is x86. (Digital’s VAX architecture was another historically

important example, since BSD was widely used on that machine.) Most other archi-

tectures are big endian. A few hardware architectures are switchable between the

two formats. The byte ordering used on a particular machine is called the host byte order.

Figure 59-1: Big-endian and little-endian byte order for 2-byte and 4-byte integers

Since port numbers and IP addresses must be transmitted between, and under-

stood by, all hosts on a network, a standard ordering must be used. This ordering is

called network byte order, and happens to be big endian.

Later in this chapter, we look at various functions that convert hostnames (e.g.,

http://www.kernel.org) and service names (e.g., http) into the corresponding numeric

forms. These functions generally return integers in network byte order, and these

integers can be copied directly into the relevant fields of a socket address structure.

1

(MSB)

0

(LSB)

address

N

address

N + 1

Big-endian

byte order

Little-endian

byte order

0

(LSB)

1

(MSB)

address

N

address

N + 1

3

(MSB)

2

address

N

address

N + 1

10

(LSB)

address

N + 2

address

N + 3

0

(LSB)

1

address

N

address

N + 1

23

(MSB)

address

N + 2

address

N + 3

2-byte integer 4-byte integer

MSB = Most Significant Byte, LSB = Least Significant Byte