Go to the previous, next section.
This chapter describes the GNU facilities for interprocess communication using sockets.
A socket is a generalized interprocess communication channel.
Like a pipe, a socket is represented as a file descriptor. But,
unlike pipes, sockets support communication between unrelated
processes, and even between processes running on different machines
that communicate over a network. Sockets are the primary means of
communicating with other machines; telnet, rlogin,
ftp, talk, and the other familiar network programs use
sockets.
Not all operating systems support sockets. In the GNU library, the header file `sys/socket.h' exists regardless of the operating system, and the socket functions always exist, but if the system does not really support sockets, these functions always fail.
Incomplete: We do not currently document the facilities for broadcast messages or for configuring Internet interfaces.
When you create a socket, you must specify the style of communication you want to use and the type of protocol that should implement it. The communication style of a socket defines the user-level semantics of sending and receiving data on the socket. Choosing a communication style specifies the answers to questions such as these:
Designing a program to use unreliable communication styles usually involves taking precautions to detect lost or misordered packets and to retransmit data as needed.
You must also choose a namespace for naming the socket. A socket name ("address") is meaningful only in the context of a particular namespace. In fact, even the data type to use for a socket name may depend on the namespace. Namespaces are also called "domains", but we avoid that word as it can be confused with other usage of the same term. Each namespace has a symbolic name that starts with `PF_'. A corresponding symbolic name starting with `AF_' designates the address format for that namespace.
Finally you must next choose the protocol to carry out the communication. The protocol determines what low-level mechanism is used to transmit and receive data. Each protocol is valid for a particular namespace and communication style; a namespace is sometimes called a protocol family because of this, which is why the namespace names start with `PF_'.
The rules of a protocol apply to the data passing between two programs, perhaps on different computers; most of these rules are handled by the operating system, and you need not know about them. What you do need to know about protocols is this:
The GNU library includes support for several different kinds of sockets, each with different characteristics. This section describes the supported socket types. The symbolic constants listed here are defined in `sys/socket.h'.
The SOCK_STREAM style is like a pipe (see section Pipes and FIFOs);
it operates over a connection with a particular remote socket, and
transmits data reliably as a stream of bytes.
Use of this style is covered in detail in section Using Sockets with Connections.
The SOCK_DGRAM style is used for sending
individually-addressed packets, unreliably.
It is the diametrical opposite of SOCK_STREAM.
Each time you write data to a socket of this kind, that data becomes
one packet. Since SOCK_DGRAM sockets do not have connections,
you must specify the recipient address with each packet.
The only guarantee that the system makes about your requests to transmit data is that it will try its best to deliver each packet you send. It may succeed with the sixth packet after failing with the fourth and fifth packets; the seventh packet may arrive before the sixth, and may arrive a second time after the sixth.
The typical use for SOCK_DGRAM is in situations where it is
acceptible to simply resend a packet if no response is seen in a
reasonable amount of time.
See section Datagram Socket Operations, for detailed information about how to use datagram sockets.
This style provides access to low-level network protocols and interfaces. Ordinary user programs usually have no need to use this style.
The name of a socket is normally called an address. The functions and symbols for dealing with socket addresses were named inconsistently, sometimes using the term "name" and sometimes using "address". You can regard these terms as synonymous where sockets are concerned.
A socket newly created with the socket function has no
address. Other processes can find it for communication only if you
give it an address. We call this binding the address to the
socket, and the way to do it is with the bind function.
You need be concerned with the address of a socket if other processes are to find it and start communicating with it. You can specify an address for other sockets, but this is usually pointless; the first time you send data from a socket, or use it to initiate a connection, the system assigns an address automatically if you have not specified one.
Occasionally a client needs to specify an address because the server
discriminates based on addresses; for example, the rsh and rlogin
protocols look at the client's socket address and don't bypass password
checking unless it is less than IPPORT_RESERVED (see section Internet Ports).
The details of socket addresses vary depending on what namespace you are using. See section The File Namespace, or section The Internet Namespace, for specific information.
Regardless of the namespace, you use the same functions bind and
getsockname to set and examine a socket's address. These
functions use a phony data type, struct sockaddr *, to accept the
address. In practice, the address lives in a structure of some other
data type appropriate to the address format you are using, but you cast
its address to struct sockaddr * when you pass it to
bind.
The functions bind and getsockname use the generic data
type struct sockaddr * to represent a pointer to a socket
address. You can't use this data type effectively to interpret an
address or construct one; for that, you must use the proper data type
for the socket's namespace.
Thus, the usual practice is to construct an address in the proper
namespace-specific type, then cast a pointer to struct sockaddr *
when you call bind or getsockname.
The one piece of information that you can get from the struct
sockaddr data type is the address format designator which tells
you which data type to use to understand the address fully.
The symbols in this section are defined in the header file `sys/socket.h'.
The struct sockaddr type itself has the following members:
short int sa_family
char sa_data[14]
sa_data is essentially arbitrary.
Each address format has a symbolic name which starts with `AF_'. Each of them corresponds to a `PF_' symbol which designates the corresponding namespace. Here is a list of address format names:
AF_FILE
PF_FILE is the name of that namespace.) See section Details of File Namespace, for information about this address format.
AF_UNIX
AF_FILE, for compatibility.
(PF_UNIX is likewise a synonym for PF_FILE.)
AF_INET
PF_INET is the name of that namespace.)
See section Internet Socket Address Format.
AF_UNSPEC
The corresponding namespace designator symbol PF_UNSPEC exists
for completeness, but there is no reason to use it in a program.
`sys/socket.h' defines symbols starting with `AF_' for many different kinds of networks, all or most of which are not actually implemented. We will document those that really work, as we receive information about how to use them.
Use the bind function to assign an address to a socket. The
prototype for bind is in the header file `sys/socket.h'.
For examples of use, see section The File Namespace, or see section Internet Socket Example.
Function: int bind (int socket, struct sockaddr *addr, size_t length)
The bind function assigns an address to the socket
socket. The addr and length arguments specify the
address; the detailed format of the address depends on the namespace.
The first part of the address is always the format designator, which
specifies a namespace, and says that the address is in the format for
that namespace.
The return value is 0 on success and -1 on failure. The
following errno error conditions are defined for this function:
EBADF
ENOTSOCK
EADDRNOTAVAIL
EADDRINUSE
EINVAL
EACCESS
IPPORT_RESERVED minus one; see
section Internet Ports.)
Additional conditions may be possible depending on the particular namespace of the socket.
Use the function getsockname to examine the address of an
Internet socket. The prototype for this function is in the header file
`sys/socket.h'.
Function: int getsockname (int socket, struct sockaddr *addr, size_t *length_ptr)
The getsockname function returns information about the
address of the socket socket in the locations specified by the
addr and length_ptr arguments. Note that the
length_ptr is a pointer; you should initialize it to be the
allocation size of addr, and on return it contains the actual
size of the address data.
The format of the address data depends on the socket namespace. The
length of the information is usually fixed for a given namespace, so
normally you can know exactly how much space is needed and can provide
that much. The usual practice is to allocate a place for the value
using the proper data type for the socket's namespace, then cast its
address to struct sockaddr * to pass it to getsockname.
The return value is 0 on success and -1 on error. The
following errno error conditions are defined for this function:
EBADF
ENOTSOCK
ENOBUFS
You can't read the address of a socket in the file namespace. This is consistent with the rest of the system; in general, there's no way to find a file's name from a descriptor for that file.
This section describes the details of the file namespace, whose
symbolic name (required when you create a socket) is PF_FILE.
In the file namespace, socket addresses are file names. You can specify any file name you want as the address of the socket, but you must have write permission on the directory containing it. In order to connect to a socket, you must have read permission for it. It's common to put these files in the `/tmp' directory.
One peculiarity of the file namespace is that the name is only used when opening the connection; once that is over with, the address is not meaningful and may not exist.
Another peculiarity is that you cannot connect to such a socket from another machine--not even if the other machine shares the file system which contains the name of the socket. You can see the socket in a directory listing, but connecting to it never succeeds. Some programs take advantage of this, such as by asking the client to send its own process ID, and using the process IDs to distinguish between clients. However, we recommend you not use this method in protocols you design, as we might someday permit connections from other machines that mount the same file systems. Instead, send each new client an identifying number if you want it to have one.
After you close a socket in the file namespace, you should delete the
file name from the file system. Use unlink or remove to
do this; see section Deleting Files.
The file namespace supports just one protocol for any communication
style; it is protocol number 0.
To create a socket in the file namespace, use the constant
PF_FILE as the namespace argument to socket or
socketpair. This constant is defined in `sys/socket.h'.
This designates the file namespace, in which socket addresses are file names, and its associated family of protocols.
This is a synonym for PF_FILE, for compatibility's sake.
The structure for specifying socket names in the file namespace is defined in the header file `sys/un.h':
This structure is used to specify file namespace socket addresses. It has the following members:
short int sun_family
AF_FILE to designate the file
namespace. See section Socket Addresses.
char sun_path[108]
Incomplete: Why is 108 a magic number? RMS suggests making
this a zero-length array and tweaking the example following to use
alloca to allocate an appropriate amount of storage based on
the length of the filename.
You should compute the length parameter for a socket address in
the file namespace as the sum of the size of the sun_family
component and the string length (not the allocation size!) of
the file name string.
Here is an example showing how to create and name a socket in the file namespace.
#include <stddef.h>
#include <stdio.h>
#include <errno.h>
#include <stdlib.h>
#include <sys/socket.h>
#include <sys/un.h>
int
make_named_socket (const char *filename)
{
struct sockaddr_un name;
int sock;
size_t size;
/* Create the socket. */
sock = socket (PF_UNIX, SOCK_DGRAM, 0);
if (sock < 0)
{
perror ("socket");
exit (EXIT_FAILURE);
}
/* Bind a name to the socket. */
name.sun_family = AF_FILE;
strcpy (name.sun_path, filename);
/* The size of the address is
the offset of the start of the filename,
plus its length,
plus one for the terminating null byte. */
size = (offsetof (struct sockaddr_un, sun_path)
+ strlen (name.sun_path) + 1);
if (bind (sock, (struct sockaddr *) &name, size) < 0)
{
perror ("bind");
exit (EXIT_FAILURE);
}
return sock;
}
This section describes the details the protocols and socket naming conventions used in the Internet namespace.
To create a socket in the Internet namespace, use the symbolic name
PF_INET of this namespace as the namespace argument to
socket or socketpair. This macro is defined in
`sys/socket.h'.
This designates the Internet namespace and associated family of protocols.
A socket address for the Internet namespace includes the following components:
You must ensure that the address and port number are represented in a canonical format called network byte order. See section Byte Order Conversion, for information about this.
In the Internet namespace, a socket address consists of a host address and a port on that host. In addition, the protocol you choose serves effectively as a part of the address because local port numbers are meaningful only within a particular protocol.
The data type for representing socket addresses in the Internet namespace is defined in the header file `netinet/in.h'.
This is the data type used to represent socket addresses in the Internet namespace. It has the following members:
short int sin_family
AF_INET in this member.
See section Socket Addresses.
struct in_addr sin_addr
unsigned short int sin_port
When you call bind or getsockname, you should specify
sizeof (struct sockaddr_in) as the length parameter if
you are using an Internet namespace socket address.
Each computer on the Internet has one or more Internet addresses, numbers which identify that computer among all those on the Internet. Users typically write numeric host addresses as sequences of four numbers, separated by periods, as in `128.52.46.32'.
Each computer also has one or more host names, which are strings of words separated by periods, as in `churchy.gnu.ai.mit.edu'.
Programs that let the user specify a host typically accept both numeric addresses and host names. But the program needs a numeric address to open a connection; to use a host name, you must convert it to the numeric address it stands for.
An Internet host address is a number containing four bytes of data. These are divided into two parts, a network number and a local network address number within that network. The network number consists of the first one, two or three bytes; the rest of the bytes are the local address.
Network numbers are registered with the Network Information Center (NIC), and are divided into three classes--A, B, and C. The local network address numbers of individual machines are registered with the administrator of the particular network.
Class A networks have single-byte numbers in the range 0 to 127. There are only a small number of Class A networks, but they can each support a very large number of hosts. Medium-sized Class B networks have two-byte network numbers, with the first byte in the range 128 to 191. Class C networks are the smallest; they have three-byte network numbers, with the first byte in the range 192-255. Thus, the first 1, 2, or 3 bytes of an Internet address specifies a network. The remaining bytes of the Internet address specify the address within that network.
The Class A network 0 is reserved for broadcast to all networks. In addition, the host number 0 within each network is reserved for broadcast to all hosts in that network.
The Class A network 127 is reserved for loopback; you can always use the Internet address `127.0.0.1' to refer to the host machine.
Since a single machine can be a member of multiple networks, it can have multiple Internet host addresses. However, there is never supposed to be more than one machine with the same host address.
There are four forms of the standard numbers-and-dots notation for Internet addresses:
a.b.c.d
a.b.c
a.b.
a.b
a
Within each part of the address, the usual C conventions for specifying the radix apply. In other words, a leading `0x' or `0X' implies hexadecimal radix; a leading `0' implies octal; and otherwise decimal radix is assumed.
Internet host addresses are represented in some contexts as integers
(type unsigned long int). In other contexts, the integer is
packaged inside a structure of type struct in_addr. It would
be better if the usage were made consistent, but it is not hard to extract
the integer from the structure or put the integer into a structure.
The following basic definitions for Internet addresses appear in the header file `netinet/in.h':
This data type is used in certain contexts to contain an Internet host
address. It has just one field, named s_addr, which records the
host address number as an unsigned long int.
Macro: unsigned long int INADDR_ANY
You can use this constant to stand for "the address of this machine,"
instead of finding its actual address. This special constant saves you
the trouble of looking up the address of your own machine. Also, if
your machine has multiple network addresses on different networks (which
is not unusual), using INADDR_ANY permits the system to choose
whichever address makes communication most efficient.
These additional functions for manipulating Internet addresses are declared in `arpa/inet.h'. They represent Internet addresses in network byte order; they represent network numbers and local-address-within-network numbers in host byte order. See section Byte Order Conversion, for an explanation of network and host byte order.
Function: unsigned long int inet_addr (const char *name)
This function converts the Internet host address name
from the standard numbers-and-dots notation into binary data.
If the input is not valid, inet_addr returns -1.
Function: unsigned long int inet_network (const char *name)
This function extracts the network number from the address name,
given in the standard numbers-and-dots notation.
If the input is not valid, inet_network returns -1.
Function: char * inet_ntoa (struct in_addr addr)
This function converts the Internet host address addr to a string in the standard numbers-and-dots notation. The return value is a pointer into a statically-allocated buffer. Subsequent calls will overwrite the same buffer, so you should copy the string if you need to save it.
Function: struct in_addr inet_makeaddr (int net, int local)
This function makes an Internet host address by combining the network number net with the local-address-within-network number local.
Function: int inet_lnaof (struct in_addr addr)
This function returns the local-address-within-network part of the Internet host address addr.
Function: int inet_netof (struct in_addr addr)
This function returns the network number part of the Internet host address addr.
Besides the standard numbers-and-dots notation for Internet addresses, you can also refer to a host by a symbolic name. The advantage of a symbolic name is that it is usually easier to remember. For example, the machine with Internet address `128.52.46.32' is also known as `churchy.gnu.ai.mit.edu'; and other machines in the `gnu.ai.mit.edu' domain can refer to it simply as `churchy'.
Internally, the system uses a database to keep track of the mapping between host names and host numbers. This database is usually either the file `/etc/hosts' or an equivalent provided by a name server. The functions and other symbols for accessing this database are declared in `netdb.h'. They are BSD features, defined unconditionally if you include `netdb.h'.
This data type is used to represent an entry in the hosts database. It has the following members:
char *h_name
char **h_aliases
int h_addrtype
AF_INET. In principle other kinds of addresses could be
represented in the data base as well as Internet addresses; if this were
done, you might find a value in this field other than AF_INET.
See section Socket Addresses.
int h_length
char **h_addr_list
char *h_addr
h_addr_list[0]; in other words, it is the
first host address.
As far as the host database is concerned, each address is just a block
of memory h_length bytes long. But in other contexts there is an
implicit assumption that you can convert this to a struct in_addr or
an unsigned long int. Host addresses in a struct hostent
structure are always given in network byte order; see section Byte Order Conversion.
You can use gethostbyname or gethostbyaddr to search the
hosts database for information about a particular host. The information
is returned in a statically-allocated structure; you must copy the
information if you need to save it across calls.
Function: struct hostent * gethostbyname (const char *name)
The gethostbyname function returns information about the host
named name. If the lookup fails, it returns a null pointer.
Function: struct hostent * gethostbyaddr (const char *addr, int length, int format)
The gethostbyaddr function returns information about the host
with Internet address addr. The length argument is the
size (in bytes) of the address at addr. format specifies
the address format; for an Internet address, specify a value of
AF_INET.
If the lookup fails, gethostbyaddr returns a null pointer.
If the name lookup by gethostbyname or gethostbyaddr
fails, you can find out the reason by looking at the value of the
variable h_errno. (It would be cleaner design for these
functions to set errno, but use of h_errno is compatible
with other systems.) Before using h_errno, you must declare it
like this:
extern int h_errno;
Here are the error codes that you may find in h_errno:
HOST_NOT_FOUND
TRY_AGAIN
NO_RECOVERY
NO_ADDRESS
You can also scan the entire hosts database one entry at a time using
sethostent, gethostent, and endhostent. Be careful
in using these functions, because they are not reentrant.
Function: void sethostent (int stayopen)
This function opens the hosts database to begin scanning it. You can
then call gethostent to read the entries.
If the stayopen argument is nonzero, this sets a flag so that
subsequent calls to gethostbyname or gethostbyaddr will
not close the database (as they usually would). This makes for more
efficiency if you call those functions several times, by avoiding
reopening the database for each call.
Function: struct hostent * gethostent ()
This function returns the next entry in the hosts database. It returns a null pointer if there are no more entries.
This function closes the hosts database.
A socket address in the Internet namespace consists of a machine's Internet address plus a port number which distinguishes the sockets on a given machine (for a given protocol). Port numbers range from 0 to 65,535.
Port numbers less than IPPORT_RESERVED are reserved for standard
servers, such as finger and telnet. There is a database
that keeps track of these, and you can use the getservbyname
function to map a service name onto a port number; see section The Services Database.
If you write a server that is not one of the standard ones defined in
the database, you must choose a port number for it. Use a number
greater than IPPORT_USERRESERVED; such numbers are reserved for
servers and won't ever be generated automatically by the system.
Avoiding conflicts with servers being run by other users is up to you.
When you use a socket without specifying its address, the system
generates a port number for it. This number is between
IPPORT_RESERVED and IPPORT_USERRESERVED.
On the Internet, it is actually legitimate to have two different
sockets with the same port number, as long as they never both try to
communicate with the same socket address (host address plus port
number). You shouldn't duplicate a port number except in special
circumstances where a higher-level protocol requires it. Normally,
the system won't let you do it; bind normally insists on
distinct port numbers. To reuse a port number, you must set the
socket option SO_REUSEADDR. See section Socket-Level Options.
These macros are defined in the header file `netinet/in.h'.
Port numbers less than IPPORT_RESERVED are reserved for
superuser use.
Macro: int IPPORT_USERRESERVED
Port numbers greater than or equal to IPPORT_USERRESERVED are
reserved for explicit use; they will never be allocated automatically.
The database that keeps track of "well-known" services is usually either the file `/etc/services' or an equivalent from a name server. You can use these utilities, declared in `netdb.h', to access the services database.
This data type holds information about entries from the services database. It has the following members:
char *s_name
char **s_aliases
int s_port
char *s_proto
To get information about a particular service, use the
getservbyname or getservbyport functions. The information
is returned in a statically-allocated structure; you must copy the
information if you need to save it across calls.
Function: struct servent * getservbyname (const char *name, const char *proto)
The getservbyname function returns information about the
service named name using protocol proto. If it can't find
such a service, it returns a null pointer.
This function is useful for servers as well as for clients; servers use it to determine which port they should listen on (see section Listening for Connections).
Function: struct servent * getservbyport (int port, const char *proto)
The getservbyport function returns information about the
service at port port using protocol proto. If it can't
find such a service, it returns a null pointer.
You can also scan the services database using setservent,
getservent, and endservent. Be careful in using these
functions, because they are not reentrant.
Function: void setservent (int stayopen)
This function opens the services database to begin scanning it.
If the stayopen argument is nonzero, this sets a flag so that
subsequent calls to getservbyname or getservbyport will
not close the database (as they usually would). This makes for more
efficiency if you call those functions several times, by avoiding
reopening the database for each call.
Function: struct servent * getservent (void)
This function returns the next entry in the services database. If there are no more entries, it returns a null pointer.
Function: void endservent (void)
This function closes the services database.
Different kinds of computers use different conventions for the ordering of bytes within a word. Some computers put the most significant byte within a word first (this is called "big-endian" order), and others put it last ("little-endian" order).
So that machines with different byte order conventions can communicate, the Internet protocols specify a canonical byte order convention for data transmitted over the network. This is known as the network byte order.
When establishing an Internet socket connection, you must make sure that
the data in the sin_port and sin_addr members of the
sockaddr_in structure are represented in the network byte order.
If you are encoding integer data in the messages sent through the
socket, you should convert this to network byte order too. If you don't
do this, your program may fail when running on or talking to other kinds
of machines.
If you use getservbyname and gethostbyname or
inet_addr to get the port number and host address, the values are
already in the network byte order, and you can copy them directly into
the sockaddr_in structure.
Otherwise, you have to convert the values explicitly. Use
htons and ntohs to convert values for the sin_port
member. Use htonl and ntohl to convert values for the
sin_addr member. (Remember, struct in_addr is equivalent
to unsigned long int.) These functions are declared in
`netinet/in.h'.
Function: unsigned short int htons (unsigned short int hostshort)
This function converts the short integer hostshort from
host byte order to network byte order.
Function: unsigned short int ntohs (unsigned short int netshort)
This function converts the short integer netshort from
network byte order to host byte order.
Function: unsigned long int htonl (unsigned long int hostlong)
This function converts the long integer hostlong from
host byte order to network byte order.
Function: unsigned long int ntohl (unsigned long int netlong)
This function converts the long integer netlong from
network byte order to host byte order.
The communications protocol used with a socket controls low-level details of how data is exchanged. For example, the protocol implements things like checksums to detect errors in transmissions, and routing instructions for messages. Normal user programs have little reason to mess with these details directly.
The default communications protocol for the Internet namespace depends on the communication style. For stream communication, the default is TCP ("transmission control protocol"). For datagram communication, the default is UDP ("user datagram protocol"). For reliable datagram communication, the default is RDP ("reliable datagram protocol"). You should nearly always use the default.
Internet protocols are generally specified by a name instead of a
number. The network protocols that a host knows about are stored in a
database. This is usually either derived from the file
`/etc/protocols', or it may be an equivalent provided by a name
server. You look up the protocol number associated with a named
protocol in the database using the getprotobyname function.
Here are detailed descriptions of the utilities for accessing the protocols database. These are declared in `netdb.h'.
This data type is used to represent entries in the network protocols database. It has the following members:
char *p_name
char **p_aliases
int p_proto
socket.
You can use getprotobyname and getprotobynumber to search
the protocols database for a specific protocol. The information is
returned in a statically-allocated structure; you must copy the
information if you need to save it across calls.
Function: struct protoent * getprotobyname (const char *name)
The getprotobyname function returns information about the
network protocol named name. If there is no such protocol, it
returns a null pointer.
Function: struct protoent * getprotobynumber (int protocol)
The getprotobynumber function returns information about the
network protocol with number protocol. If there is no such
protocol, it returns a null pointer.
You can also scan the whole protocols database one protocol at a time by
using setprotoent, getprotoent, and endprotoent.
Be careful in using these functions, because they are not reentrant.
Function: void setprotoent (int stayopen)
This function opens the protocols database to begin scanning it.
If the stayopen argument is nonzero, this sets a flag so that
subsequent calls to getprotobyname or getprotobynumber will
not close the database (as they usually would). This makes for more
efficiency if you call those functions several times, by avoiding
reopening the database for each call.
Function: struct protoent * getprotoent (void)
This function returns the next entry in the protocols database. It returns a null pointer if there are no more entries.
Function: void endprotoent (void)
This function closes the protocols database.
Here is an example showing how to create and name a socket in the
Internet namespace. The newly created socket exists on the machine that
the program is running on. Rather than finding and using the machine's
Internet address, this example specifies INADDR_ANY as the host
address; the system replaces that with the machine's actual address.
#include <stdio.h>
#include <stdlib.h>
#include <sys/socket.h>
#include <netinet/in.h>
int
make_socket (unsigned short int port)
{
int sock;
struct sockaddr_in name;
/* Create the socket. */
sock = socket (PF_INET, SOCK_STREAM, 0);
if (sock < 0)
{
perror ("socket");
exit (EXIT_FAILURE);
}
/* Give the socket a name. */
name.sin_family = AF_INET;
name.sin_port = htons (port);
name.sin_addr.s_addr = htonl (INADDR_ANY);
if (bind (sock, (struct sockaddr *) &name, sizeof (name)) < 0)
{
perror ("bind");
exit (EXIT_FAILURE);
}
return sock;
}
Here is another example, showing how you can fill in a sockaddr_in
structure, given a host name string and a port number:
#include <stdio.h>
#include <stdlib.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include <netdb.h>
void
init_sockaddr (struct sockaddr_in *name,
const char *hostname, unsigned short int port)
{
struct hostent *hostinfo;
name->sin_family = AF_INET;
name->sin_port = htons (port);
hostinfo = gethostbyname (serverhost);
if (hostinfo == NULL)
{
fprintf (stderr, "Unknown host %s.\n", hostname);
exit (EXIT_FAILURE);
}
name->sin_addr = *(struct in_addr *) hostinfo->h_addr;
}
Certain other namespaces and associated protocol families are supported
but not documented yet because they are not often used. PF_NS
refers to the Xerox Network Software protocols. PF_ISO stands
for Open Systems Interconnect. PF_CCITT refers to protocols from
CCITT. `socket.h' defines these symbols and others naming protocols
not actually implemented.
PF_IMPLINK is used for communicating between hosts and Internet
Message Processors. For information on this, and on PF_ROUTE, an
occasionally-used local area routing protocol, see the GNU Hurd Manual
(to appear in the future).
This section describes the actual library functions for opening and closing sockets. The same functions work for all namespaces and connection styles.
The primitive for creating a socket is the socket function,
declared in `sys/socket.h'.
Function: int socket (int namespace, int style, int protocol)
This function creates a socket and specifies communication style
style, which should be one of the socket styles listed in
section Communication Styles. The namespace argument specifies
the namespace; it must be PF_FILE (see section The File Namespace) or
PF_INET (see section The Internet Namespace). protocol
designates the specific protocol (see section Socket Concepts); zero is
usually right for protocol.
The return value from socket is the file descriptor for the new
socket, or -1 in case of error. The following errno error
conditions are defined for this function:
EPROTONOSUPPORT
EMFILE
ENFILE
EACCESS
ENOBUFS
The file descriptor returned by the socket function supports both
read and write operations. But, like pipes, sockets do not support file
positioning operations.
For examples of how to call the socket function,
see section The File Namespace, or section Internet Socket Example.
When you are finished using a socket, you can simply close its
file descriptor with close; see section Opening and Closing Files.
If there is still data waiting to be transmitted over the connection,
normally close tries to complete this transmission. You
can control this behavior using the SO_LINGER socket option to
specify a timeout period; see section Socket Options.
You can also shut down only reception or only transmission on a
connection by calling shutdown, which is declared in
`sys/socket.h'.
Function: int shutdown (int socket, int how)
The shutdown function shuts down the connection of socket
socket. The argument how specifies what action to
perform:
0
1
2
The return value is 0 on success and -1 on failure. The
following errno error conditions are defined for this function:
EBADF
ENOTSOCK
ENOTCONN
A socket pair consists of a pair of connected (but unnamed)
sockets. It is very similar to a pipe and is used in much the same
way. Socket pairs are created with the socketpair function,
declared in `sys/socket.h'. A socket pair is much like a pipe; the
main difference is that the socket pair is bidirectional, whereas the
pipe has one input-only end and one output-only end (see section Pipes and FIFOs).
Function: int socketpair (int namespace, int style, int protocol, int filedes[2])
This function creates a socket pair, returning the file descriptors in
filedes[0] and filedes[1]. The socket pair
is a full-duplex communications channel, so that both reading and writing
may be performed at either end.
The namespace, style, and protocol arguments are
interpreted as for the socket function. style should be
one of the communication styles listed in section Communication Styles.
The namespace argument specifies the namespace, which must be
AF_FILE (see section The File Namespace); protocol specifies the
communications protocol, but zero is the only meaningful value.
If style specifies a connectionless communication style, then the two sockets you get are not connected, strictly speaking, but each of them knows the other as the default destination address, so they can send packets to each other.
The socketpair function returns 0 on success and -1
on failure. The following errno error conditions are defined
for this function:
EMFILE
EAFNOSUPPORT
EPROTONOSUPPORT
EOPNOTSUPP
The most common communication styles involve making a connection to a particular other socket, and then exchanging data with that socket over and over. Making a connection is asymmetric; one side (the client) acts to request a connection, while the other side (the server) makes a socket and waits for the connection request.
In making a connection, the client makes a connection while the server
waits for and accepts the connection. Here we discuss what the client
program must do, using the connect function.
Function: int connect (int socket, struct sockaddr *addr, size_t length)
The connect function initiates a connection from the socket
with file descriptor socket to the socket whose address is
specified by the addr and length arguments. (This socket
is typically on another machine, and it must be already set up as a
server.) See section Socket Addresses, for information about how these
arguments are interpreted.
Normally, connect waits until the server responds to the request
before it returns. You can set nonblocking mode on the socket
socket to make connect return immediately without waiting
for the response. See section File Status Flags, for information about
nonblocking mode.
The normal return value from connect is 0. If an error
occurs, connect returns -1. The following errno
error conditions are defined for this function:
EBADF
ENOTSOCK
EADDRNOTAVAIL
EAFNOSUPPORT
EISCONN
ETIMEDOUT
ECONNREFUSED
ENETUNREACH
EADDRINUSE
EINPROGRESS
EALREADY
Now let us consider what the server process must do to accept
connections on a socket. This involves the use of the listen
function to enable connection requests on the socket, and later using
the accept function (see section Accepting Connections) to act on a
request. The listen function is not allowed for sockets using
connectionless communication styles.
You can write a network server that does not even start running until a
connection to it is requested. See section inetd Servers.
In the Internet namespace, there are no special protection mechanisms for controlling access to connect to a port; any process on any machine can make a connection to your server. If you want to restrict access to your server, make it examine the addresses associated with connection requests or implement some other handshaking or identification protocol.
In the File namespace, the ordinary file protection bits control who has access to connect to the socket.
Function: int listen (int socket, unsigned int n)
The listen function enables the socket socket to
accept connections, thus making it a server socket.
The argument n specifies the length of the queue for pending connections.
The listen function returns 0 on success and -1
on failure. The following errno error conditions are defined
for this function:
EBADF
ENOTSOCK
EOPNOTSUPP
When a server receives a connection request, it can complete the
connection by accepting the request. Use the function accept
to do this.
A socket that has been established as a server can accept connection
requests from multiple clients. The server's original socket
does not become part of the connection; instead, accept
makes a new socket which participates in the connection.
accept returns the descriptor for this socket. The server's
original socket remains available for listening for further connection
requests.
The number of pending connection requests on a server socket is finite.
If connection requests arrive from clients faster than the server can
act upon them, the queue can fill up and additional requests are refused
with a ECONNREFUSED error. You can specify the maximum length of
this queue as an argument to the listen function, although the
system may also impose its own internal limit on the length of this
queue.
Function: int accept (int socket, struct sockaddr *addr, size_t *length_ptr)
This function is used to accept a connection request on the server socket socket.
The accept function waits if there are no connections pending,
unless the socket socket has nonblocking mode set. (You can use
select to wait for a pending connection, with a nonblocking
socket.) See section File Status Flags, for information about nonblocking
mode.
The addr and length_ptr arguments are used to return information about the name of the client socket that initiated the connection. See section Socket Addresses, for information about the format of the information.
Accepting a connection does not make socket part of the
connection. Instead, it creates a new socket which becomes
connected. The normal return value of accept is the file
descriptor for the new socket.
After accept, the original socket socket remains open and
unconnected, and continues listening until you close it. You can
accept further connections with socket by calling accept
again.
If an error occurs, accept returns -1. The following
errno error conditions are defined for this function:
EBADF
ENOTSOCK
EOPNOTSUPP
EWOULDBLOCK
The accept function is not allowed for sockets using
connectionless communication styles.
Function: int getpeername (int socket, struct sockaddr *addr, size_t *length_ptr)
The getpeername function returns the address of the socket that
socket is connected to; it stores the address in the memory space
specified by addr and length_ptr. It stores the length of
the address in *length_ptr.
See section Socket Addresses, for information about the format of the
address. In some operating systems, getpeername works only for
sockets in the Internet domain.
The return value is 0 on success and -1 on error. The
following errno error conditions are defined for this function:
EBADF
ENOTSOCK
ENOTCONN
ENOBUFS
Once a socket has been connected to a peer, you can use the ordinary
read and write operations (see section Input and Output Primitives) to
transfer data. A socket is a two-way communications channel, so read
and write operations can be performed at either end.
There are also some I/O modes that are specific to socket operations.
In order to specify these modes, you must use the recv and
send functions instead of the more generic read and
write functions. The recv and send functions take
an additional argument which you can use to specify various flags to
control the special I/O modes. For example, you can specify the
MSG_OOB flag to read or write out-of-band data, the
MSG_PEEK flag to peek at input, or the MSG_DONTROUTE flag
to control inclusion of routing information on output.
The send function is declared in the header file
`sys/socket.h'. If your flags argument is zero, you can just
as well use write instead of send; see section Input and Output Primitives. If the socket was connected but the connection has broken,
you get a SIGPIPE signal for any use of send or
write (see section Miscellaneous Signals).
Function: int send (int socket, void *buffer, size_t size, int flags)
The send function is like write, but with the additional
flags flags. The possible values of flags are described
in section Socket Data Options.
This function returns the number of bytes transmitted, or -1 on
failure. If the socket is nonblocking, then send (like
write) can return after sending just part of the data.
See section File Status Flags, for information about nonblocking mode.
Note, however, that a successful return value merely indicates that the message has been sent without error, not necessarily that it has been received without error.
The following errno error conditions are defined for this function:
EBADF
EINTR
ENOTSOCK
EMSGSIZE
EWOULDBLOCK
send blocks until the operation can be
completed.)
ENOBUFS
ENOTCONN
EPIPE
send generates a SIGPIPE signal first; if that
signal is ignored or blocked, or if its handler returns, then
send fails with EPIPE.
The recv function is declared in the header file
`sys/socket.h'. If your flags argument is zero, you can
just as well use read instead of recv; see section Input and Output Primitives.
Function: int recv (int socket, void *buffer, size_t size, int flags)
The recv function is like read, but with the additional
flags flags. The possible values of flags are described
In section Socket Data Options.
If nonblocking mode is set for socket, and no data is available to
be read, recv fails immediately rather than waiting. See section File Status Flags, for information about nonblocking mode.
This function returns the number of bytes received, or -1 on failure.
The following errno error conditions are defined for this function:
EBADF
ENOTSOCK
EWOULDBLOCK
recv blocks until there is input
available to be read.)
EINTR
ENOTCONN
The flags argument to send and recv is a bit
mask. You can bitwise-OR the values of the following macros together
to obtain a value for this argument. All are defined in the header
file `sys/socket.h'.
Send or receive out-of-band data. See section Out-of-Band Data.
Look at the data but don't remove it from the input queue. This is
only meaningful with input functions such as recv, not with
send.
Don't include routing information in the message. This is only meaningful with output operations, and is usually only of interest for diagnostic or routing programs. We don't try to explain it here.
Here is an example client program that makes a connection for a byte stream socket in the Internet namespace. It doesn't do anything particularly interesting once it has connected to the server; it just sends a text string to the server and exits.
#include <stdio.h>
#include <errno.h>
#include <stdlib.h>
#include <unistd.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include <netdb.h>
#define PORT 5555
#define MESSAGE "Yow!!! Are we having fun yet?!?"
#define SERVERHOST "churchy.gnu.ai.mit.edu"
void
write_to_server (int filedes)
{
int nbytes;
nbytes = write (filedes, MESSAGE, strlen (MESSAGE) + 1);
if (nbytes < 0)
{
perror ("write");
exit (EXIT_FAILURE);
}
}
int
main (void)
{
extern void init_sockaddr (struct sockaddr_in *name,
const char *hostname, unsigned short int port);
int sock;
struct sockaddr_in servername;
/* Create the socket. */
sock = socket (PF_INET, SOCK_STREAM, 0);
if (sock < 0)
{
perror ("socket (client)");
exit (EXIT_FAILURE);
}
/* Connect to the server. */
init_sockaddr (&servername, SERVERHOST, PORT);
if (0 > connect (sock,
(struct sockaddr *) &servername,
sizeof (servername)))
{
perror ("connect (client)");
exit (EXIT_FAILURE);
}
/* Send data to the server. */
write_to_server (sock);
close (sock);
exit (EXIT_SUCCESS);
}
The server end is much more complicated. Since we want to allow
multiple clients to be connected to the server at the same time, it
would be incorrect to wait for input from a single client by simply
calling read or recv. Instead, the right thing to do is
to use select (see section Waiting for Input or Output) to wait for input on
all of the open sockets. This also allows the server to deal with
additional connection requests.
This particular server doesn't do anything interesting once it has gotten a message from a client. It does close the socket for that client when it detects an end-of-file condition (resulting from the client shutting down its end of the connection).
This program uses make_socket and init_sockaddr to set
up the socket address; see section Internet Socket Example.
#include <stdio.h>
#include <errno.h>
#include <stdlib.h>
#include <unistd.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include <netdb.h>
#define PORT 5555
#define MAXMSG 512
int
read_from_client (int filedes)
{
char buffer[MAXMSG];
int nbytes;
nbytes = read (filedes, buffer, MAXMSG);
if (nbytes < 0)
{
/* Read error. */
perror ("read");
exit (EXIT_FAILURE);
}
else if (nbytes == 0)
/* End-of-file. */
return -1;
else
{
/* Data read. */
fprintf (stderr, "Server: got message: `%s'\n", buffer);
return 0;
}
}
int
main (void)
{
extern int make_socket (unsigned short int port);
int sock;
int status;
fd_set active_fd_set, read_fd_set;
int i;
struct sockaddr_in clientname;
size_t size;
/* Create the socket and set it up to accept connections. */
sock = make_socket (PORT);
if (listen (sock, 1) < 0)
{
perror ("listen");
exit (EXIT_FAILURE);
}
/* Initialize the set of active sockets. */
FD_ZERO (&active_fd_set);
FD_SET (sock, &active_fd_set);
while (1)
{
/* Block until input arrives on one or more active sockets. */
read_fd_set = active_fd_set;
if (select (FD_SETSIZE, &read_fd_set, NULL, NULL, NULL) < 0)
{
perror ("select");
exit (EXIT_FAILURE);
}
/* Service all the sockets with input pending. */
for (i = 0; i < FD_SETSIZE; ++i)
if (FD_ISSET (i, &read_fd_set))
{
if (i == sock)
{
/* Connection request on original socket. */
size = sizeof (clientname);
if (accept (sock,
(struct sockaddr *) &clientname, &size) < 0)
{
perror ("accept");
exit (EXIT_FAILURE);
}
fprintf (stderr, "Server: connect from host %s, port %hd.\n",
inet_ntoa (clientname.sin_addr),
ntohs (clientname.sin_port));
FD_SET (status, &active_fd_set);
}
else
{
/* Data arriving on an already-connected socket. */
if (read_from_client (i) < 0)
{
close (i);
FD_CLR (i, &active_fd_set);
}
}
}
}
}
Streams with connections permit out-of-band data that is
delivered with higher priority than ordinary data. Typically the
reason for sending out-of-band data is to send notice of an
exceptional condition. The way to send out-of-band data is using
send, specifying the flag MSG_OOB (see section Sending Data).
Out-of-band data is received with higher priority because the
receiving process need not read it in sequence; to read the next
available out-of-band data, use recv with the MSG_OOB
flag (see section Receiving Data). Ordinary read operations do not read
out-of-band data; they read only the ordinary data.
When a socket finds that out-of-band data is on its way, it sends a
SIGURG signal to the owner process or process group of the
socket. You can specify the owner using the F_SETOWN command
to the fcntl function; see section Interrupt-Driven Input. You must
also establish a handler for this signal, as described in section Signal Handling, in order to take appropriate action such as reading the
out-of-band data.
Alternatively, you can test for pending out-of-band data, or wait
until there is out-of-band data, using the select function; it
can wait for an exceptional condition on the socket. See section Waiting for Input or Output, for more information about select.
Notification of out-of-band data (whether with SIGURG or with
select) indicates that out-of-band data is on the way; the data
may not actually arrive until later. If you try to read the
out-of-band data before it arrives, recv fails with an
EWOULDBLOCK error.
Sending out-of-band data automatically places a "mark" in the stream of ordinary data, showing where in the sequence the out-of-band data "would have been". This is useful when the meaning of out-of-band data is "cancel everything sent so far". Here is how you can test, in the receiving process, whether any ordinary data was sent before the mark:
success = ioctl (socket, SIOCATMARK, &result);
Here's a function to discard any ordinary data preceding the out-of-band mark:
int
discard_until_mark (int socket)
{
while (1)
{
/* This is not an arbitrary limit; any size will do. */
char buffer[1024];
int result, success;
/* If we have reached the mark, return. */
success = ioctl (socket, SIOCATMARK, &result);
if (success < 0)
perror ("ioctl");
if (result)
return;
/* Otherwise, read a bunch of ordinary data and discard it.
This is guaranteed not to read past the mark
if it starts before the mark. */
success = read (socket, buffer, sizeof buffer);
if (success < 0)
perror ("read");
}
}
If you don't want to discard the ordinary data preceding the mark, you
may need to read some of it anyway, to make room in internal system
buffers for the out-of-band data. If you try to read out-of-band data
and get an EWOULDBLOCK error, try reading some ordinary data
(saving it so that you can use it when you want it) and see if that
makes room. Here is an example:
struct buffer
{
char *buffer;
int size;
struct buffer *next;
};
/* Read the out-of-band data from SOCKET and return it
as a `struct buffer', which records the address of the data
and its size.
It may be necessary to read some ordinary data
in order to make room for the out-of-band data.
If so, the ordinary data is saved as a chain of buffers
found in the `next' field of the value. */
struct buffer *
read_oob (int socket)
{
struct buffer *tail = 0;
struct buffer *list = 0;
while (1)
{
/* This is an arbitrary limit.
Does anyone know how to do this without a limit? */
char *buffer = (char *) xmalloc (1024);
struct buffer *link;
int success;
int result;
/* Try again to read the out-of-band data. */
success = recv (socket, buffer, sizeof buffer, MSG_OOB);
if (success >= 0)
{
/* We got it, so return it. */
struct buffer *link
= (struct buffer *) xmalloc (sizeof (struct buffer));
link->buffer = buffer;
link->size = success;
link->next = list;
return link;
}
/* If we fail, see if we are at the mark. */
success = ioctl (socket, SIOCATMARK, &result);
if (success < 0)
perror ("ioctl");
if (result)
{
/* At the mark; skipping past more ordinary data cannot help.
So just wait a while. */
sleep (1);
continue;
}
/* Otherwise, read a bunch of ordinary data and save it.
This is guaranteed not to read past the mark
if it starts before the mark. */
success = read (socket, buffer, sizeof buffer);
if (success < 0)
perror ("read");
/* Save this data in the buffer list. */
{
struct buffer *link
= (struct buffer *) xmalloc (sizeof (struct buffer));
link->buffer = buffer;
link->size = success;
/* Add the new link to the end of the list. */
if (tail)
tail->next = link;
else
list = link;
tail = link;
}
}
}
This section describes how to use communication styles that don't use
connections (styles SOCK_DGRAM and SOCK_RDM). Using
these styles, you group data into packets and each packet is an
independent communication. You specify the destination for each
packet individually.
Datagram packets are like letters: you send each one independently, with its own destination address, and they may arrive in the wrong order or not at all.
The listen and accept functions are not allowed for
sockets using connectionless communication styles.
The normal way of sending data on a datagram socket is by using the
sendto function, declared in `sys/socket.h'.
You can call connect on a datagram socket, but this only
specifies a default destination for further data transmission on the
socket. When a socket has a default destination, then you can use
send (see section Sending Data) or even write (see section Input and Output Primitives) to send a packet there. You can cancel the default
destination by calling connect using an address format of
AF_UNSPEC in the addr argument. See section Making a Connection, for
more information about the connect function.
Function: int sendto (int socket, void *buffer. size_t size, int flags, struct sockaddr *addr, size_t length)
The sendto function transmits the data in the buffer
through the socket socket to the destination address specified
by the addr and length arguments. The size argument
specifies the number of bytes to be transmitted.
The flags are interpreted the same way as for send; see
section Socket Data Options.
The return value and error conditions are also the same as for
send, but you cannot rely on the system to detect errors and
report them; the most common error is that the packet is lost or there
is no one at the specified address to receive it, and the operating
system on your machine usually does not know this.
It is also possible for one call to sendto to report an error
due to a problem related to a previous call.
The recvfrom function reads a packet from a datagram socket and
also tells you where it was sent from. This function is declared in
`sys/socket.h'.
Function: int recvfrom (int socket, void *buffer, size_t size, int flags, struct sockaddr *addr, size_t *length_ptr)
The recvfrom function reads one packet from the socket
socket into the buffer buffer. The size argument
specifies the maximum number of bytes to be read.
If the packet is longer than size bytes, then you get the first size bytes of the packet, and the rest of the packet is lost. There's no way to read the rest of the packet. Thus, when you use a packet protocol, you must always know how long a packet to expect.
The addr and length_ptr arguments are used to return the address where the packet came from. See section Socket Addresses. For a socket in the file domain, the address information won't be meaningful, since you can't read the address of such a socket (see section The File Namespace). You can specify a null pointer as the addr argument if you are not interested in this information.
The flags are interpreted the same way as for recv
(see section Socket Data Options). The return value and error conditions
are also the same as for recv.
You can use plain recv (see section Receiving Data) instead of
recvfrom if you know don't need to find out who sent the packet
(either because you know where it should come from or because you
treat all possible senders alike). Even read can be used if
you don't want to specify flags (see section Input and Output Primitives).
Here is a set of example programs that send messages over a datagram
stream in the file namespace. Both the client and server programs use the
make_named_socket function that was presented in section The File Namespace, to create and name their sockets.
First, here is the server program. It sits in a loop waiting for messages to arrive, bouncing each message back to the sender. Obviously, this isn't a particularly useful program, but it does show the general ideas involved.
#include <stdio.h>
#include <errno.h>
#include <stdlib.h>
#include <sys/socket.h>
#include <sys/un.h>
#define SERVER "/tmp/serversocket"
#define MAXMSG 512
int
main (void)
{
int sock;
char message[MAXMSG];
struct sockaddr_un name;
size_t size;
int nbytes;
/* Make the socket, then loop endlessly. */
sock = make_named_socket (SERVER);
while (1)
{
/* Wait for a datagram. */
size = sizeof (name);
nbytes = recvfrom (sock, message, MAXMSG, 0,
(struct sockaddr *) & name, &size);
if (nbytes < 0)
{
perror ("recfrom (server)");
exit (EXIT_FAILURE);
}
/* Give a diagnostic message. */
fprintf (stderr, "Server: got message: %s\n", message);
/* Bounce the message back to the sender. */
nbytes = sendto (sock, message, nbytes, 0,
(struct sockaddr *) & name, size);
if (nbytes < 0)
{
perror ("sendto (server)");
exit (EXIT_FAILURE);
}
}
}
Here is the client program corresponding to the server above.
It sends a datagram to the server and then waits for a reply. Notice that the socket for the client (as well as for the server) in this example has to be given a name. This is so that the server can direct a message back to the client. Since the socket has no associated connection state, the only way the server can do this is by referencing the name of the client.
#include <stdio.h>
#include <errno.h>
#include <unistd.h>
#include <stdlib.h>
#include <sys/socket.h>
#include <sys/un.h>
#define SERVER "/tmp/serversocket"
#define CLIENT "/tmp/mysocket"
#define MAXMSG 512
#define MESSAGE "Yow!!! Are we having fun yet?!?"
int
main (void)
{
extern int make_named_socket (const char *name);
int sock;
char message[MAXMSG];
struct sockaddr_un name;
size_t size;
int nbytes;
/* Make the socket. */
sock = make_named_socket (CLIENT);
/* Initialize the server socket address. */
name.sun_family = AF_UNIX;
strcpy (name.sun_path, SERVER);
size = strlen (name.sun_path) + sizeof (name.sun_family);
/* Send the datagram. */
nbytes = sendto (sock, MESSAGE, strlen (MESSAGE) + 1, 0,
(struct sockaddr *) & name, size);
if (nbytes < 0)
{
perror ("sendto (client)");
exit (EXIT_FAILURE);
}
/* Wait for a reply. */
nbytes = recvfrom (sock, message, MAXMSG, 0, NULL, 0);
if (nbytes < 0)
{
perror ("recfrom (client)");
exit (EXIT_FAILURE);
}
/* Print a diagnostic message. */
fprintf (stderr, "Client: got message: %s\n", message);
/* Clean up. */
remove (CLIENT);
close (sock);
}
Keep in mind that datagram socket communications are unreliable. In
this example, the client program waits indefinitely if the message
never reaches the server or if the server's response never comes
back. It's up to the user running the program to kill it and restart
it, if desired. A more automatic solution could be to use
select (see section Waiting for Input or Output) to establish a timeout period
for the reply, and in case of timeout either resend the message or
shut down the socket and exit.
inetd DaemonWe've explained above how to write a server program that does its own listening. Such a server must already be running in order for anyone to connect to it.
Another way to provide service for an Internet port is to let the daemon
program inetd do the listening. inetd is a program that
runs all the time and waits (using select) for messages on a
specified set of ports. When it receives a message, it accepts the
connection (if the socket style calls for connections) and then forks a
child process to run the corresponding server program. You specify the
ports and their programs in the file `/etc/inetd.conf'.
inetd Servers
Writing a server program to be run by inetd is very simple. Each time
someone requests a connection to the appropriate port, a new server
process starts. The connection already exists at this time; the
socket is available as the standard input descriptor and as the
standard output descriptor (descriptors 0 and 1) in the server
process. So the server program can begin reading and writing data
right away. Often the program needs only the ordinary I/O facilities;
in fact, a general-purpose filter program that knows nothing about
sockets can work as a byte stream server run by inetd.
You can also use inetd for servers that use connectionless
communication styles. For these servers, inetd does not try to accept
a connection, since no connection is possible. It just starts the
server program, which can read the incoming datagram packet from
descriptor 0. The server program can handle one request and then
exit, or you can choose to write it to keep reading more requests
until no more arrive, and then exit. You must specify which of these
two techniques the server uses, when you configure inetd.
inetd
The file `/etc/inetd.conf' tells inetd which ports to listen to
and what server programs to run for them. Normally each entry in the
file is one line, but you can split it onto multiple lines provided
all but the first line of the entry start with whitespace. Lines that
start with `#' are comments.
Here are two standard entries in `/etc/inetd.conf':
ftp stream tcp nowait root /libexec/ftpd ftpd talk dgram udp wait root /libexec/talkd talkd
An entry has this format:
service style protocol wait username program arguments
The service field says which service this program provides. It
should be the name of a service defined in `/etc/services'.
inetd uses service to decide which port to listen on for
this entry.
The fields style and protocol specify the communication style and the protocol to use for the listening socket. The style should be the name of a communication style, converted to lower case and with `SOCK_' deleted--for example, `stream' or `dgram'. protocol should be one of the protocols listed in `/etc/protocols'. The typical protocol names are `tcp' for byte stream connections and `udp' for unreliable datagrams.
The wait field should be either `wait' or `nowait'.
Use `wait' if style is a connectionless style and the
server, once started, handles multiple requests, as many as come in.
Use `nowait' if inetd should start a new process for each message
or request that comes in. If style uses connections, then
wait must be `nowait'.
user is the user name that the server should run as. inetd runs
as root, so it can set the user ID of its children arbitrarily. It's
best to avoid using `root' for user if you can; but some
servers, such as Telnet and FTP, read a username and password
themselves. These servers need to be root initially so they can log
in as commanded by the data coming over the network.
program together with arguments specifies the command to run to start the server. program should be an absolute file name specifying the executable file to run. arguments consists of any number of whitespace-separated words, which become the command-line arguments of program. The first word in arguments is argument zero, which should by convention be the program name itself (sans directories).
If you edit `/etc/inetd.conf', you can tell inetd to reread the
file and obey its new contents by sending the inetd process the
SIGHUP signal. You'll have to use ps to determine the
process ID of the inetd process, as it is not fixed.
This section describes how to read or set various options that modify the behavior of sockets and their underlying communications protocols.
When you are manipulating a socket option, you must specify which level the option pertains to. This describes whether the option applies to the socket interface, or to a lower-level communications protocol interface.
Here are the functions for examining and modifying socket options. They are declared in `sys/socket.h'.
Function: int getsockopt (int socket, int level, int optname, void *optval, size_t *optlen_ptr)
The getsockopt function gets information about the value of
option optname at level level for socket socket.
The option value is stored in a buffer that optval points to.
Before the call, you should supply in *optlen_ptr the
size of this buffer; on return, it contains the number of bytes of
information actually stored in the buffer.
Most options interpret the optval buffer as a single int
value.
The actual return value of getsockopt is 0 on success
and -1 on failure. The following errno error conditions
are defined:
EBADF
ENOTSOCK
ENOPROTOOPT
Function: int setsockopt (int socket, int level, int optname, void *optval, size_t optlen)
This function is used to set the socket option optname at level level for socket socket. The value of the option is passed in the buffer optval, which has size optlen.
The return value and error codes are the same as for getsockopt.
Use this constant as the level argument to getsockopt or
setsockopt to manipulate the socket-level options described in
this section.
Here is a table of socket-level option names; all are defined in the header file `sys/socket.h'.
SO_DEBUG
This option toggles recording of debugging information in the underlying
protocol modules. The value has type int; a nonzero value means
"yes".
SO_REUSEADDR
bind (see section Setting a Socket's Address)
should permit reuse of local addresses for this socket. If you enable
this option, you can actually have two sockets with the same Internet
port number; but the system won't allow you to use the two
identically-named sockets in a way that would confuse the Internet. The
reason for this option is that some higher-level Internet protocols,
including FTP, require you to keep reusing the same socket number.
The value has type int; a nonzero value means "yes".
SO_KEEPALIVE
int; a nonzero value means
"yes".
SO_DONTROUTE
int; a nonzero
value means "yes".
SO_LINGER
struct linger.
This structure type has the following members:
int l_onoff
close
blocks until the data is transmitted or the timeout period has expired.
int l_linger
int; a nonzero value means "yes".
read or recv without specifying the MSG_OOB
flag. See section Out-of-Band Data. The value has type int; a
nonzero value means "yes".
size_t, which is the size in bytes.
size_t, which is the size in bytes.
getsockopt only. It is used to
get the socket's communication style. SO_TYPE is the
historical name, and SO_STYLE is the preferred name in GNU.
The value has type int and its value designates a communication
style; see section Communication Styles.
This option can be used with getsockopt only. It is used to reset
the error status of the socket. The value is an int, which represents
the previous error status.
Many systems come with a database that records a list of networks known
to the system developer. This is usually kept either in the file
`/etc/networks' or in an equivalent from a name server. This data
base is useful for routing programs such as route, but it is not
useful for programs that simply communicate over the network. We
provide functions to access this data base, which are declared in
`netdb.h'.
This data type is used to represent information about entries in the networks database. It has the following members:
char *n_name
char **n_aliases
int n_addrtype
AF_INET for Internet networks.
unsigned long int n_net
Use the getnetbyname or getnetbyaddr functions to search
the networks database for information about a specific network. The
information is returned in a statically-allocated structure; you must
copy the information if you need to save it.
Function: struct netent * getnetbyname (const char *name)
The getnetbyname function returns information about the network
named name. It returns a null pointer if there is no such
network.
Function: struct netent * getnetbyaddr (long net, int type)
The getnetbyaddr function returns information about the network
of type type with number net. You should specify a value of
AF_INET for the type argument for Internet networks.
getnetbyaddr returns a null pointer if there is no such
network.
You can also scan the networks database using setnetent,
getnetent, and endnetent. Be careful in using these
functions, because they are not reentrant.
Function: void setnetent (int stayopen)
This function opens and rewinds the networks database.
If the stayopen argument is nonzero, this sets a flag so that
subsequent calls to getnetbyname or getnetbyaddr will
not close the database (as they usually would). This makes for more
efficiency if you call those functions several times, by avoiding
reopening the database for each call.
Function: struct netent * getnetent (void)
This function returns the next entry in the networks database. It returns a null pointer if there are no more entries.
Function: void endnetent (void)
This function closes the networks database.
Go to the previous, next section.