NAME¶

STANDARD NETWORKING MODULE¶

IP address
The IP based communication uses a standard address to reference a particular peer. With IP version 4, the standard dot notation is with 4 bytes. With IP version 6, the standard semicolon notation is with 16 bytes. The current implementation supports both versions.

127.0.0.1       # ipv4 localhost
0:0:0:0:0:0:0:1 # ipv6 localhost

IP address architecture and behavior are described in various documents as listed in the bibliography.

Domain name system
The translation between a host name and an IP address is performed by a resolverwhich uses the Domain Name Systemor DNS. Access to the DNS is automatic with the implementation. Depending on the machine resolver configuration, a particular domain name translation might result in an IP version 4 or IP version 6 address. Most of the time, an IP version 4 address is returned. The mapping between an IP address and a host name returns the associated canonical namefor that IP address. This is the reverse of the preceding operation.

The Address class
The Addressclass allows manipulation of IP address. The constructor takes a string as its arguments. The argument string can be either an IP address or a host name which can be qualified or not. When the address is constructed with a host name, the IP address resolution is done immediately.

Name to address translation
The most common operation is to translate a host name to its equivalent IP address. Once the Addressobject is constructed, the get-addressmethod returns a string representation of the internal IP address. The following example prints the IP address of the localhost, that is 127.0.0.1with IP version 4.

# load network module
interp:library "afnix-net"
# get the localhost address
const addr (afnix:net:Address "localhost")
# print the ip address
println (addr:get-address)

As another example, the get-host-namefunction returns the host name of the running machine. The previous example can be used to query its IP address.

Address to name translation
The reverse operation of name translation maps an IP address to a canonical name. It shall be noted that the reverse lookup is not done automatically, unless the reverse flagis set in the constructoor. The get-canonical-namemethod of the Addressclass returns such name. Example XNET001.alsis a demonstration program which prints the address original name, the IP address and the canonical name. Fell free to use it with your favorite site to check the equivalence between the original name and the canonical name.

# print the ip address information of the arguments
# usage: axi XNET001.als [hosts ...]
# get the network module
interp:library "afnix-net"
# print the ip address
const ip-address-info (host) {
  try {
    const addr (afnix:net:Address host true)
    println "host name        : " (addr:get-name)
    println "  ip address     : " (addr:get-address)
    println "  canonical name : " (
      addr:get-canonical-name)
    # get aliases
    const size (addr:get-alias-size)
    loop (trans i 0) (< i size) (i:++) {
      println "  alias address  : " (
        addr:get-alias-address i)
      println "  alias name     : " (
        addr:get-alias-name i)
    }      
  } (errorln "error: " what:reason)
}
# get the hosts 
for (s) (interp:argv) (ip-address-info s)
zsh> axi net-0001.als localhost
host name        : localhost
ip address     : 127.0.0.1
canonical name : localhost

Address operations
The Addressclass provides several methods and operators that ease the address manipulation in a protocol indepedant way. For example, the ==operator compares two addresses. The ++operator can also be used to get the next IP address.

Transport layers
The two transport layer protocols supported by the Internet protocol is the TCP, a full-duplex oriented protocol, and UDP, a datagram protocol. TCP is a reliable protocol while UDP is not. By reliable, we mean that the protocol provides automatically some mechanisms for error recovery, message delivery, acknowledgment of reception, etc... The use of TCP vs. UDP is dictated mostly by the reliability concerns, while UDP reduces the traffic congestion.

Service port
In the client-server model, a connection is established between two hosts. The connections is made via the IP address and the port number. For a given service, a port identifies that service at a particular address. This means that multiple services can exist at the same address. More precisely, the transport layer protocol is also used to distinguish a particular service. The network module provides a simple mechanism to retrieve the port number, given its name and protocol. The function get-tcp-serviceand get-udp-servicereturns the port number for a given service by name. For example, the daytimeserver is located at port number 13.

assert 13 (afnix:net:get-tcp-service "daytime")
assert 13 (afnix:net:get-udp-service "daytime")

Host and peer
With the client server model, the only information needed to identify a particular client or server is the address and the port number. When a client connects to a server, it specify the port number the server is operating. The client uses a random port number for itself. When a server is created, the port number is used to bind the server to that particular port. If the port is already in use, that binding will fail. From a reporting point of view, a connection is therefore identified by the running host address and port, and the peer address and port. For a client, the peer is the server. For a server, the peer is the client.

TCP client socket
The TcpClientclass creates an TCP client object by address and port. The address can be either a string or an Addressobject. During the object construction, the connection is established with the server. Once the connection is established, the client can use the readand writemethod to communicate with the server. The TcpClientclass is derived from the Socketclass which is derived from the InputStreamand OutputStreamclasses.

Day time client
The simplest example is a client socket which communicates with the daytime server. The server is normally running on all machines and is located at port 13.

# get the network module
interp:library "afnix-net"
# get the daytime server port
const port (afnix:net:get-tcp-service "daytime")
# create a tcp client socket
const s (afnix:net:TcpClient "localhost" port)
# read the data - the server close the connection
while (s:valid-p) (println (s:readln))

Example 3201.alsin the example directory prints the day time of the local host without argument or the day time of the argument. Feel free to use it with www.afnix.org. If the server you are trying to contact does not have a day time server, an exception will be raised and the program terminates.

zsh> axi 3201.als www.afnix.org

HTTP request example
Another example which illustrates the use of the TcpClientobject is a simple client which download a web page. At this stage we are not concern with the URL but rather the mechanics involved. The request is made by opening a TCP client socket on port 80 (the HTTP server port) and sending a request by writing some HTTP commands. When the commands have been sent, the data sent by the server are read and printed on the standard output. Note that this example is not concerned by error detection.

# fetch an html page by host and page
# usage: axi 3203.als [host] [page]
# get the network module
interp:library "afnix-net"
interp:library "afnix-sys"
# connect to the http server and issue a request
const send-http-request (host page) {
  # create a client sock on port 80
  const s     (afnix:net:TcpClient host 80)
  const saddr (s:get-socket-address)
  # format the request
  s:writeln "GET " page " HTTP/1.1"
  s:writeln "Host: " (saddr:get-canonical-name)
  s:writeln "Connection: close"
  s:writeln "User-Agent: afnix tcp client example"
  s:newline
  # write the result
  while (s:valid-p) (println (s:readln))
}
# get the argument
if (!= (interp:argv:length) 2) (afnix:sys:exit 1)
const host (interp:argv:get 0)
const page (interp:argv:get 1)
# send request
send-http-request host page

UDP client socket
UDP client socket is similar to TCP client socket. However, due to the unreliable nature of UDP, UDP clients are somehow more difficult to manage. Since there is no flow control, it becomes more difficult to assess whether or not a datagram has reached its destination. The same apply for a server, where a reply datagram might be lost. The UdpClientclass is the class which creates a UDP client object. Its usage is similar to the TcpClient.

The time client
The UDP time server normally runs on port 37 is the best place to enable it. A UDP client is created with the UdpClientclass. Once the object is created, the client sends an empty datagram to the server. The server send a reply datagram with 4 bytes, in network byte order, corresponding to the date as of January 1st 1900. Example 3204.alsprints date information after contacting the local host time server or the host specified as the first argument.

# get the libraries
interp:library "afnix-net"
interp:library "afnix-sys"
# get the daytime server port
const port (afnix:net:get-udp-service "time")
# create a client socket and read the data
const print-time (host) {
  # create a udp client socket
  const s (afnix:net:UdpClient host port)
  # send an empty datagram
  s:write
  # read the 4 bytes data and adjust to epoch
  const buf (s:read 4)
  const val (- (buf:get-quad) 2208988800)
  # format the date
  const time (afnix:sys:Time val)
  println (time:format-date) ' ' (time:format-time)
}
# check for one argument or use localhost
const host (if (== (interp:argv:length) 0) 
  "localhost" (interp:argv:get 0))
print-time host

This example calls for several comments. First the writemethod without argument sends an empty datagram. It is the datagram which trigger the server. The readmethod reads 4 bytes from the reply datagram and places them in a Bufferobject. Since the bytes are in network byte order, the conversion into an integer value is done with the get-quadmethod. Finally, in order to use the Timeclass those epoch is January 1st 1970, the constant 2208988800is subtracted from the result. Remember that the time server sends the date in reference to January 1st 1900. More information about the time server can be found in RFC738.

More on reliability
The previous example has some inherent problems due to the unreliability of UDP. If the first datagram is lost, the readmethod will block indefinitely. Another scenario which causes the readmethod to block is the loss of the server reply datagram. Both problem can generally be fixed by checking the socket with a timeout using the valid-pmethod. With one argument, the method timeout and return false. In this case, a new datagram can be send to the server. Example 3205.alsillustrates this point. We print below the extract of code.

# create a client socket and read the data
const print-time (host) {
  # create a udp client socket
  const s (afnix:net:UdpClient host port)
  # send an empty datagram until the socket is valid
  s:write
  # retransmit datagram each second
  while (not (s:valid-p 1000)) (s:write)
  # read the 4 bytes data and adjust to epoch
  const buf (s:read 4)
  const val (- (buf:get-quad) 2208988800)
  # format the date
  const time (afnix:sys:Time val)
  println (time:format-date) ' ' (time:format-time)
}

Note that this solution is a naive one. In the case of multiple datagrams, a sequence number must be placed because there is no clue about the lost datagram. A simple rule of thumb is to use TCP as soon as reliability is a concern, but this choice might not so easy.

Error detection
Since UDP is not reliable, there is no simple solution to detect when a datagram has been lost. Even worse, if the server is not running, it is not easy to detect that the client datagram has been lost. In such situation, the client might indefinitely send datagram without getting an answer. One solution to this problem is again to count the number of datagram re-transmit and eventually give up after a certain time.

Socket class
The Socketclass is the base class for both TcpClientand UdpClient. The class provides methods to query the socket port and address as well as the peer port and address. Note at this point that the UDP socket is a connected socket. Therefore, these methods will work fine. The get-socket-addressand get-socket-portreturns respectively the address and port of the connected socket. The get-peer-addressand get-peer-portreturns respectively the address and port of the connected socket's peer. Example 3206.alsillustrates the use of these methods.

# create a client socket and read the data
const print-socket-info (host) {
  # create a tcp client socket
  const s (afnix:net:TcpClient host port)
  # print socket address and port
  const saddr (s:get-socket-address)
  const sport (s:get-socket-port)
  println "socket ip address     : " (
    saddr:get-address)
  println "socket canonical name : " (
    saddr:get-canonical-name)
  println "socket port           : " sport
  # print peer address and port
  const paddr (s:get-peer-address)
  const pport (s:get-peer-port)
  println "peer ip address       : " (
    paddr:get-address)
  println "peer canonical name   : " (
    paddr:get-canonical-name)
  println "peer port             : " pport
}

Socket predicates
The Socketclass is associated with the socket-ppredicate. The respective client objects have the tcp-client-ppredicate and udp-client-ppredicate.

TCP server socket
The TcpServerclass creates an TCP server object. There are several constructors for the TCP server. In its simplest form, without port, a TCP server is created on the localhostwith an ephemeral port number (i.e port 0 during the call). With a port number, the TCP server is created on the localhost. For a multi-homed host, the address to use to run the server can be specified as the first argument. The address can be either a string or an Addressobject. In both cases, the port is specified as the second argument. Finally, a third argument called the backlogcan be specified to set the number of acceptable incoming connection. That is the maximum number of pending connection while processing a connection. The following example shows various ways to create a TCP server.

trans s (afnix:net:TcpServer)
trans s (afnix:net:TcpServer 8000)
trans s (afnix:net:TcpServer 8000 5)
trans s (afnix:net:TcpServer "localhost" 8000)
trans s (afnix:net:TcpServer "localhost" 8000 5)
trans s (afnix:net:TcpServer (
    Address "localhost") 8000)
trans s (afnix:net:TcpServer (
    Address "localhost") 8000 5)

Echo server example
A simple echo servercan be built and tested with the standard telnetapplication. The application will echo all lines that are typed with the telnetclient. The server is bound on the port 8000, since ports 0 to 1024 are privileged ports.

# get the network module
interp:library "afnix-net"
# create a tcp server on port 8000
const srv (afnix:net:TcpServer 8000)
# wait for a connection
const s (srv:accept)
# echo the line until the end
while (s:valid-p) (s:writeln (s:readln))

The telnetsession is then quite simple. The line hello worldis echoed by the server.

zsh> telnet localhost 8000
Trying 127.0.0.1...
Connected to localhost.
Escape character is '^]'.
hello world
^D

The accept method
The previous example illustrates the mechanics of a server. When the server is created, the server is ready to accept connection. The acceptmethod blocks until a client connect with the server. When the connection is established, the acceptmethod returns a socket object which can be used to read and write data.

Multiple connections
One problem with the previous example is that the server accepts only one connection. In order to accept multiple connection, the acceptmethod must be placed in a loop, and the server operation in a thread (There are some situations where a new process might be more appropriate than a thread). Example 3302.alsillustrates such point.

# get the network module
interp:library "afnix-net"
# this function echo a line from the client
const echo-server (s) {
  while (s:valid-p) (s:writeln (s:readln))
}
# create a tcp server on port 8000
const srv (afnix:net:TcpServer 8000)
# wait for a connection
while true {
  trans s (srv:accept)
  launch  (echo-server s)
}

UDP server socket
The UdpServerclass is similar to the TcpServerobject, except that there is no backlog parameters. In its simplest form, the UDP server is created on the localhostwith an ephemeral port (i.e port 0). With a port number, the server is created on the localhost. For a multi-homed host, the address used to run the server can be specified as the first argument. The address can be either a string or an Addressobject. In both cases, the port is specified as the second argument.

trans s (afnix:net:UdpServer)
trans s (afnix:net:UdpServer 8000)
trans s (afnix:net:UdpServer "localhost" 8000)
trans s (afnix:net:UdpServer (
    Address "localhost") 8000)

Echo server example
The echo servercan be revisited to work with udp datagram. The only difference is the use of the acceptmethod. For a UDP server, the method return a Datagramobject which can be used to read and write data.

# get the network module
interp:library "afnix-net"
# create a udp server on port 8000
const srv (afnix:net:UdpServer 8000)
# wait for a connection
while true {
  trans dg   (srv:accept)
  dg:writeln (dg:readln)
}

Datagram object
With a UDP server, the acceptmethod returns a Datagramobject. Because a UDP is connection-less, the server has no idea from whom the datagram is coming until that one has been received. When a datagram arrives, the Datagramobject is constructed with the peer address being the source address. Standard i/o methods can be used to read or write. When a write method is used, the data are sent back to the peer in a form of another datagram.

# wait for a datagram
trans dg (s:accept)
# assert datagram type
assert true (datagram-p dg)
# get contents length
println "datagram buffer size : " (dg:get-buffer-length)
# read a line from this datagram
trans line (dg:readln)
# send it back to the sender
s:writeln line

Input data buffer
For a datagram, and generally speaking, for a UDP socket, all input operations are buffered. This means that when a datagram is received, the acceptmethod places all data in an input buffer. This means that a read operation does not necessarily flush the whole buffer but rather consumes only the requested character. For example, if one datagram contains the string hello world. A call to readlnwill return the entire string. A call to read will return only the character 'h'. Subsequent call will return the next available characters. A call like read 5will return a buffer with 5 characters. Subsequent calls will return the remaining string. In any case, the get-buffer-lengthwill return the number of available characters in the buffer. A call to valid-pwill return true if there are some characters in the buffer or if a new datagram has arrived. Care should be taken with the readmethod. For example if there is only 4 characters in the input buffer and a call to read for 10 characters is made, the method will block until a new datagram is received which can fill the remaining 6 characters. Such situation can be avoided by using the get-buffer-lengthand the valid-pmethods. Note also that a timeout can be specified with the valid-pmethod.

Low level socket methods
Some folks always prefer to do everything by themselves. Most of the time for good reasons. If this is your case, you might have to use the low level socket methods. Instead of using a client or server class, the implementation let's you create a TcpSocketor UdpSocket. Once this done, the bind, connectand other methods can be used to create the desired connection.

A socket client
A simple TCP socket client is created with the TcpSocketclass. Then the connectmethod is called to establish the connection.

# create an address and a tcp socket
const addr (afnix:net:Address "localhost")
const sid  (afnix:net:TcpSocket)
# connect the socket
sid:connect 13 addr

Once the socket is connected, normal read and write operations can be performed. After the socket is created, it is possible to set some options. A typical one is NO-DELAYwhich disable the Naggle algorithm.

# create an address and a tcp socket
const addr (afnix:net:Address "localhost")
const sid  (afnix:net:TcpSocket)
# disable the naggle algorithm
sid:set-option sid:NO-DELAY true
# connect the socket
sid:connect 13 addr

NETWORKING REFERENCE¶

Predicate

Inheritance

Constructors

Operators

Methods

Functions

Socket
The Socketclass is a base class for the AFNIX network services. The class is automatically constructed by a derived class and provide some common methods for all socket objects.

Predicate

Inheritance

Constants

Methods

TcpSocket
The TcpSocketclass is a base class for all tcp socket objects. The class is derived from the Socketclass and provides some specific tcp methods. If a TcpSocketis created, the user is responsible to connect it to the proper address and port.

Predicate

Inheritance

Constructors

Methods

TcpClient
The TcpClientclass creates a tcp client by host and port. The host argument can be either a name or an address object. The port argument is the server port to contact. The TcpClientclass is derived from the TcpSocketclass. This class has no specific methods.

Predicate

Inheritance

Constructors

TcpServer
The TcpServerclass creates a tcp server by port. An optional host argument can be either a name or an address object. The port argument is the server port to bind. The TcpServerclass is derived from the TcpSocketclass. This class has no specific methods. With one argument, the server bind the port argument on the local host. The backlog can be specified as the last argument. The host name can also be specified as the first argument, the port as second argument and eventually the backlog. Note that the host can be either a string or an address object.

Predicate

Inheritance

Constructors

Datagram
The Datagramclass is a socket class used by udp socket. A datagram is constructed by the UdpSocketacceptmethod. The purpose of a datagram is to store the peer information so one can reply to the sender. The datagram also stores in a buffer the data sent by the peer. This class does not have any constructor nor any specific method.

Predicate

Inheritance

UdpSocket
The UdpSocketclass is a base class for all udp socket objects. The class is derived from the Socketclass and provides some specific udp methods.

Predicate

Inheritance

Constructors

Methods

UdpClient
The UdpClientclass creates a udp client by host and port. The host argument can be either a name or an address object. The port argument is the server port to contact. The UdpClientclass is derived from the UdpSocketclass. This class has no specific methods.

Predicate

Inheritance

Constructors

UdpServer
The UdpServerclass creates a udp server by port. An optional host argument can be either a name or an address object. The port argument is the server port to bind. The UdpServerclass is derived from the UdpSocketclass. This class has no specific methods. With one argument, the server bind the port argument on the local host. The host name can also be specified as the first argument, the port as second argument. Note that the host can be either a string or an address object.

Predicate

Inheritance

Constructors

Multicast
The Multicastclass creates a udp multicast socket by port. An optional host argument can be either a name or an address object. The port argument is the server port to bind. The Multicastclass is derived from the UdpSocketclass. This class has no specific methods. With one argument, the server bind the port argument on the local host. The host name can also be specified as the first argument, the port as second argument. Note that the host can be either a string or an address object. This class is similar to the UdpServerclass, except that the socket join the multicast group at construction and leave it at destruction.

Predicate

Inheritance

Constructors