6 library(socket): Network socket (TCP and UDP) library

The library(socket) provides TCP and UDP inet-domain sockets from SWI-Prolog, both client and server-side communication. The interface of this library is very close to the Unix socket interface, also supported by the MS-Windows winsock API. SWI-Prolog applications that wish to communicate with multiple sources have three options:

• Use I/O multiplexing based on wait_for_input/3. On Windows systems this can only be used for sockets, not for general (device-) file handles.
• Use multiple threads, handling either a single blocking socket or a pool using I/O multiplexing as above.
• Using XPCE's class socket which synchronises socket events in the GUI event-loop.

6.1 Client applications

Using this library to establish a TCP connection to a server is as simple as opening a file. See also http_open/3.

dump_swi_homepage :-
    setup_call_cleanup(
        tcp_connect(www.swi-prolog.org:http, Stream, []),
        ( format(Stream,
                 'GET / HTTP/1.1~n\c
                  Host: www.swi-prolog.org~n\c
                  Connection: close~n~n', []),
          flush_output(Stream),
          copy_stream_data(Stream, current_output)
        ),
        close(S)).

To deal with timeouts and multiple connections, threads, wait_for_input/3 and/or non-blocking streams (see tcp_fcntl/3) can be used.

6.2 Server applications

The typical sequence for generating a server application is given below. To close the server, use close/1 on AcceptFd.

create_server(Port) :-
      tcp_socket(Socket),
      tcp_bind(Socket, Port),
      tcp_listen(Socket, 5),
      tcp_open_socket(Socket, AcceptFd, _),
      <dispatch>

There are various options for <dispatch>. The most commonly used option is to start a Prolog thread to handle the connection. Alternatively, input from multiple clients can be handled in a single thread by listening to these clients using wait_for_input/3. Finally, on Unix systems, we can use fork/1 to handle the connection in a new process. Note that fork/1 and threads do not cooperate well. Combinations can be realised but require good understanding of POSIX thread and fork-semantics.

Below is the typical example using a thread. Note the use of setup_call_cleanup/3 to guarantee that all resources are reclaimed, also in case of failure or exceptions.

dispatch(AcceptFd) :-
        tcp_accept(AcceptFd, Socket, Peer),
        thread_create(process_client(Socket, Peer), _,
                      [ detached(true)
                      ]),
        dispatch(AcceptFd).

process_client(Socket, Peer) :-
        setup_call_cleanup(
            tcp_open_socket(Socket, StreamPair),
            handle_service(StreamPair),
            close(StreamPair)).

handle_service(StreamPair) :-
        ...

6.3 Socket exceptions

Errors that are trapped by the low-level library are mapped to an exception of the shape below. In this term, Code is a lower case atom that corresponds to the C macro name, e.g., epipe for a broken pipe. Message is the human readable string for the error code returned by the OS or the same as Code if the OS does not provide this functionality. Note that Code is derived from a static set of macros that may or may not be defines for the target OS. If the macro name is not known, Code is ERROR_nnn, where nnn is an integer.

error(socket_error(Code, Message), _)

Note that on Windows Code is a wsa* code which makes it hard to write portable code that handles specific socket errors. Even on POSIX systems the exact set of errors produced by the network stack is not defined.

6.4 TCP socket predicates

[det]tcp_socket(-SocketId)

Creates an INET-domain stream-socket and unifies an identifier to it with SocketId. On MS-Windows, if the socket library is not yet initialised, this will also initialise the library.

[det]tcp_close_socket(+SocketId)

Closes the indicated socket, making SocketId invalid. Normally, sockets are closed by closing both stream handles returned by open_socket/3. There are two cases where tcp_close_socket/1 is used because there are no stream-handles:

• If, after tcp_accept/3, the server uses fork/1 to handle the client in a sub-process. In this case the accepted socket is not longer needed from the main server and must be discarded using tcp_close_socket/1.
• If, after discovering the connecting client with tcp_accept/3, the server does not want to accept the connection, it should discard the accepted socket immediately using tcp_close_socket/1.

[det]tcp_open_socket(+SocketId, -StreamPair)

Create streams to communicate to SocketId. If SocketId is a master socket (see tcp_bind/2), StreamPair should be used for tcp_accept/3. If SocketId is a connected (see tcp_connect/2) or accepted socket (see tcp_accept/3), StreamPair is unified to a stream pair (see stream_pair/3) that can be used for reading and writing. The stream or pair must be closed with close/1, which also closes SocketId.

[det]tcp_open_socket(+SocketId, -InStream, -OutStream)

Similar to tcp_open_socket/2, but creates two separate sockets where tcp_open_socket/2 would have created a stream pair.

deprecated

New code should use tcp_open_socket/2 because closing a stream pair is much easier to perform safely.

[det]tcp_bind(SocketId, ?Address)

Bind the socket to Address on the current machine. This operation, together with tcp_listen/2 and tcp_accept/3 implement the server-side of the socket interface. Address is either an plain Port or a term HostPort. The first form binds the socket to the given port on all interfaces, while the second only binds to the matching interface. A typical example is below, causing the socket to listen only on port 8080 on the local machine's network.

  tcp_bind(Socket, localhost:8080)

If Port is unbound, the system picks an arbitrary free port and unifies Port with the selected port number. Port is either an integer or the name of a registered service. See also tcp_connect/4.

[det]tcp_listen(+SocketId, +BackLog)

Tells, after tcp_bind/2, the socket to listen for incoming requests for connections. Backlog indicates how many pending connection requests are allowed. Pending requests are requests that are not yet acknowledged using tcp_accept/3. If the indicated number is exceeded, the requesting client will be signalled that the service is currently not available. A commonly used default value for Backlog is 5.

[det]tcp_accept(+Socket, -Slave, -Peer)

This predicate waits on a server socket for a connection request by a client. On success, it creates a new socket for the client and binds the identifier to Slave. Peer is bound to the IP-address of the client or the atom af_unix if Socket is an AF_UNIX socket (see unix_domain_socket/1).

[det]tcp_connect(+SocketId, +Address)

Connect SocketId. After successful completion, tcp_open_socket/3 can be used to create I/O-Streams to the remote socket. This predicate is part of the low level client API. A connection to a particular host and port is realised using these steps:

    tcp_socket(Socket),
    tcp_connect(Socket, Host:Port),
    tcp_open_socket(Socket, StreamPair)

Typical client applications should use the high level interface provided by tcp_connect/3 which avoids resource leaking if a step in the process fails, and can be hooked to support proxies. For example:

    setup_call_cleanup(
        tcp_connect(Host:Port, StreamPair, []),
        talk(StreamPair),
        close(StreamPair))

If SocketId is an AF_UNIX socket (see unix_domain_socket/1), Address is an atom or string denoting a file name.

[det]tcp_connect(+Socket, +Address, -Read, -Write)

Connect a (client) socket to Address and return a bi-directional connection through the stream-handles Read and Write. This predicate may be hooked by defining socket:tcp_connect_hook/4 with the same signature. Hooking can be used to deal with proxy connections. E.g.,

:- multifile socket:tcp_connect_hook/4.

socket:tcp_connect_hook(Socket, Address, Read, Write) :-
    proxy(ProxyAdress),
    tcp_connect(Socket, ProxyAdress),
    tcp_open_socket(Socket, Read, Write),
    proxy_connect(Address, Read, Write).

deprecated

New code should use tcp_connect/3 called as tcp_connect(+Address, -StreamPair, +Options).

[det]tcp_connect(+Address, -StreamPair, +Options)

[det]tcp_connect(+Socket, +Address, -StreamPair)

Establish a TCP communication as a client. The +,-,+ mode is the preferred way for a client to establish a connection. This predicate can be hooked to support network proxies. To use a proxy, the hook proxy_for_url/3 must be defined. Permitted options are:

bypass_proxy(+Boolean)

Defaults to false. If true, do not attempt to use any proxies to obtain the connection

nodelay(+Boolean)

Defaults to false. If true, set nodelay on the resulting socket using tcp_setopt(Socket, nodelay)

The +,+,- mode is deprecated and does not support proxies. It behaves like tcp_connect/4, but creates a stream pair (see stream_pair/3).

Address

is either a Host:Port term or a file name (atom or string). The latter connects to an AF_UNIX socket and requires unix_domain_socket/1.

Errors

proxy_error(tried(ResultList)) is raised by mode (+,-,+) if proxies are defines by proxy_for_url/3 but no proxy can establsh the connection. ResultList contains one or more terms of the form false(Proxy) for a hook that simply failed or error(Proxy, ErrorTerm) for a hook that raised an exception.

See also

library(http/http_proxy) defines a hook that allows to connect through HTTP proxies that support the CONNECT method.

tcp_select(+ListOfStreams, -ReadyList, +TimeOut)

Same as the built-in wait_for_input/3. Used to allow for interrupts and timeouts on Windows. A redesign of the Windows socket interface makes it impossible to do better than Windows select() call underlying wait_for_input/3. As input multiplexing typically happens in a background thread anyway we accept the loss of timeouts and interrupts.

deprecated

Use wait_for_input/3

[semidet,multifile]try_proxy(+Proxy, +TargetAddress, -Socket, -StreamPair)

Attempt a socket-level connection via the given proxy to TargetAddress. The Proxy argument must match the output argument of proxy_for_url/3. The predicate tcp_connect/3 (and http_open/3 from the library(http/http_open)) collect the results of failed proxies and raise an exception no proxy is capable of realizing the connection.

The default implementation recognises the values for Proxy described below. The library(http/http_proxy) adds proxy(Host,Port) which allows for HTTP proxies using the CONNECT method.

direct

Do not use any proxy

socks(Host, Port)

Use a SOCKS5 proxy

[nondet,multifile]proxy_for_url(+URL, +Hostname, -Proxy)

This hook can be implemented to return a proxy to try when connecting to URL. Returned proxies are tried in the order in which they are returned by the multifile hook try_proxy/4. Pre-defined proxy methods are:

direct

connect directly to the resource

proxy(Host, Port)

Connect to the resource using an HTTP proxy. If the resource is not an HTTP URL, then try to connect using the CONNECT verb, otherwise, use the GET verb.

socks(Host, Port)

Connect to the resource via a SOCKS5 proxy

These correspond to the proxy methods defined by PAC Proxy auto-config. Additional methods can be returned if suitable clauses for http:http_connection_over_proxy/6 or try_proxy/4 are defined.

[det]tcp_setopt(+SocketId, +Option)

Set options on the socket. Defined options are:

reuseaddr

Allow servers to reuse a port without the system being completely sure the port is no longer in use.

bindtodevice(+Device)

Bind the socket to Device (an atom). For example, the code below binds the socket to the loopback device that is typically used to realise the localhost. See the manual pages for setsockopt() and the socket interface (e.g., socket(7) on Linux) for details.

tcp_socket(Socket),
tcp_setopt(Socket, bindtodevice(lo))

nodelay

nodelay(true)

If true, disable the Nagle optimization on this socket, which is enabled by default on almost all modern TCP/IP stacks. The Nagle optimization joins small packages, which is generally desirable, but sometimes not. Please note that the underlying TCP_NODELAY setting to setsockopt() is not available on all platforms and systems may require additional privileges to change this option. If the option is not supported, tcp_setopt/2 raises a domain_error exception. See Wikipedia for details.

broadcast

UDP sockets only: broadcast the package to all addresses matching the address. The address is normally the address of the local subnet (i.e. 192.168.1.255). See udp_send/4.

ip_add_membership(+MultiCastGroup)

ip_add_membership(+MultiCastGroup, +LocalInterface)

ip_add_membership(+MultiCastGroup, +LocalInterface, +InterfaceIndex)

ip_drop_membership(+MultiCastGroup)

ip_drop_membership(+MultiCastGroup, +LocalInterface)

ip_drop_membership(+MultiCastGroup, +LocalInterface, +InterfaceIndex)

Join/leave a multicast group. Calls setsockopt() with the corresponding arguments.

dispatch(+Boolean)

In GUI environments (using XPCE or the Windows swipl-win.exe executable) this flags defines whether or not any events are dispatched on behalf of the user interface. Default is true. Only very specific situations require setting this to false.

sndbuf(+Integer)

Sets the send buffer size to Integer (bytes). On Windows this defaults (now) to 64kb. Higher latency links may benefit from increasing this further since the maximum theoretical throughput on a link is given by buffer-size / latency. See https://support.microsoft.com/en-gb/help/823764/slow-performance-occurs-when-you-copy-data-to-a-tcp-server-by-using-a for Microsoft's discussion

[det]tcp_fcntl(+Stream, +Action, ?Argument)

Interface to the fcntl() call. Currently only suitable to deal switch stream to non-blocking mode using:

  tcp_fcntl(Stream, setfl, nonblock),

An attempt to read from a non-blocking stream while there is no data available returns -1 (or end_of_file for read/1), but at_end_of_stream/1 fails. On actual end-of-input, at_end_of_stream/1 succeeds.

[semidet]tcp_getopt(+Socket, ?Option)

Get information about Socket. Defined properties are below. Requesting an unknown option results in a domain_error exception.

file_no(-File)

Get the OS file handle as an integer. This may be used for debugging and integration.

[det]tcp_host_to_address(?HostName, ?Address)

Translate between a machines host-name and it's (IP-)address. If HostName is an atom, it is resolved using getaddrinfo() and the IP-number is unified to Address using a term of the format ip(Byte1,Byte2,Byte3,Byte4). Otherwise, if Address is bound to an ip(Byte1,Byte2,Byte3,Byte4) term, it is resolved by gethostbyaddr() and the canonical hostname is unified with HostName.

To be done

This function should support more functionality provided by gethostbyaddr, probably by adding an option-list.

[det]gethostname(-Hostname)

Return the canonical fully qualified name of this host. This is achieved by calling gethostname() and return the canonical name returned by getaddrinfo().

[det]negotiate_socks_connection(+DesiredEndpoint, +StreamPair)

Negotiate a connection to DesiredEndpoint over StreamPair. DesiredEndpoint should be in the form of either:

• hostname : port
• ip(A,B,C,D) : port

Errors

socks_error(Details) if the SOCKS negotiation failed.

6.5 Unix domain sockets

Unix domain sockets (sockets with address family AF_UNIX) are represented as a (socket) file in the file system. They can only be used to connect processes on the same host. The main advantage of AF_UNIX sockets is that name conflicts are much easier to manage than port conflicts and that access is determined by the file system permission rules.

unix_domain_socket(-Socket)

Creates an AF_UNIX-domain stream-socket and unifies an identifier to it with Socket. This predicate does not exist if the OS does not support the AF_UNIX address family (e.g. MS-Windows).

Unix domain socket affect tcp_connect/2 (for clients) and tcp_bind/2 and tcp_accept/3 (for servers). The address is an atom or string that is handled as a file name. On most systems the length of this file name is limited to 128 bytes (including null terminator), but according to the Linux documentation (unix(7)), portable applications must keep the address below 92 bytes. Note that these lengths are in bytes. Non-ascii characters may be represented as multiple bytes. If the length limit is exceeded a representation_error(af_unix_name) exception is raised.

6.6 UDP protocol support

The socket library provides support for UDP sockets. The UDP protocol is a connection-less and unreliable datagram based protocol. That means that messages sent may or may not arrive at the client side, may arrive in a different order then they were sent or even may arrive multiple times. UDP messages are often used for data replication, streaming media, or service discovery.

The library library(udp_broadcast) provides a high-level interface that builds on top of the library(broadcast) facilitating an unreliable publish-subscribe pattern based communication channel. The library(udp_broadcast) library supports multiple network architectures, e.g., classical LAN broadcasting, multicast and unicast.

udp_socket(-Socket)

Similar to tcp_socket/1, but create a socket using the SOCK_DGRAM protocol, ready for UDP connections.

udp_receive(+Socket, -Data, -From, +Options)

Wait for and return the next datagram. The data is returned as a Prolog string object (see string_to_list/2). From is a term of the format ip(A,B,C,D):Port indicating the sender of the message. Socket can be waited for using wait_for_input/3. Defined Options:

as(+Type)

Defines the returned term-type. Type is one of atom, codes or string (default).

max_message_size(+Size)

Specify the maximum number of bytes to read from a UDP datagram. Size must be within the range 0-65535. If unspecified, a maximum of 4096 bytes will be read.

The typical sequence to receive UDP data is:

receive(Port) :-
        udp_socket(S),
        tcp_bind(S, Port),
        repeat,
            udp_receive(Socket, Data, From, [as(atom)]),
            format('Got ~q from ~q~n', [Data, From]),
            fail.

udp_send(+Socket, +Data, +To, +Options)

Send a UDP message. Data is a string, atom or code-list providing the data. To is an address of the form Host:Port where Host is either the hostname or a term ip/4. Options is currently unused.

A simple example to send UDP data is:

send(Host, Port, Message) :-
        udp_socket(S),
        udp_send(S, Message, Host:Port, []),
        tcp_close_socket(S).

A broadcast is achieved by using tcp_setopt(Socket, broadcast) prior to sending the datagram and using the local network broadcast address as a ip/4 term.

The normal mechanism to discover a service on the local network is for the client to send a broadcast message to an agreed port. The server receives this message and replies to the client with a message indicating further details to establish the communication.