|
| Network support for the Lua language |
|
|
| Network support for the Lua language |
LuaSocket is a Lua extension library that is composed by two parts: a C core that provides support for the TCP and UDP transport layers, and a set of Lua modules that add support for the SMTP (sending e-mails), HTTP (WWW access) and FTP (uploading and downloading files) protocols and other functionality commonly needed by applications that deal with the Internet. This introduction is about the C core.
Communication in LuaSocket is performed via I/O objects. These can represent different network domains. Currently, support is provided for TCP and UDP, but nothing prevents other developers from implementing SSL, Local Domain, Pipes, File Descriptors etc. I/O objects provide a standard interface to I/O across different domains and operating systems.
The API design had two goals in mind. First, users experienced with the C API to sockets should feel comfortable using LuaSocket. Second, the simplicity and the feel of the Lua language should be preserved. To achieve these goals, the LuaSocket API keeps the function names and semantics the C API whenever possible, but their usage in Lua has been greatly simplified.
One of the simplifications is the receive pattern capability. Applications can read data from stream domains (such as TCP) line by line, block by block, or until the connection is closed. All I/O reads are buffered and the performance differences between different receive patterns are negligible.
Another advantage is the flexible timeout control mechanism. As in C, all I/O operations are blocking by default. For example, the send, receive and accept methods of the TCP domain will block the caller application until the operation is completed (if ever!). However, with a call to the settimeout method, an application can specify upper limits on the time it can be blocked by LuaSocket (the "total" timeout), on the time LuaSocket can internally be blocked by any OS call (the "block" timeout) or a combination of the two. Each LuaSocket call might perform several OS calls, so that the two timeout values are not equivalent.
Finally, the host name resolution is transparent, meaning that most functions and methods accept both IP addresses and host names. In case a host name is given, the library queries the system's resolver and tries the main IP address returned. Note that direct use of IP addresses is more efficient, of course. The toip and tohostname functions from the DNS module are provided to convert between host names and IP addresses.
Together, these changes make network programming in LuaSocket much simpler than it is in C, as the following sections will show.
TCP (Transfer Control Protocol) is reliable stream protocol. In other words, applications communicating through TCP can send and receive data as an error free stream of bytes. Data is split in one end and reassembled transparently on the other end. There are no boundaries in the data transfers. The library allows users to read data from the sockets in several different granularities: patterns are available for lines, arbitrary sized blocks or "read up to connection closed", all with good performance.
The library distinguishes three types of TCP sockets: master, client and server sockets.
Master sockets are newly created TCP sockets returned by the function socket.tcp. A master socket is transformed into a server socket after it is associated with a local address by a call to the bind method followed by a call to the listen. Conversely, a master socket can be changed into a client socket with the method connect, which associates it with a remote address.
On server sockets, applications can use the accept method to wait for a client connection. Once a connection is established, a client socket object is returned representing this connection. The other methods available for server socket objects are getsockname, setoption, settimeout, and close.
Client sockets are used to exchange data between two applications over the Internet. Applications can call the methods send and receive to send and receive data. The other methods available for client socket objects are getsockname, getpeername, setoption, settimeout, shutdown, and close.
Example:
A simple echo server, using LuaSocket. The program binds to an ephemeral port (one that is chosen by the operating system) on the local host and awaits client connections on that port. When a connection is established, the program reads a line from the remote end and sends it back, closing the connection immediately. You can test it using the telnet program.
-- load namespace local socket = require("socket") -- create a TCP socket and bind it to the local host, at any port local server = assert(socket.bind("*", 0)) -- find out which port the OS chose for us local ip, port = server:getsockname() -- print a message informing what's up print("Please telnet to localhost on port " .. port) print("After connecting, you have 10s to enter a line to be echoed") -- loop forever waiting for clients while 1 do -- wait for a connection from any client local client = server:accept() -- make sure we don't block waiting for this client's line client:settimeout(10) -- receive the line local line, err = client:receive() -- if there was no error, send it back to the client if not err then client:send(line .. "\n") end -- done with client, close the object client:close() end
UDP (User Datagram Protocol) is a non-reliable datagram protocol. In other words, applications communicating through UDP send and receive data as independent blocks, which are not guaranteed to reach the other end. Even when they do reach the other end, they are not guaranteed to be error free. Data transfers are atomic, one datagram at a time. Reading only part of a datagram discards the rest, so that the following read operation will act on the next datagram. The advantages are in simplicity (no connection setup) and performance (no error checking or error correction).
Note that although no guarantees are made, these days networks are so good that, under normal circumstances, few errors happen in practice.
An UDP socket object is created by the socket.udp function. UDP sockets do not need to be connected before use. The method sendto can be used immediately after creation to send a datagram to IP address and port. Host names are not allowed because performing name resolution for each packet would be forbiddingly slow. Methods receive and receivefrom can be used to retrieve datagrams, the latter returning the IP and port of the sender as extra return values (thus being slightly less efficient).
When communication is performed repeatedly with a single peer, an application should call the setpeername method to specify a permanent partner. Methods sendto and receivefrom can no longer be used, but the method send can be used to send data directly to the peer, and the method receive will only return datagrams originating from that peer. There is about 30% performance gain due to this practice.
To associate an UDP socket with a local address, an application calls the setsockname method before sending any datagrams. Otherwise, the socket is automatically bound to an ephemeral address before the first data transmission and once bound the local address cannot be changed. The other methods available for UDP sockets are getpeername, getsockname, settimeout, setoption and close.
Example:
A simple daytime client, using LuaSocket. The program connects to a remote server and tries to retrieve the daytime, printing the answer it got or an error message.
-- change here to the host an port you want to contact local host, port = "localhost", 13 -- load namespace local socket = require("socket") -- convert host name to ip address local ip = assert(socket.dns.toip(host)) -- create a new UDP object local udp = assert(socket.udp()) -- contact daytime host assert(udp:sendto("anything", ip, port)) -- retrieve the answer and print results io.write(assert(udp:receive()))
Although not covered in the introduction, LuaSocket offers much more than TCP and UDP functionality. As the library evolved, support for HTTP, FTP, and SMTP were built on top of these. These modules and many others are covered by the reference manual.