Notes

Web server basics (1): HTTP and sockets

April 6, 2019

Say we have some HTML:

<html>
  <body>
    <h1>Hello world!</h1>
  </body>
</html>

And say we'd like to be able to render this page in a web browser. If the server is hosted locally we may want to enter localhost:9000/hello-world.html in the address bar, hit enter, make a request (done by the browser), receive a response (sent by some server), and render the result (done by the browser).

Here is a minimal, often incomplete, and unsafe Node.js program (about 100 LoC) that would serve this (code available on Github):

const fs = require('fs');
const net = require('net');

const CRLF = '\r\n';

const HELLO_WORLD = `<html>
  <body>
    <h1>Hello world!</h1>
  </body>
</html>`;
const NOT_FOUND = `<html>
  <body>
    <h1>Not found</h1>
  </body>
</html>`;

class HTTPRequestHandler {
  constructor(connection) {
    this.connection = connection;
    this.request = {
      statusLine: null,
      headers: {},
      body: null,
    };
  }

  parse(buffer) {
    const lines = buffer.toString().split(CRLF);

    // Parse/store status line if necessary
    if (!this.request.statusLine) {
      const [method, path, protocol] = lines.shift().split(' ');
      this.request.statusLine = { method, path, protocol };
    }

    // Parse/store headers if the body hasn't begun
    if (!this.request.body) {
      for (let line = lines.shift(); lines.length; line = lines.shift()) {
        // Reached the end of headers, double CRLF
        if (line === '') {
          this.request.body = '';
          break;
        }

        const [key, value] = line.split(':');

        const safeKey = key.toLowerCase();
        if (!this.request.headers[safeKey]) {
          this.request.headers[safeKey] = [];
        }

        this.request.headers[safeKey].push(value.trimStart());
      }
    }

    this.request.body += lines.join(CRLF);
  }

  requestComplete() {
    if (!this.request.statusLine || !Object.keys(this.request.headers).length || this.request.body === null) {
      return false;
    }

    const [contentLength] = this.request.headers['content-length'] || [];
    if (this.request.statusLine.method !== 'GET' && this.request.body.length !== contentLength) {
      return false;
    }

    return true;
  }

  sendResponse() {
    const response = { status: 200, statusMessage: 'OK', body: '' };

    if (this.request.statusLine.path === '/hello-world.html') {
      response.body = HELLO_WORLD;
    } else {
      response.status = 404;
      response.statusMessage = 'NOT FOUND';
      response.body = NOT_FOUND;
    }

    const serialized = 'HTTP/1.1 ${response.status} ${response.statusMessage}' + CRLF +
                       'Content-Length: ' + response.body.length + CRLF + CRLF +
                       response.body;
    this.connection.write(serialized);
  }

  handle(buffer) {
    this.parse(buffer);

    if (!this.requestComplete()) {
      return;
    }

    this.sendResponse();

    // Other-wise the connection may attempt to be re-used, we don't support this.
    this.connection.end();
  }
}

function handleConnection(connection) {
  const handler = new HTTPRequestHandler(connection);
  connection.on('data', (buffer) => handler.handle(buffer));
}

const server = net.createServer(handleConnection);

server.listen('9000');

So what's going on?

The protocol

HTTP (version 1.1, specifically) is a convention for connecting over TCP/IP and sending plain-text messages between two processes. HTTP messages are broken into two categories: requests (the sender of a request is called a "client") and responses (the sender of a response is called a "server").

HTTP is important because it is the default protocol of web browsers. When we type in localhost:9000/hello-world.html and hit enter, the browser will open an TCP/IP connection to the location localhost on the port 9000 and send an HTTP request. If/when it receives the HTTP response from the server it will try to render the response.

An HTTP request

A bare minimum HTTP/1.1 request (defined here) based on the request for localhost:9000/hello-world.html is the following:

GET /hello-world.html HTTP/1.1\r\nHost: localhost:9000\r\n\r\n

NOTE: the spec explicitly requires the \r\n combo to represent a newline instead of simply \n.

If we printed out this request it would look like this:

GET /hello-world.html HTTP/1.1
Host: localhost:9000

Components of an HTTP request

An HTTP/1.1 request is made up of a few parts:

The status line consists of the request method (e.g. GET, POST, PUT, etc.), the path for the request, and the protocol -- all separated by a space.

An HTTP header is a key-value pair separated by a colon. Spaces following the colon are ignored. The key is case insensitive. Only the Host header appears to be mandatory. Since these headers are sent by the client they are intended for the server's use.

The request body is text and is only relevant for requests of certain methods (e.g. POST but not GET).

An HTTP response

A bare minimum HTTP/1.1 response (defined here) based on the file we wanted to send back is the following:

HTTP/1.1 200 OK\r\n\r\n<html>\n  <body>\n    <h1>Hello world!</h1>\n  </body>\n</html>

If we printed out this response it would look like this:

HTTP/1.1 200 OK

<html>
  <body>
    <h1>Hello world!</h1>
  </body>
</html>

Components of an HTTP response

An HTTP/1.1 response is made up of a few parts:

The status line consists of the protocol, the status code, and the status message -- all separated by a space.

HTTP response headers are the same as HTTP request headers although in a response they are directives from the server to the client. There are many standard headers that are used for such things as setting cache rules, setting cookies, settings response type (e.g. HTML vs CSS vs PNG so the browser knows how to handle the response).

The response body is similar to the HTTP request body.

Sockets

Most operating systems have a built-in means of connecting over TCP/IP (and sending and receiving messages) called "sockets". Sockets allow us to treat TCP/IP connections like files in memory. Most programming languages have a built-in socket library. Node.js provides a high-level interface for listening on a port and handling new connections.

function handleConnection(connection) {
  connection.on('data', (buffer) => doSomething???);
}

const server = net.createServer(handleConnection);

server.listen('9000');

Once the program is listening, clients can open TCP/IP connections to the address (localhost) and port (9000) and our program takes over from there. Each connection is handled separately and receives "data" events. Each data event includes new bytes available for us to handle.

Let's encapsulate the state of each connection in HTTPRequestHandler class. Its function will be to parse data as it becomes available and respond to the request when the request is done.

class HTTPRequestHandler {
  constructor(connection) {
    this.connection = connection;
  }

  parse(buffer) {}

  requestComplete() {}

  sendResponse() {}

  handle(buffer) {
    this.parse(buffer);

    if (!this.requestComplete()) {
      return;
    }

    this.sendResponse();

    // Other-wise the connection may attempt to be re-used, we don't support this.
    this.connection.end();
  }
}

function handleConnection(connection) {
  const handler = new HTTPRequestHandler(connection);
  connection.on('data', (buffer) => handler.handle(buffer));
}

...

There are three functions we need to implement now: parse(buffer), requestComplete(), and sendResponse.

parse(buffer)

This function will be responsible for progressively pulling out data from the buffer. If the status line has not been received, it will try to grab the status line. If the body has not yet started, it will accumulate headers. Then it will continue accumulating the body until we close the connection (this will happen implicitly when requestComplete() returns true).

class HTTPRequestHandler {
  constructor(connection) {
    this.connection = connection;
    this.request = {
      statusLine: null,
      headers: {},
      body: null,
    };
  }

  parse(buffer) {
    const lines = buffer.toString().split(CRLF);

    // Parse/store status line if necessary
    if (!this.request.statusLine) {
      const [method, path, protocol] = lines.shift().split(' ');
      this.request.statusLine = { method, path, protocol };
    }

    // Parse/store headers if the body hasn't begun
    if (this.request.body === null) {
      for (let line = lines.shift(); lines.length; line = lines.shift()) {
        // Reached the end of headers, double CRLF
        if (line === '') {
          this.request.body = '';
          break;
        }

        const [key, value] = line.split(':');

        const safeKey = key.toLowerCase();
        if (!this.request.headers[safeKey]) {
          this.request.headers[safeKey] = [];
        }

        this.request.headers[safeKey].push(value.trimStart());
      }
    }

    this.request.body += lines.join(CRLF);
  }

...

}

requestComplete()

This function will look at the internal request state and return false if the status line has not been received, no headers have been received (although this is stricter than the HTTP/1.1 standard requires), or if the body length is not equal to the value of the Content-Length header.

class HTTPRequestHandler {

...

  requestComplete() {
    if (!this.request.statusLine || !Object.keys(this.request.headers).length || this.request.body === null) {
      return false;
    }

    const [contentLength] = this.request.headers['content-length'] || [];
    if (this.request.statusLine.method !== 'GET' && this.request.body.length !== contentLength) {
      return false;
    }

    return true;
  }

...

}

sendResponse()

Finally we'll hard-code two responses (one for the valid request for /hello-world.html and a catch-all 404 response for every other request). These responses need to be serialized according the HTTP response format described above and written to the connection.

class HTTPRequestHandler {

...

  sendResponse() {
    const response = { status: 200, statusMessage: 'OK', body: '' };

    if (this.request.statusLine.path === '/hello-world.html') {
      response.body = HELLO_WORLD;
    } else {
      response.status = 404;
      response.statusMessage = 'NOT FOUND';
      response.body = NOT_FOUND;
    }

    const serialized = 'HTTP/1.1 ${response.status} ${response.statusMessage}' + CRLF +
                       'Content-Length: ' + response.body.length + CRLF + CRLF +
                       response.body;
    this.connection.write(serialized);
  }  

...

}

Run it

Now that we've got all the pieces we can finally run the initial program:

$ node uweb.js &
$ open localhost:9000/hello-world.html

And we see the page! Try any other path and we receive a 404.

Review and next steps

We covered the basics of HTTP/1.1: a very simple, plain-text protocol oriented around requests and responses over a TCP/IP connection. We realize we need to know little about anything but parsing and formatting text on top of the TCP/IP blackbox called sockets. We created a simple application that returns different responses based on the request. But we're a far shot from a more general library, a web framework. Future posts will explore this transition as well as performance and more features.

Code is available on Github.

Share