goldilocks.js/lib/tcp/http.js

'use strict';

module.exports.create = function (deps, conf, tcpMods) {
  var PromiseA = require('bluebird');
  var statAsync = PromiseA.promisify(require('fs').stat);
  var domainMatches = require('../domain-utils').match;
  var separatePort = require('../domain-utils').separatePort;

  function parseHeaders(conn, opts) {
    // There should already be a `firstChunk` on the opts, but because we might sometimes
    // need more than that to get all the headers it's easier to always read the data off
    // the connection and put it back later if we need to.
    opts.firstChunk = Buffer.alloc(0);

    // First we make sure we have all of the headers.
    return new PromiseA(function (resolve, reject) {
      if (opts.firstChunk.includes('\r\n\r\n')) {
        resolve(opts.firstChunk.toString());
        return;
      }

      var errored = false;
      function handleErr(err) {
        errored = true;
        reject(err);
      }
      conn.once('error', handleErr);

      function handleChunk(chunk) {
        if (!errored) {
          opts.firstChunk = Buffer.concat([opts.firstChunk, chunk]);
          if (!opts.firstChunk.includes('\r\n\r\n')) {
            conn.once('data', handleChunk);
            return;
          }

          conn.removeListener('error', handleErr);
          conn.pause();
          resolve(opts.firstChunk.toString());
        }
      }
      conn.once('data', handleChunk);
    }).then(function (firstStr) {
      var headerSection = firstStr.split('\r\n\r\n')[0];
      var lines = headerSection.split('\r\n');
      var result = {};

      lines.slice(1).forEach(function (line) {
        var match = /([^:]*?)\s*:\s*(.*)/.exec(line);
        if (match) {
          result[match[1].toLowerCase()] = match[2];
        } else {
          console.error('HTTP header line does not match pattern', line);
        }
      });

      var match = /^([a-zA-Z]+)\s+(\S+)\s+HTTP/.exec(lines[0]);
      if (!match) {
        throw new Error('first line of "HTTP" does not match pattern: '+lines[0]);
      }
      result.method = match[1].toUpperCase();
      result.url = match[2];

      return result;
    });
  }

  function hostMatchesDomains(req, domainList) {
    var host = separatePort((req.headers || req).host).host.toLowerCase();

    return domainList.some(function (pattern) {
      return domainMatches(pattern, host);
    });
  }

  function determinePrimaryHost() {
    var result;
    if (Array.isArray(conf.domains)) {
      conf.domains.some(function (dom) {
        if (!dom.modules || !dom.modules.http) {
          return false;
        }
        return dom.names.some(function (domain) {
          if (domain[0] !== '*') {
            result = domain;
            return true;
          }
        });
      });
    }
    if (result) {
      return result;
    }

    if (Array.isArray(conf.http.modules)) {
      conf.http.modules.some(function (mod) {
        return mod.domains.some(function (domain) {
          if (domain[0] !== '*') {
            result = domain;
            return true;
          }
        });
      });
    }
    return result;
  }

  // We handle both HTTPS and HTTP traffic on the same ports, and we want to redirect
  // any unencrypted requests to the same port they came from unless it came in on
  // the default HTTP port, in which case there wont be a port specified in the host.
  var redirecters = {};
  var ipv4Re = /^\d{1,3}\.\d{1,3}\.\d{1,3}\.\d{1,3}$/;
  var ipv6Re = /^\[[0-9a-fA-F:]+\]$/;
  function redirectHttps(req, res) {
    var host = separatePort(req.headers.host);

    if (!redirecters[host.port]) {
      redirecters[host.port] = require('redirect-https')({ port: host.port });
    }

    // localhost and IP addresses cannot have real SSL certs (and don't contain any useful
    // info for redirection either), so we direct some hosts to either localhost.daplie.me
    // or the "primary domain" ie the first manually specified domain.
    if (host.host === 'localhost') {
      req.headers.host = 'localhost.daplie.me' + (host.port ? ':'+host.port : '');
    }
    // Test for IPv4 and IPv6 addresses. These patterns will match some invalid addresses,
    // but since those still won't be valid domains that won't really be a problem.
    if (ipv4Re.test(host.host) || ipv6Re.test(host.host)) {
      var dest;
      if (conf.http.primaryDomain) {
        dest = conf.http.primaryDomain;
      } else {
        dest = determinePrimaryHost();
      }
      if (dest) {
        req.headers.host = dest + (host.port ? ':'+host.port : '');
      }
    }

    redirecters[host.port](req, res);
  }

  function emitConnection(server, conn, opts) {
    server.emit('connection', conn);

    // We need to put back whatever data we read off to determine the connection was HTTP
    // and to parse the headers. Must be done after data handlers added but before any new
    // data comes in.
    process.nextTick(function () {
      conn.unshift(opts.firstChunk);
      conn.resume();
    });

    // Convenience return for all the check* functions.
    return true;
  }

  var acmeServer;
  function checkAcme(conn, opts, headers) {
    if (headers.url.indexOf('/.well-known/acme-challenge/') !== 0) {
      return false;
    }

    if (deps.tunnelServer.isClientDomain(separatePort(headers.host).host)) {
      deps.tunnelServer.handleClientConn(conn);
      process.nextTick(function () {
        conn.unshift(opts.firstChunk);
        conn.resume();
      });
      return true;
    }

    if (!acmeServer) {
      acmeServer = require('http').createServer(tcpMods.tls.middleware);
    }
    return emitConnection(acmeServer, conn, opts);
  }

  function checkLoopback(conn, opts, headers) {
    if (headers.url.indexOf('/.well-known/cloud-challenge/') !== 0) {
      return false;
    }
    return emitConnection(deps.ddns.loopbackServer, conn, opts);
  }

  var httpsRedirectServer;
  function checkHttps(conn, opts, headers) {
    if (conf.http.allowInsecure || conn.encrypted) {
      return false;
    }
    if (conf.http.trustProxy && 'https' === headers['x-forwarded-proto']) {
      return false;
    }

    if (!httpsRedirectServer) {
      httpsRedirectServer = require('http').createServer(redirectHttps);
    }
    return emitConnection(httpsRedirectServer, conn, opts);
  }

  var adminDomains;
  var adminServer;
  function checkAdmin(conn, opts, headers) {
    var host = separatePort(headers.host).host;

    if (!adminDomains) {
      adminDomains = require('../admin').adminDomains;
    }
    if (adminDomains.indexOf(host) !== -1) {
      if (!adminServer) {
        adminServer = require('../admin').create(deps, conf);
      }
      return emitConnection(adminServer, conn, opts);
    }

    if (deps.tunnelServer.isAdminDomain(host)) {
      deps.tunnelServer.handleAdminConn(conn);
      process.nextTick(function () {
        conn.unshift(opts.firstChunk);
        conn.resume();
      });
      return true;
    }

    return false;
  }

  var proxyServer;
  function createProxyServer() {
    var http = require('http');
    var agent = new http.Agent();
    agent.createConnection = deps.net.createConnection;

    var proxy = require('http-proxy').createProxyServer({
      agent: agent
    , toProxy: true
    });

    proxy.on('error', function (err, req, res) {
      res.statusCode = 502;
      res.setHeader('Connection', 'close');
      res.setHeader('Content-Type', 'text/html');
      res.end(tcpMods.proxy.getRespBody(err, conf.debug));
    });

    proxyServer = http.createServer(function (req, res) {
      proxy.web(req, res, req.connection.proxyOpts);
    });
    proxyServer.on('upgrade', function (req, socket, head) {
      proxy.ws(req, socket, head, socket.proxyOpts);
    });
  }
  function proxyRequest(mod, conn, opts, xHeaders) {
    if (!proxyServer) {
      createProxyServer();
    }

    conn.proxyOpts = {
      target: 'http://'+(mod.address || (mod.host || 'localhost')+':'+mod.port)
    , headers: xHeaders
    };
    return emitConnection(proxyServer, conn, opts);
  }

  function proxyWebsocket(mod, conn, opts, headers, xHeaders) {
    var index = opts.firstChunk.indexOf('\r\n\r\n');
    var body = opts.firstChunk.slice(index);

    var head = opts.firstChunk.slice(0, index).toString();
    var headLines = head.split('\r\n');
    // First strip any existing `X-Forwarded-*` headers (for security purposes?)
    headLines = headLines.filter(function (line) {
      return !/^x-forwarded/i.test(line);
    });
    // Then add our own `X-Forwarded` headers at the end.
    Object.keys(xHeaders).forEach(function (key) {
      headLines.push(key + ': ' +xHeaders[key]);
    });
    // Then convert all of the head lines back into a header buffer.
    head = Buffer.from(headLines.join('\r\n'));

    opts.firstChunk = Buffer.concat([head, body]);

    var newConnOpts = separatePort(mod.address || '');
    newConnOpts.port = newConnOpts.port || mod.port;
    newConnOpts.host = newConnOpts.host || mod.host || 'localhost';
    newConnOpts.servername = separatePort(headers.host).host;
    newConnOpts.data = opts.firstChunk;

    newConnOpts.remoteFamily  = opts.family  || conn.remoteFamily;
    newConnOpts.remoteAddress = opts.address || conn.remoteAddress;
    newConnOpts.remotePort    = opts.port    || conn.remotePort;

    tcpMods.proxy(conn, newConnOpts, opts.firstChunk);
  }

  function checkProxy(mod, conn, opts, headers) {
    var xHeaders = {};
    // Then add our own `X-Forwarded` headers at the end.
    if (conf.http.trustProxy && headers['x-forwarded-proto']) {
      xHeaders['X-Forwarded-Proto'] = headers['x-forwarded-proto'];
    } else {
      xHeaders['X-Forwarded-Proto'] = conn.encrypted ? 'https' : 'http';
    }
    var proxyChain = (headers['x-forwarded-for'] || '').split(/ *, */).filter(Boolean);
    proxyChain.push(opts.remoteAddress || opts.address || conn.remoteAddress);
    xHeaders['X-Forwarded-For'] = proxyChain.join(', ');
    xHeaders['X-Forwarded-Host'] = headers.host;

    if ((headers.connection || '').toLowerCase() === 'upgrade') {
      proxyWebsocket(mod, conn, opts, headers, xHeaders);
    } else {
      proxyRequest(mod, conn, opts, xHeaders);
    }
    return true;
  }

  function checkRedirect(mod, conn, opts, headers) {
    if (!mod.fromRe || mod.fromRe.origSrc !== mod.from) {
      // Escape any characters that (can) have special meaning in regular expression
      // but that aren't the special characters we have interest in.
      var from = mod.from.replace(/[-\/\\^$+?.()|[\]{}]/g, '\\$&');
      // Then modify the characters we are interested in so they do what we want in
      // the regular expression after being compiled.
      from = from.replace(/\*/g, '(.*)');
      var fromRe = new RegExp('^' + from + '/?$');
      fromRe.origSrc = mod.from;
      // We don't want this property showing up in the actual config file or the API,
      // so we define it this way so it's not enumberable.
      Object.defineProperty(mod, 'fromRe', {value: fromRe, configurable: true});
    }

    var match = mod.fromRe.exec(headers.url);
    if (!match) {
      return false;
    }

    var to = mod.to;
    match.slice(1).forEach(function (globMatch, index) {
      to = to.replace(':'+(index+1), globMatch);
    });
    var status = mod.status || 301;
    var code = require('http').STATUS_CODES[status] || 'Unknown';

    conn.end([
      'HTTP/1.1 ' + status + ' ' + code
    , 'Date: ' + (new Date()).toUTCString()
    , 'Location: ' + to
    , 'Connection: close'
    , 'Content-Length: 0'
    , ''
    , ''
    ].join('\r\n'));
    return true;
  }

  var staticServer;
  var staticHandlers = {};
  var indexHandlers = {};
  function serveStatic(req, res) {
    var rootDir = req.connection.rootDir;
    var modOpts = req.connection.modOpts;

    if (!staticHandlers[rootDir]) {
      staticHandlers[rootDir] = require('express').static(rootDir, {
        dotfiles: modOpts.dotfiles
      , fallthrough: false
      , redirect: modOpts.redirect
      , index: modOpts.index
      });
    }

    staticHandlers[rootDir](req, res, function (err) {
      function doFinal() {
        if (err) {
          res.statusCode = err.statusCode;
        } else {
          res.statusCode = 404;
        }
        res.setHeader('Content-Type', 'text/html');

        if (res.statusCode === 404) {
          res.end('File Not Found');
        } else {
          res.end(require('http').STATUS_CODES[res.statusCode]);
        }
      }

      var handlerHandle = rootDir
        + (modOpts.hidden||'')
        + (modOpts.icons||'')
        + (modOpts.stylesheet||'')
        + (modOpts.template||'')
        + (modOpts.view||'')
        ;

      function pathMatchesUrl(pathname) {
        if (req.url === pathname) {
          return true;
        }
        if (0 === req.url.replace(/\/?$/, '/').indexOf(pathname.replace(/\/?$/, '/'))) {
          return true;
        }
      }
      if (!modOpts.indexes || ('*' !== modOpts.indexes[0] && !modOpts.indexes.some(pathMatchesUrl))) {
        doFinal();
        return;
      }

      if (!indexHandlers[handlerHandle]) {
        // https://www.npmjs.com/package/serve-index
        indexHandlers[handlerHandle] = require('serve-index')(rootDir, {
          hidden: modOpts.hidden
        , icons: modOpts.icons
        , stylesheet: modOpts.stylesheet
        , template: modOpts.template
        , view: modOpts.view
        });
      }

      indexHandlers[handlerHandle](req, res, function (_err) {
        err = _err || err;

        doFinal();
      });

    });
  }
  function checkStatic(modOpts, conn, opts, headers) {
    var rootDir = modOpts.root.replace(':hostname', separatePort(headers.host).host);
    return statAsync(rootDir)
      .then(function (stats) {
        if (!stats || !stats.isDirectory()) {
          return false;
        }

        if (!staticServer) {
          staticServer = require('http').createServer(serveStatic);
        }
        conn.rootDir = rootDir;
        conn.modOpts = modOpts;
        return emitConnection(staticServer, conn, opts);
      })
      .catch(function (err) {
        if (err.code !== 'ENOENT') {
          console.warn('errored stating', rootDir, 'for serving static files', err);
        }
        return false;
      })
      ;
  }

  // The function signature is as follows
  // function module(moduleOptions, tcpConnection, connectionOptions, headers) { ... }
  var moduleChecks = {
    proxy:    checkProxy
  , redirect: checkRedirect
  , static:   checkStatic
  };

  function handleConnection(conn) {
    var opts = conn.__opts;
    parseHeaders(conn, opts)
      .then(function (headers) {
        if (checkAcme(conn, opts, headers))  { return; }
        if (checkLoopback(conn, opts, headers))  { return; }
        if (checkHttps(conn, opts, headers)) { return; }
        if (checkAdmin(conn, opts, headers)) { return; }

        var prom = PromiseA.resolve(false);
        (conf.domains || []).forEach(function (dom) {
          prom = prom.then(function (handled) {
            if (handled) {
              return handled;
            }
            if (!dom.modules || !dom.modules.http) {
              return false;
            }
            if (!hostMatchesDomains(headers, dom.names)) {
              return false;
            }

            var subProm = PromiseA.resolve(false);
            dom.modules.http.forEach(function (mod) {
              if (moduleChecks[mod.type]) {
                subProm = subProm.then(function (handled) {
                  if (handled) { return handled; }
                  return moduleChecks[mod.type](mod, conn, opts, headers);
                });
              } else {
                console.warn('unknown HTTP module under domains', dom.names.join(','), mod);
              }
            });
            return subProm;
          });
        });
        (conf.http.modules || []).forEach(function (mod) {
          prom = prom.then(function (handled) {
            if (handled) {
              return handled;
            }
            if (!hostMatchesDomains(headers, mod.domains)) {
              return false;
            }

            if (moduleChecks[mod.type]) {
              return moduleChecks[mod.type](mod, conn, opts, headers);
            }
            console.warn('unknown HTTP module found', mod);
          });
        });


        prom.then(function (handled) {
          // XXX TODO SECURITY html escape
          var host = (headers.host || '[no host header]').replace(/</, '&lt;');
          // TODO specify filepath of config file or database connection, etc
          var msg = "Bad Gateway: Goldilocks accepted '" + host + "' but no module (neither static nor proxy) was designated to handle it. Check your config file.";
          if (!handled) {
            conn.end([
              'HTTP/1.1 502 Bad Gateway'
            , 'Date: ' + (new Date()).toUTCString()
            , 'Content-Type: text/html'
            , 'Content-Length: ' + msg.length
            , 'Connection: close'
            , ''
            , msg
            ].join('\r\n'));
          }
        });
      })
      ;
  }

  return {
    emit: function (type, value) {
      if (type === 'connection') {
        handleConnection(value);
      }
    }
  };
};