SSO as a service (SSOaaS)

Our concept of SSOaaS

Access management provides 3 services:

  • Global Authentication: Single Sign-On
  • Authorization: to grant authentication is not enough. User rights must be checked
  • Accounting: SSO logs (access) + application logs (transactions and results)

LL::NG affords all these services (except application logs of course, but headers are provided to permit this).

Headers setting is an another LL::NG service. LL::NG can provide any user attributes to an application (see Rules and headers)

*aaS means that application can drive underlying layer (IaaS for infrastructure, PaaS for platform,…). So for us, SSOaaS must provide the ability for an app to manage authorizations and choose user attributes to set. Authentication can not be really “*aaS”: app must just use it, not manage it.

LL::NG affords some features that can be used to provide SSO as a service: a web application can manage its rules and headers. Docker or VM images (Nginx only) includes LL::NG Nginx configuration that aims to a global LL::NG authorization server. By default, all authenticated users can access and one header is set: Auth-User. If application gives a RULES_URL parameter that refers to a JSON file, authorization server will read it, apply specified rules and set required headers (see DevOps Handler).

There are two different architectures to do this:

In both case, Handler type must be set to DevOps.

Examples of webserver configuration for Docker/VM images

Using a global FastCGI (or uWSGI) server

Nginx

In this example, web server templates (Nginx only) are configured to request authorization from a central FastCGI server:

test-nginx.conf
server {
  server_name myapp.domain.com;
  location = /lmauth {
    internal;
    include /etc/nginx/fastcgi_params;
    # Pass authorization requests to Central FastCGI server:
    fastcgi_pass 10.1.2.3:9090;
    fastcgi_param VHOSTTYPE DevOps;
    # Drop post datas
    fastcgi_pass_request_body  off;
    fastcgi_param CONTENT_LENGTH "";
    # Keep original hostname
    fastcgi_param HOST $http_host;
    # Keep original request (LLNG server will received /lmauth)
    fastcgi_param X_ORIGINAL_URI  $request_uri;
 
    # Set dynamically rules (LLNG will poll it every 10 mn)
    fastcgi_param RULES_URL http://rulesserver/my.json
  }
  location /rules.json {
    auth_request off;
    allow 10.1.2.3;
    deny all;
  }
  location ~ ^(.*\.php)$ {
    auth_request /lmauth;
    auth_request_set $lmremote_user $upstream_http_lm_remote_user;
    auth_request_set $lmlocation $upstream_http_location;
    error_page 401 $lmlocation;
    include /etc/lemonldap-ng/nginx-lua-headers.conf;
    ...
    # Example with php-fpm:
    include snippets/fastcgi-php.conf;
    fastcgi_pass unix:/var/run/php/php7.0-fpm.sock;
  }
  location / {
    try_files $uri $uri/ =404;
  }
}

Apache

There is an experimental FactCGI client in LLNG. You just have to install FCGI::Client and add this in your apache2.conf:

apache2.conf
<VirtualHost ...>
    PerlHeaderParserHandler Lemonldap::NG::Handler::ApacheMP2::FCGIClient
    PerlSetVar LLNG_SERVER 127.0.0.1:9090
    PerlSetVar VHOSTTYPE DevOps
    PerlSetVar RULES_URL http://app.tld/rules.json
    ...
</VirtualHost>

Node.js

Using express and fastcgi-authz-client, you can protect also an Express server. Example:

app.js
var express = require('express');
var app = express();
var FcgiAuthz = require('fastcgi-authz-client');
var handler = FcgiAuthz({
  host: '127.0.0.1',
  port: 9090,
  PARAMS: {
    RULES_URL: 'http://my-server/rules.json'
  }
});
 
app.use(handler);
 
// Simple express application
app.get('/', function(req, res) {
  return res.send('Hello ' + req.upstreamHeaders['auth-user'] + ' !');
});
 
// Launch server
app.listen(3000, function() {
  return console.log('Example app listening on port 3000!');
});

Plack application

You just have to enable Plack::Middleware::Auth::FCGI. Simple example:

app.psgi
use Plack::Builder;
 
my $app   = sub {
  my $env = shift;
  my $user = $env->{fcgiauth-auth-user};
  return [ 200, [ 'Content-Type' => 'text/plain' ], [ "Hello $user" ] ];
};
 
# Optionally ($fcgiResponse is the PSGI response of remote FCGI auth server)
#sub on_reject {
#    my($self,$env,$fcgiResponse) = @_;
#    my $statusCode = $fcgiResponse->{status};
#    ...
#}
 
builder
{
  enable "Auth::FCGI",
    host => '127.0.0.1',
    port => '9090',
    fcgi_auth_params => {
      RULES_URL => 'https://my-server/my.json',
    },
    # Optional rejection subroutine
    #on_reject => \&on_reject;
    ;
  $app;
};

Using front reverse-proxies

This is a simple Nginx configuration file. It looks like a standard LL::NG nginx configuration file except for:

  • VHOSTTYPE parameter forced to use DevOps handler
  • /rules.json must not be protected by LL::NG but by the web server itself

This configuration handles *.dev.sso.my.domain URL and forwards authenticated requests to <vhost>.internal.domain. Rules can be defined in /rules.json which is located at the website root directory.

test-nginx.conf
server {
  server_name "~^(?<vhost>.+?)\.dev\.sso\.my\.domain$";
  location = /lmauth {
    internal;
    include /etc/nginx/fastcgi_params;
    fastcgi_pass unix:/home/xavier/dev/lemonldap/e2e-tests/conf/llng-fastcgi.sock;
    # Force handler type:
    fastcgi_param VHOSTTYPE DevOps;
    # Drop post datas
    fastcgi_pass_request_body  off;
    fastcgi_param CONTENT_LENGTH "";
    # Keep original hostname
    fastcgi_param HOST $http_host;
    # Keep original request (LLNG server will received /lmauth)
    fastcgi_param X_ORIGINAL_URI  $request_uri;
  }
  location /rules.json {
    auth_request off;
    allow 127.0.0.0/8;
    deny all;
  }
  location / {
    auth_request /lmauth;
    auth_request_set $lmremote_user $upstream_http_lm_remote_user;
    auth_request_set $lmlocation $upstream_http_location;
    error_page 401 $lmlocation;
    include /etc/lemonldap-ng/nginx-lua-headers.conf;
    proxy_pass https://$vhost.internal.domain;
  }
}