Install NGINX

 

Background to the recommended installation

Using NGINX mainline

We recommend using NGINX mainline - by default, most Linux distributions will instead install NGINX stable. While the stable branch will receive security updates, the version number will not always reflect the latest published version of NGINX, and scanning tools and other security compliance frameworks in your organisation may deem this as a failure to patch to the latest version.

Using a custom error page

We recommend installing custom error pages for common HTTP errors. Without this, certain errors are handled by NGINX, and this can reveal the version number of NGINX; other errors are handled by Tomcat, and this can reveal information about the version of Tomcat in use, and a stack trace for the error.

The instructions below are for a simple configuration showing a text error message for all 4xx and 5xx HTTP errors.

  • If you want to add details such as the web address, phone number or email address for your own support desk, update the text in the custom error file custom_error.html below.

  • If you wish to show different pages for different errors, or to include images in your error pages, configuring NGINX to do this is well documented on the web.

In the phixflow.conf file provided below, the following lines specify the errors to be handed using a custom page, and the location of the page. Update these if you want to configure something different:

proxy_intercept_errors on; error_page 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 421 422 423 424 425 426 428 429 431 451 500 501 502 503 504 505 506 507 508 510 511 /custom_error.html; location = /custom_error.html { internal; root /usr/share/nginx/html; }

Including a redirect to connections to port 80

Leaving port 80 open and including a redirect is widely recommended (including, for example, by Let’s Encrypt: https://letsencrypt.org/docs/allow-port-80/). This is done in the NGINX configuration file below. This means that you will need to allow incoming connections through your firewall to port 80 on the server you are using to host NGINX. Doing this also allows Let’s Encrypt to automatically renew the certificate using the default HTTP-01 challenge.

Hide version number

In this installation, we hide the version number on the default NGINX page. This is a useful security measure. Because errors are already redirected to a standard error page (see above), in this installation you are unlikely to see this in a browser. But this can be viewed using a client such as curl that does not apply the port 80 redirect, or by querying for the header information for the service.

Using Let’s Encrypt to provide certificates

The method below uses https://certbot.eff.org/ to issue and install certificates from https://letsencrypt.org/ to provide secure connections over HTTPS. Let’s Encrypt is a well known Certificate Authority (CA) that is free to use. Check with your organisation on standards for certificates. You may need to use a different public CA, or an internal CA. Using CAs aside from Let’s Encrypt is well documented on the web. In particular, most public CAs will document how their certificates can be installed into NGINX.

If you do not use Certbot, skip the section ‘Install certificate using Certbot’ below, and follow instructions appropriate for your CA.

Setting client_max_body_size

The client_max_body_size is set in the configuration below to allow files of sufficient size to be loaded into PhixFlow. (Note, this is specified in the "Content-Length" request header field.) In particular, this setting must allow configuration files to be uploaded into PhixFlow. Configuration files are used to move applications developed in the PhixFlow GUI between PhixFlow instances (e.g. from a test to a production system).

If PhixFlow users encounter problems loading files into PhixFlow, you may need to further increase this setting. Only authenticated users are able to trigger a file upload, and even then only certain users will have access to funtions that cause a file upload. Even in an instance of PhixFlow that is available through the public internet, general users who are not athenticated on PhixFlow are not able to upload arbitrary files into PhixFlow.

Installing with apt

The instructions below are based on installation on a Debian-based distribution of Linux, and use the apt command. If you are installing on a RedHat-based distribution of Linux, the equivalent yum commands for NGINX installation are well documented on the web.

Single server, single PhixFlow webapp

The following instructions assume that NGINX is installed on the same server as PhixFlow itself (i.e. the same server as the Tomcat installation), and with a single installation of PhixFlow (a “webapp”). If you have multiple webapps on a single server, multiple webapps across several servers, or a single webapp on a different server from the reverse proxy, see ‘Multiple PhixFlow webapps or multiple servers’ below.

Set up a domain record

Choose a domain, e.g. phixflow.mycompany.com, and set up an A record on your DNS to point this domain at the public IP address of the server that you are installing NGINX on.

Install NGINX

Install NGINX from the repository. Instructions for various Linux distributions can be found here: http://nginx.org/en/linux_packages.html.

The example given here is taken from those instructions, and is for installation of NGINX mainline on Ubuntu.

Install prerequisites:

sudo apt install curl gnupg2 ca-certificates lsb-release ubuntu-keyring

Import the official NGINX signing key so that the package can be verified:

curl https://nginx.org/keys/nginx_signing.key | gpg --dearmor | sudo tee /usr/share/keyrings/nginx-archive-keyring.gpg >/dev/null

Verify the downloaded file:

The output should contain the full fingerprint 573BFD6B3D8FBC641079A6ABABF5BD827BD9BF62 as follows:

Add the mainline packages to your repository list:

Set up repository pinning:

Install NGINX:

Install certificate using Certbot

The following instructions are based on the recommended installation of Certbot with snap: https://certbot.eff.org/instructions?ws=nginx&os=ubuntubionic. This link also includes alternative instructions for installing Certbot if you don’t want to use snap.

These instructions also assume that snap is installed on your system. If it is not, you can following instructions here: https://snapcraft.io/docs/installing-snapd.

Make sure that any firewalls on the server are open to port 80 (HTTP).

Ensure that snapd is up to date:

Install Certbot:

Link Certbot command to make it accessible:

Request a certificate:

When the final command runs, it will prompt for several responses:

  • An email address (for urgent renewal and security notices): e.g. security.notifications@mycompany.com

  • Terms of service: you must agree to these (press Y)

  • Agreement to share your email address with EFF: you can choose either option, i.e. Y or N

  • The domain name assigned to this service: e.g. phixflow.mycompany.com

You can run sudo certbot certificates to list the installed certificates and confirm that the certificate has been successfully requested

Configure NGINX

Most distributions of NGINX no longer include sites-enabled and sites-available directories. This example installation is based on distribution that does not include these directories, and places the configuration file in the /etc/nginx/conf.d directory. If your installed version of NGINX includes sites-enabled and sites-available directories, consult the NGINX documentation for further guidance.

Drop HTTP v1.0 traffic: in the configuration below, we drop HTTP v1.0 traffic - see comment in the configuration. As noted, there is a possible very minor impact if you are building a website that you want to be discoverable on all search engines, but this is unlikely to affect major search engines. We have taken the option to drop HTTP v1.0 traffic because this is very straightforward; the alternative configuration to allow HTTP v1.0 but mitigate specific risks is more complex, and in our opinion, overall a bigger risk to reliably providing the service because it is more prone to being configured incorrectly or having unintended consequences.

Save a backup of the default configuration file:

Create configuration file for PhixFlow: open editing on a file /etc/nginx/conf.d/phixflow.conf (e.g. with sudo nano /etc/nginx/conf.d/phixflow.conf), and paste in the following, replacing [domain] with your domain, e.g. phixflow.mycompany.com:

Create a custom error page: open editing on a file /usr/share/nginx/html/custom_error.html (e.g. with sudo nano /usr/share/nginx/html/custom_error.html), and paste in the following:

Hide version number: edit the file /etc/nginx/nginx.conf, and in the http context add the line server_tokens off;. This will result in a file that looks similar to:

Restart NGINX:

Notes

  • You may need to reboot the server as well in order for NGINX to restart successfully.

  • Run netstat -tln to check the server is listening on port 443 and port 80.

  • Check the PhixFlow application loads in the browser.

Configure SSL cipher restriction

This is based on recommendations given at: https://wiki.mozilla.org/Security/Server_Side_TLS#Recommended_configurations.

Open the file at /etc/letsencrypt/options-ssl-nginx.conf. It should look similar to the following:

openssl-1.1.1 and later

At version 1.1.1 OpenSSL changed the method of configuring ciphersuites for TLS1.3. This has an impact on configuration of ciphersuites in NGINX (https://trac.nginx.org/nginx/ticket/1529​).

To determine if you are using OpenSSL at a version of 1.1.1 or later, take note of the version of libssl on your linux distribution rather than openssl itself. On Debian based distributions of linux, you can find these packges with the command dpkg --list | grep ssl (this will probably show other ssl related packages, but you can ignore them for this assessment).

E.g. from Ubuntu 18.04:

In practice on Ubuntu, for example, on 22.04 the newer version of OpenSSL is used; on 18.04, the older version.

If you aren’t sure, try the configuration in this section, if you get an error when trying to start NGINX, try the other configuration below.

Edit the ssl_protocols parameter to be:

Update the file to replace the current line that starts ssl_ciphers to be:

Save and close the file.

Before openssl-1.1.1

Edit the ssl_protocols parameter to be:

Edit the ssl_ciphers parameter to be:

Save and close the file.

Restart NGINX

Run the following to reload the NGINX configuration:

Restart the NGINX service:

Multiple PhixFlow webapps or multiple servers

A single reverse proxy can handle connections to different webapps (i.e. installations of PhixFlow), whether these are hosted on a single server, or across several servers. See *** for common network configuration for PhixFlow installation. However, whichever configuration you choose, the configuration is the same: you need an IP address and a webapp name to define each PhixFlow installation.

Note that, since NGINX is acting as the the point of SSL termination, if you specify a PhixFlow installation at an IP address you must make sure that the route from the reverse proxy to the PhixFlow installation is secure. The suggested configurations in *** will provide this since traffic is sent within a private network.

In the following example, a single reverse proxy on a dedicated server is handling incoming connections to four different webapps:

System name

Server IP address

Webapp name

System name

Server IP address

Webapp name

Production

10.154.0.12

prod, installed at: /opt/tomcat/webapps/prod

UAT

10.154.0.13

uat, installed at: /opt/tomcat/webapps/uat

Dev 1

10.154.0.14

dev1, installed at /opt/tomcat/webapps/dev1

Dev 2

10.154.0.14

dev2, installed at /opt/tomcat/webapps/dev2

To support connections to all these systems replace the directive

in the example phixflow.conf file above with: