Docker Whitelister

What?

A whitelist proxy that uses socat. 🔌😼

Why?

tl;dr: To workaround https://github.com/moby/moby/issues/36174.

Basically, Docker supports internal networks; but when you use them, you simply cannot open ports from those services, which is not very convenient: you either have full or none isolation.

This proxy allows some whitelist endpoints to have network connectivity. It can be used for:

• Allowing connection only to some APIs, but not to the rest of the WWW.
• Exposing ports from a container while still not letting the container access the WWW.

How?

Use these environment variables:

TARGET

Required. It's the host name where the incoming connections will be redirected to.

HTTP_HEALTHCHECK

Default: 0

Set to 1 to enable healthcheck with pycurl http requests. This is useful if the target uses a deployment where the ip of the service gets changed frequently (e.g. accounts.google.com) and you are using PRE_RESOLVE

Automatically restarting unhealthy proxies

When you enable the http healthcheck the container marks itself as unhealthy but does nothing. (see https://github.com/moby/moby/pull/22719)

If you want to restart your proxies automatically, you can use https://github.com/willfarrell/docker-autoheal.

HTTP_HEALTHCHECK_URL

Default: http://$TARGET/

Url to use in HTTP_HEALTHCHECK if enabled. $TARGET gets replaced inside the url by the configured TARGET.

HTTP_HEALTHCHECK_TIMEOUT_MS

Default: 2000

Timeout in milliseconds for http healthcheck. This is used as a timeout for connecting and receiving an answer. You may end up with twice the time spend.

MODE

Default: tcp

Set to udp to proxy in UDP mode.

MAX_CONNECTIONS

Default: 100

Limits the maximum number of accepted connections at once per port.

Setting "unlimited" connections

For each port and open connection a subprocess is spawned. Setting a number too high might make your host system unresponsive and prevent you from logging in to it. So be very careful with setting this setting to a large number.

The typical linux system can handle up to 32768 so if you need a lot more parallel open connections make sure to also set the corresponding variables on your host system. See https://stackoverflow.com/questions/6294133/maximum-pid-in-linux for reference. And divide this number by at least the number of ports you are running through docker-whitelist.

What happens when the limit is hit?

docker-whitelist basically starts socat so the behaviour is the same. In case no more subprocesses can be forked:

• UDP mode: You won't see a difference on the connecting side. But no more packets are forwarded for new connections until the number of connections for this port is reduced.
• TCP mode: docker-whitelist no longer accepts the connection and your connection will wait until the number of connections for this port is reduced. Your connection may time out.

NAMESERVERS

Default: 208.67.222.222 8.8.8.8 208.67.220.220 8.8.4.4 to use OpenDNS and Google DNS resolution servers by default.

Only used when pre-resolving is enabled.

PORT

Default: 80 443. If you're proxying HTTP/S services, no need to specify!

The port where this service will listen, and where the target service is expected to be listening on also.

PRE_RESOLVE

Default: 0

Set to 1 to force using the specified nameservers to resolve the target before proxying.

This is especially useful when using a network alias to whitelist an external API.

SMTP_HEALTHCHECK

Default: 0

Set to 1 to enable healthcheck with pycurl smtp requests. This is useful if the target uses a deployment where the ip of the service gets changed frequently (e.g. smtp.eu.sparkpostmail.com) and you are using PRE_RESOLVE

Automatically restarting unhealthy proxies

see HTTP_HEALTHCHECK

SMTP_HEALTHCHECK_URL

Default: smtp://$TARGET/

Url to use in SMTP_HEALTHCHECK if enabled. $TARGET gets replaced inside the url by the configured TARGET.

SMTP_HEALTHCHECK_COMMAND

Default: HELP

Enables changing the healthcheck command for servers that do not support HELP (e.g. for MailHog you can use QUIT)

SMTP_HEALTHCHECK_TIMEOUT_MS

Default: 2000

Timeout in milliseconds for smtp healthcheck. This is used as a timeout for connecting and receiving an answer. You may end up with twice the time spend.

UDP_ANSWERS

Default: 1

1 means the process will wait for an answer from the server before the forked child process terminates (until this happens the connection counts towards the connection limit). Set to 0 if no answers are expected from the server, this prevents subprocesses waiting for an answer indefinitely.

Setting to 0 is recommended if you are using this to connect to a syslog server like graylog.

VERBOSE

Default: 0

Set to 1 to log all connections.

Example

So say you have a production app called coolapp that sends and reads emails, and uses Google Font APIs to render some PDF reports.

It is defined in a docker-compose.yaml file like this:

# Production deployment
version: "2.0"
services:
    app:
        image: Tecnativa/coolapp
        ports:
            - "80:80"
        environment:
            DB_HOST: db
        depends_on:
            - db

    db:
        image: postgres:alpine
        volumes:
            - dbvol:/var/lib/postgresql/data:z

volumes:
    dbvol:

Now you want to set up a staging environment for your QA team, which includes a fresh copy of the production database. To avoid the app to send or read emails, you put all into a safe internal network:

# Staging deployment
version: "2.0"
services:
    proxy:
        image: traefik
        networks:
            default:
            public:
        ports:
            - "8080:8080"
        volumes:
            # Here you redirect incoming connections to the app container
            - /etc/traefik/traefik.toml

    app:
        image: Tecnativa/coolapp
        environment:
            DB_HOST: db
        depends_on:
            - db

    db:
        image: postgres:alpine

networks:
    default:
        internal: true
    public:

Now, it turns out your QA detects font problems. Logic! app cannot contact fonts.google.com. Yikes! What to do? 🤷

tecnativa/whitelist to the rescue!! 💪🤠

# Staging deployment
version: "2.0"
services:
    fonts_googleapis_proxy:
        image: tecnativa/whitelist
        environment:
            TARGET: fonts.googleapis.com
            PRE_RESOLVE: 1 # Otherwise it would resolve to localhost
        networks:
            # Containers in default restricted network will ask here for fonts
            default:
                aliases:
                    - fonts.googleapis.com
            # We need public access to "open the door"
            public:

    fonts_gstatic_proxy:
        image: tecnativa/whitelist
        networks:
            default:
                aliases:
                    - fonts.gstatic.com
            public:
        environment:
            TARGET: fonts.gstatic.com
            PRE_RESOLVE: 1

    proxy:
        image: traefik
        networks:
            default:
            public:
        ports:
            - "8080:8080"
        volumes:
            # Here you redirect incoming connections to the app container
            - /etc/traefik/traefik.toml

    app:
        image: Tecnativa/coolapp
        environment:
            DB_HOST: db
        depends_on:
            - db

    db:
        image: postgres:alpine

networks:
    default:
        internal: true
    public:

And voilà! app has fonts, but nothing more. ✋👮

Development

All the dependencies you need to develop this project (apart from Docker itself) are managed with poetry.

To set up your development environment, run:

pip install pipx  # If you don't have pipx installed
pipx install poetry  # Install poetry itself
poetry install  # Install the python dependencies and setup the development environment

Testing

To run the tests locally, add --prebuild to autobuild the image before testing:

poetry run pytest --prebuild

By default, the image that the tests use (and optionally prebuild) is named test:docker-whitelist. If you prefer, you can build it separately before testing, and remove the --prebuild flag, to run the tests with that image you built:

docker image build -t test:docker-whitelist .
poetry run pytest

If you want to use a different image, pass the --image command line argument with the name you want:

# To build it automatically
poetry run pytest --prebuild --image my_custom_image

# To prebuild it separately
docker image build -t my_custom_image .
poetry run pytest --image my_custom_image