A Docker firewall for your reverse proxy network
Why do you need something like TrafficJam? Reverse proxies are often used to authenticate external access to internal services, providing benefits such as centralized user management, access control, 2FA and SSO. In a typical Docker setup, multiple services are connected to the reverse proxy via a single network. If a user authenticates to one service and is able to compromise that service (such as by using this Pi-Hole vulnerability), that user will gain access to the entire network behind the reverse proxy, and can access every service on the network whether they would normally have permission or not.
Potential solutions include:
- Use each service's own authentication
- Not all services provide 2FA :(
- Many services do not support centralized user management (LDAP) :(
- Many services do not support SSO :(
- Have each service on a unique network
- Reverse proxy network connections must be updated every time a service is added or removed :(
- Manually configuring every service and reverse proxy entry is painful and error-prone even with tools like Ansible :(
- Use a reverse proxy with auto-discovery and a firewall to isolate services
- Enables 2FA, LDAP, ACL, SSO, etc. regardless of service support :)
- Routes are automatically discovered by the proxy without manual configuration :)
- Every service only needs a connection to one network :)
TrafficJam allows you to safely and easily connect all of your backend containers to your reverse proxy using a single docker network by preventing the backend containers from communicating with each other.
TrafficJam works by adding some firewall (iptables
) rules to the docker network you specify. First, it blocks all traffic on the network. Then it adds a rule that only allows traffic to/from the container(s) you specify in the whitelist. It continually monitors the docker network to make sure the rules stay up to date as you add or remove containers.
docker-cli
:
docker run \
--name trafficjam \
--cap-add NET_ADMIN \
--network host \
--volume "/var/run/docker.sock:/var/run/docker.sock" \
--env NETWORK=traefik_public \
--env WHITELIST_FILTER="ancestor=traefik:latest" \
--env TZ="America/Los_Angeles" \
--detach \
kaysond/trafficjam
docker-compose.yml
:
version: '3.8'
services:
trafficjam:
container_name: trafficjam
image: kaysond/trafficjam
cap_add:
- NET_ADMIN
network_mode: host
volumes:
- /var/run/docker.sock:/var/run/docker.sock
environment:
NETWORK: traefik_public
WHITELIST_FILTER: ancestor=traefik:latest
TZ: America/Los_Angeles
traefik:
container_name: traefik
image: traefik:latest
networks:
traefik_public:
whoami:
container_name: whoami
image: traefik/whoami
networks:
traefik_public:
networks:
traefik_public:
docker-cli
:
docker service create \
--name trafficjam \
--mount type=bind,source=/var/run/docker.sock,destination=/var/run/docker.sock \
--env NETWORK=traefik_public \
--env WHITELIST_FILTER=ancestor=traefik:v2.3.7@sha256:0aca29bb8e51aa69569b15b8b7f08328e6957cbec201dd532304b3329e5a82a9 \
--env SWARM_DAEMON=true \
--env TZ=America/Los_Angeles \
--replicas 1 \
--constraint node.role==manager \
kaysond/trafficjam
docker-compose.yml
:
version: '3.8'
services:
trafficjam:
image: trafficjam
volumes:
- /var/run/docker.sock:/var/run/docker.sock
environment:
NETWORK: traefik_network
WHITELIST_FILTER: ancestor=traefik:v2.3.7@sha256:0aca29bb8e51aa69569b15b8b7f08328e6957cbec201dd532304b3329e5a82a9
SWARM_DAEMON: "true"
TZ: America/Los_Angeles
deploy:
replicas: 1
placement:
constraints: ['node.role==manager']
Notes:
Docker Swarm services tag images with a sha256 hash to guarantee that every node runs the exact same container (since tags are mutable). When using the ancestor
tag, ensure that the appropriate hash is included as shown in the examples.
trafficjam
requires the NET_ADMIN
Linux capability in order to manipulate iptables
rules. For Docker Swarm setups, SYS_ADMIN
is also required in order to enter namespaces, though the setting of container capabilities is automatically handled by the trafficjam
swarm daemon.
TrafficJam is configured via several environment variables:
- NETWORK - The name of the Docker network this instance of TrafficJam should manage (multiple instances can be run for different networks)
- WHITELIST_FILTER - A Docker
--filter
parameter that designates which containers should be permitted to openly access the network. See Docker Docs - filtering - TZ - Timezone (for logging)
- INSTANCE_ID - A unique alphanumeric instance ID that is required to run multiple instances of trafficjam
- SWARM_DAEMON - Setting this variable is required for swarm and activates a daemon that determines network load balancer IP addresses and properly configures the trafficjam service
- SWARM_IMAGE - The image the trafficjam swarm daemon should deploy (defaults to
kaysond/trafficjam
). The best practice is to pin this to a particular image hash (e.g.kaysond/trafficjam:v1.0.0@sha256:8d41599fa564e058f7eb396016e229402730841fa43994124a8fb3a14f1a9122
) - POLL_INTERVAL - How often TrafficJam checks Docker for changes
- ALLOW_HOST_TRAFFIC - Allow containers to initiate communication with the docker host, and thus any port-mapped containers. Most users do not need this setting enabled. (See ARCHITECTURE.md). Note that if this setting is enabled while old rules exist, some will not be cleared automatically and must be done so manually (See Clearing Rules).
- DEBUG - Setting this variable turns on debug logging
- Linux with iptables whose version is compatible with the iptables in TrafficJam (currently
1.8.7 (legacy)
or1.8.7 (nf_tables)
) - Docker >20.10.0
trafficjam
can be run with the --clear
argument to remove all rules that have been set. Note that the host docker socket must be mounted within the container. The rules can also be cleared by sending the SIGUSR1
signal to the container. This will cause trafficjam
to exit.
Examples:
docker run --volume "/var/run/docker.sock:/var/run/docker.sock" --cap-add NET_ADMIN --network host kaysond/trafficjam --clear
docker kill --signal SIGUSR1 trafficjam