Access your docker socket safely as read-only, rootless and distroless
These are the main tags for the image. There is also a tag for each commit and its shorthand sha256 value.
What can I do with this? This image will run a proxy to access your docker socket as read-only. The exposed proxy socket is run as 1000:1000, not as root, although the image starts the proxy process as root to interact with the actual docker socket. There is also a TCP endpoint started at 2375 that will also proxy to the actual docker socket if needed. It is not exposed by default and must be exposed via using - "2375:2375/tcp"
in your compose.
Make sure that the docker socket is accessible by the user:
specification in your compose, if the UID/GID are not correct, the image will print out the correct UID/GID for you to set:
socket-proxy-1 | 2025/03/26 10:16:33 can’t access docker socket as GID 0 owned by GID 991
socket-proxy-1 | please change the user setting in your compose to the correct UID/GID pair like this:
socket-proxy-1 | services:
socket-proxy-1 | socket-proxy:
socket-proxy-1 | user: "0:991"
Why should I run this image and not the other image(s) that already exist? Good question! All the other images on the market that do exactly the same don’t do or offer these options:
Important
- This image runs the proxy part as a specific UID/GID (not root), most other images run everything as root
- This image uses a single binary, most other images use apps like Nginx or HAProxy (bloat)
- This image has no shell since it is 100% distroless, most other images run on a distro like Debian or Alpine with full shell access (security)
- This image does not ship with any critical or high rated CVE and is automatically maintained via CI/CD, most other images mostly have no CVE scanning or code quality tools in place
- This image is created via a secure, pinned CI/CD process and immune to upstream attacks, most other images have upstream dependencies that can be exploited
- This image contains a proper health check that verifies the app is actually working, most other images have either no health check or only check if a port is open or ping works
- This image exposes the socket as a UNIX socket and TCP socket, most other images only expose it via a TCP socket
- This image works as read-only, most other images need to write files to the image filesystem
If you value security, simplicity and the ability to interact with the maintainer and developer of an image. Using my images is a great start in that direction.
name: "traefik" # this is a compose example for Traefik
services:
socket-proxy:
image: "11notes/socket-proxy:2.1.2"
read_only: true
user: "0:0" # make sure to use the same UID/GID as the owner of your docker socket!
volumes:
- "/run/docker.sock:/run/docker.sock:ro" # mount host docker socket, the :ro does not mean read-only for the socket, just for the actual file
- "socket-proxy:/run/proxy" # this socket is run as 1000:1000, not as root!
restart: "always"
traefik:
image: "11notes/traefik:3.2.0"
depends_on:
socket-proxy:
condition: "service_healthy"
restart: true
command:
- "--global.checkNewVersion=false"
- "--global.sendAnonymousUsage=false"
- "--api.dashboard=true"
- "--api.insecure=true"
- "--log.level=INFO"
- "--log.format=json"
- "--providers.docker.exposedByDefault=false" # use docker provider but do not expose by default
- "--entrypoints.http.address=:80"
- "--entrypoints.https.address=:443"
- "--serversTransport.insecureSkipVerify=true" # do not verify downstream SSL certificates
ports:
- "80:80/tcp"
- "443:443/tcp"
- "8080:8080/tcp"
networks:
frontend:
backend:
volumes:
- "socket-proxy:/var/run"
sysctls:
net.ipv4.ip_unprivileged_port_start: 80
restart: "always"
nginx: # example container
image: "11notes/nginx:1.26.2"
labels:
- "traefik.enable=true"
- "traefik.http.routers.default.priority=1"
- "traefik.http.routers.default.rule=PathPrefix(`/`)"
- "traefik.http.routers.default.entrypoints=http"
- "traefik.http.routers.default.service=default"
- "traefik.http.services.default.loadbalancer.server.port=8443"
- "traefik.http.services.default.loadbalancer.server.scheme=https" # proxy from http to https since this image runs by default on https
networks:
backend: # allow container only to be accessed via traefik
restart: "always"
volumes:
socket-proxy:
networks:
frontend:
backend:
internal: true
Parameter | Value | Default |
---|---|---|
TZ |
Time Zone | |
DEBUG |
Will activate debug option for container image and app (if available) | |
SOCKET_PROXY_VOLUME |
path to the docker volume used to expose the prox socket | /run/proxy |
SOCKET_PROXY_DOCKER_SOCKET |
path to the actual docker socket | /run/docker.sock |
SOCKET_PROXY_UID |
the UID used to run the proxy parts | 1000 |
SOCKET_PROXY_GID |
the GID used to run the proxy parts | 1000 |
Important
This image is not based on another image but uses scratch as the starting layer.
Tip
- Use a reverse proxy like Traefik, Nginx, HAproxy to terminate TLS and to protect your endpoints
- Use Let’s Encrypt DNS-01 challenge to obtain valid SSL certificates for your services
This image is provided to you at your own risk. Always make backups before updating an image to a different version. Check the releases for breaking changes. If you have any problems with using this image simply raise an issue, thanks. If you have a question or inputs please create a new discussion instead of an issue. You can find all my other repositories on github.
created 31.03.2025, 21:27:39 (CET)