Files
mailsink/README.md
T
jcoffey-dev 92b8688033 Add mailsink, a discard-only SMTP server for test networks
Accepts mail for an allowlist of domains from allowlisted client networks
and discards it on receipt. No outbound code, no logging, scratch image;
all settings are Dockerfile build arguments compiled into the binary.
2026-09-24 12:36:16 -07:00

63 lines
2.7 KiB
Markdown

# mailsink
An SMTP server for test and development networks. Applications on other
machines send mail to it as they would to a real server. It accepts mail for
the domains you allow and discards it.
- **Nothing is kept.** Message bodies are read off the socket and discarded
as they arrive. They are never held in memory as a whole or written to
disk, and the container's filesystem is read-only.
- **Nothing is relayed.** The server contains no code that opens an outbound
connection, and the runtime image has only its one static binary: no MTA,
no shell, no network tools. Mail to a domain outside the allowlist gets
`550 5.7.1` at `RCPT TO`.
- **Only local clients.** Connections from addresses outside
`ALLOWED_NETWORKS` get `554` and are closed.
- **No logging.** The server writes nothing while it runs, and the compose
file sets Docker's logging driver to `none`.
## Configure
All settings are build arguments at the top of the `Dockerfile`, with
comments explaining each one. They are compiled into the binary, so the
running container reads no environment variables or config files. To change
a setting, edit the Dockerfile and rebuild. A bad value fails the build.
| Setting | Default |
|---|---|
| `ALLOWED_DOMAINS` | `example.test,*.example.test` (`*` accepts any domain) |
| `ALLOWED_NETWORKS` | the RFC 1918 ranges and `fc00::/7` |
| `SMTP_HOSTNAME` | `mailsink.test` |
| `SMTP_PORT` | `25` |
| `MAX_MESSAGE_SIZE` | `26214400` (25 MiB) |
| `MAX_RECIPIENTS` | `100` |
| `MAX_CONNECTIONS` | `100` |
## Run
```sh
docker compose up -d --build
sudo ./egress-lockdown.sh # optional second layer; see below
```
Point applications at the Docker host on port 25. Any username and password
are accepted, so apps configured for authenticated SMTP work unchanged. There
is no TLS: clients must allow a plaintext connection.
`ALLOWED_NETWORKS` is checked against the client's source address. Clients on
the network reach the container through Docker's NAT, which keeps their real
address. A connection from the Docker host itself goes through Docker's
userland proxy, so it appears to come from the bridge gateway (in
`172.16.0.0/12`). If you narrow `ALLOWED_NETWORKS`, tests run on the host
itself will be refused.
## Blocking egress at the firewall
Docker can't publish a port from an `internal` network, so the container sits
on a normal bridge (`br-mailsink`). `egress-lockdown.sh` adds iptables rules
that drop every new connection starting from that bridge, both out to the
network and to the Docker host itself. Inbound SMTP and its replies still
pass. Run it as root after the container is up, and again after a reboot.
`--remove` takes the rules out. The script assumes Docker's iptables firewall
backend.