The client role now automatically collects and merges searchDomains from ALL servers in the instance when not explicitly configured. This eliminates redundant configuration and ensures clients trust certificates from all servers. Also uses lib.mkIf with .exists check to safely handle the openssh-cert generator access, checking searchDomains first to enable lazy evaluation.
What it does
- Generates and persists SSH host keys via
vars. - Optionally issues CA-signed host certificates for servers.
- Installs the
serverCA public key intoclientsknown_hostsfor TOFU-less verification.
When to use it
- Zero-TOFU SSH for dynamic fleets: admins/CI can connect to frequently rebuilt hosts (e.g., server-1.example.com) without prompts or per-host
known_hostschurn.
Roles
- Server: runs sshd, presents a CA-signed host certificate for
<machine>.<domain>. - Client: trusts the CA for the given domains to verify servers' certificates. Tip: assign both roles to a machine if it should both present a cert and verify others.
Quick start (with host certificates) Useful if you never want to get a prompt about trusting the ssh fingerprint.
{
inventory.instances = {
sshd-with-certs = {
module = { name = "sshd"; input = "clan-core"; };
# Servers present certificates for <machine>.example.com
roles.server.tags.all = { };
roles.server.settings = {
certificate.searchDomains = [ "example.com" ];
# Optional: also add RSA host keys
# hostKeys.rsa.enable = true;
};
# Clients trust the CA for *.example.com
roles.client.tags.all = { };
roles.client.settings = {
certificate.searchDomains = [ "example.com" ];
};
};
};
}
Basic: only add persistent host keys (ed25519), no certificates Useful if you want to get an ssh "trust this server" prompt once and then never again.
{
inventory.instances = {
sshd-basic = {
module = {
name = "sshd";
input = "clan-core";
};
roles.server.tags.all = { };
};
};
}
Example: selective trust per environment Admins should trust only production; CI should trust prod and staging. Servers are reachable under both domains.
{
inventory.instances = {
sshd-env-scoped = {
module = { name = "sshd"; input = "clan-core"; };
# Servers present certs for both prod and staging FQDNs
roles.server.tags.all = { };
roles.server.settings = {
certificate.searchDomains = [ "prod.example.com" "staging.example.com" ];
};
# Admin laptop: trust prod only
roles.client.machines."admin-laptop".settings = {
certificate.searchDomains = [ "prod.example.com" ];
};
# CI runner: trust prod and staging
roles.client.machines."ci-runner-1".settings = {
certificate.searchDomains = [ "prod.example.com" "staging.example.com" ];
};
};
};
}
Explanation
- Admin -> server1.prod.example.com: zero-TOFU (verified via cert).
- Admin -> server1.staging.example.com: falls back to TOFU (or is blocked by policy).
- CI -> either prod or staging: zero-TOFU for both. Note: server and client searchDomains don't have to be identical; they only need to overlap for the hostnames you actually use.
Notes
- Connect using a name that matches a cert principal (e.g.,
server1.example.com); wildcards are not allowed inside the certificate. - CA private key stays in
vars(not deployed); only the CA public key is distributed. - Logins still require your user SSH keys on the server (passwords are disabled).