WIP janus secrets

This commit is contained in:
John Lancaster
2026-07-04 11:49:58 -05:00
parent cddc369687
commit 69f492c7cc
10 changed files with 318 additions and 58 deletions
@@ -0,0 +1,96 @@
---
description: "Use when modifying the Janus host, Janus step-ca deployment, Step SSH CA, mTLS issuance, CA bootstrap artifacts, or Janus SOPS secrets. Covers idiomatic Nix, secret placement, and troubleshooting."
name: "Janus Host Instructions"
applyTo: "modules/hosts/janus/**, modules/services/step-ca/step-ca.nix, modules/features/step-client.nix"
---
# Janus Host Instructions
Use this instruction when changing Janus host wiring, Janus CA material, or the shared Step CA module that Janus consumes.
## Host Intent And Boundaries
- Treat Janus as the homelab certificate authority host. It runs `step-ca` for X.509 issuance, SSH user and host certificates, and mTLS certificates.
- Assume Janus CA endpoints are private/LAN-only unless the task explicitly changes the exposure model.
- Keep Janus as a concrete host under `modules/hosts/janus`; keep reusable behavior in shared modules such as `modules/services/step-ca/step-ca.nix`, `modules/services/step-ca/ssh-host.nix`, `modules/features/step-client.nix`, and `modules/features/mtls.nix`.
- Preserve host identity unless the task explicitly renames the deployed CA: `hostname = "janus"`, `flake.nixosConfigurations."${hostname}"`, and `networking.hostName = hostname`.
- Keep Janus-specific paths, DNS names, IP SANs, and CA artifacts in the Janus host area or in Janus-specific option values. Do not move them into generic shared modules unless they become a deliberate module interface.
## Required Module Composition
When editing `modules/hosts/janus/default.nix`, preserve this composition unless the task explicitly changes Janus architecture:
- `nixos.lxc` for the container host shape.
- `nixos.mysops` with `sops.defaultSopsFile = ./secrets.yaml` for host-local encrypted CA runtime material.
- `nixos.step-ca` to run the CA daemon and render `/etc/smallstep/ca.json` from Nix plus SOPS paths.
- `nixos.step-client` to install Step trust bootstrap defaults and root CA material.
- `nixos.step-ssh-host` so Janus itself participates in SSH host certificate flow.
- `nixos.mtls` so Janus can issue and renew its own mTLS certificate bundle.
## Step CA Deployment Pattern
- Prefer structured Nix values rendered with `builtins.toJSON` over hand-written JSON strings for `ca.json`.
- Prefer the NixOS `services.step-ca.settings` interface for normal `ca.json` settings. Use a `sops.templates`-rendered config only when the rendered JSON needs sops-nix placeholders or runtime secret paths.
- Keep secret-bearing `ca.json` content rendered through `sops.templates` when it includes values from `config.sops.placeholder`; do not put decrypted private keys, passwords, or provisioner encrypted private-key values directly in the Nix store.
- Continue using `services.step-ca` for the daemon contract: `enable`, `settings`, `address`, `port`, `openFirewall`, `package`, and `intermediatePasswordFile`.
- Remember that the NixOS module overrides `services.step-ca.settings.address` with `services.step-ca.address` plus `services.step-ca.port`. Change the high-level address/port options when changing the listener.
- Keep the intermediate password as a runtime string path from `config.sops.secrets.<name>.path`; do not use a Nix path literal or the password file can be copied to the globally readable Nix store. Do not pass passwords through command-line arguments or environment variables.
- Account for upstream NixOS service behavior when debugging: the module writes `/etc/smallstep/ca.json`, starts `step-ca.service` as `Type=notify`, sets `DynamicUser = true`, uses `StateDirectory = "step-ca"`, and passes `intermediatePasswordFile` through systemd `LoadCredential`.
- If `step-ca` cannot read key files referenced by `ca.json`, check sops-nix owner/group/mode and systemd credential handling before loosening permissions. The password file path and CA signing key paths are separate contracts.
- Keep Step CA runtime state in `/var/lib/step-ca/db` unless the task explicitly migrates storage. If moving away from embedded Badger, document database backup and HA implications.
- Keep `root`, `crt`, `key`, `ssh.hostKey`, and `ssh.userKey` in `ca.json` pointing at public cert paths or sops-nix materialized secret paths as appropriate.
- Use an ACME provisioner for automated X.509 issuance, an `SSHPOP` provisioner for SSH certificate renewal/rekey behavior, and the `admin` JWK provisioner for explicit administrative or scripted issuance. Ensure JWK claims include `enableSSHCA = true` when it is expected to sign SSH certificates.
- Treat Nix and SOPS as the source of truth. If `step ca provisioner add`, `remove`, or `update` is used for investigation, copy the intended result back into Nix/SOPS and reload or restart `step-ca`; do not leave live `ca.json` drift on the host.
- Use `authority.claims` or provisioner-level claims for certificate lifetimes, renewal behavior, and SSH CA enablement. Avoid `allowRenewalAfterExpiry` unless the task explicitly accepts the added risk for intermittently connected clients.
- Add `authority.policy` when constraining which X.509 SANs or SSH principals Janus may issue. This is mandatory for any move away from private/LAN-only operation.
## CA Material And Secret Placement
Store each kind of material in the narrowest place that matches who needs it:
- `modules/hosts/janus/secrets.yaml`: Janus runtime CA secrets used by `step-ca`, including `janus/ca_password`, `janus/intermediate_ca_key`, `janus/ssh_host_ca_key`, `janus/ssh_user_ca_key`, and `janus/admin_provisioner_encrypted_key`.
- `keys/secrets.yaml`: shared provisioner credential `janus/admin_jwk`, because client and host certificate modules outside Janus consume it to request certs.
- `modules/hosts/janus/root_ca.crt`, `intermediate_ca.crt`, `fingerprint`, `ssh_user_ca_key.pub`, and `ssh_host_ca_key.pub`: public trust artifacts that may be committed and reviewed.
- Offline secret storage only: `root_ca_key` and the root key password. The running CA should not need the root private key for normal operation.
When editing secrets:
- Use SOPS commands (`sops`, `edit-secrets`, or `sops set`) rather than editing `ENC[...]` payloads.
- Use YAML literal blocks for PEM or OpenSSH private keys so newlines are preserved.
- After changing Janus recipient rules, run `sops --config .sops.yaml updatekeys -y modules/hosts/janus/secrets.yaml` so the host-local file is actually rewrapped.
- Keep `.sops.yaml` scoped: Janus host runtime secrets should match `modules/hosts/janus/secrets.yaml`; shared credentials should match `keys/secrets.yaml`.
## Public Trust Artifacts
- If `root_ca.crt` changes, update `fingerprint` from `step certificate fingerprint <root_ca.crt>` and update `modules/features/step-client.nix` defaults in the same change.
- If `intermediate_ca.crt` changes, confirm it still chains to the committed root and that the running CA uses the matching `intermediate_ca_key` from SOPS.
- If SSH CA keys rotate, update the matching public key files and every SSH trust consumer, including `modules/services/ssh.nix` known-host CA values and `TrustedUserCAKeys` behavior.
- Treat `README.md` bootstrap commands as part of the operational contract. Keep them aligned with secret names, public artifact paths, CA URL, DNS names, and IP SANs.
## mTLS And Client Trust
- Janus mTLS values should stay aligned with Step CA identity: update `step-ca.dnsNames`, `mtls.subject`, `mtls.san`, and Step client defaults together when CA names or addresses change.
- `modules/features/step-client.nix` is the shared bootstrap surface. Its default CA URL, root certificate, and fingerprint should describe Janus unless the repo intentionally migrates to another CA.
- mTLS certificate files are runtime outputs under the configured cert directory, not tracked source files. Do not store mTLS private keys in Git or the Nix store.
## Step/Nix Troubleshooting Checklist
Start with the smallest check that matches the change:
1. Host evaluation: `nix eval .#nixosConfigurations.janus.config.system.build.toplevel.drvPath`
2. Build planning: `nix build .#nixosConfigurations.janus.config.system.build.toplevel --no-link --dry-run`
3. Secret contract: confirm every `step-ca.secrets.*` value has a matching `sops.secrets` declaration and a matching key in `modules/hosts/janus/secrets.yaml`.
4. Rendered config: on the host, inspect `/etc/smallstep/ca.json` or `/etc/step-ca` paths without copying secret content into logs.
5. Service health: check `systemctl status step-ca.service`, `journalctl -u step-ca.service`, and `step ca health --ca-url https://janus.john-stream.com` after trust is bootstrapped.
6. Provisioners: use `step ca provisioner list` to confirm ACME, SSHPOP, and admin JWK availability.
7. SSH cert flow: use `ssh-host-cert-check`, `systemctl status step-ssh-host-renew.service`, and `systemctl status step-ssh-host-renew.timer`.
8. mTLS flow: use `mtls-check`, `mtls-generate`, and `systemctl status mtls-renew.timer` for host certificate renewal issues.
## Change Safety Rules
- Do not deploy a root private key to Janus unless the task explicitly changes the trust model.
- Do not rename `janus/admin_jwk` without updating all consumers in the same change.
- Do not expose ACME or `/provisioners` endpoints publicly without an explicit migration away from the private/LAN-only model plus policy and endpoint exposure review.
- Do not use certificate templates casually; if templates are added, keep them narrow and use `toJson` for user-controlled values.
- If changing certificate durations, prefer short-lived leaf certs with automated renewal over long-lived leaf certs.
- If changing service ports or proxying, account for Step renewal behavior: mTLS renewal is sensitive to layer-7 proxies unless configured for token-based renewal.