3.6 KiB
modules/openwrt/generator/
openwrt-gen — a small Rust CLI that converts a single JSON network description into OpenWrt
config files. Built by ../default.nix with rustPlatform.buildRustPackage and invoked by
the openwrt-deploy script.
Layout
generator/
├── Cargo.toml # crate metadata; deps: serde, serde_json
├── Cargo.lock # pinned for the Nix build (cargoLock.lockFile)
├── src/
│ ├── main.rs # CLI entry: reads JSON from stdin, dispatches on argv[1]
│ ├── config.rs # serde Deserialize structs for the input JSON (+ unit tests)
│ ├── network.rs # process_network — expands per-network device lists
│ └── generators/
│ ├── mod.rs # re-exports dnsmasq + ethers + firewall
│ ├── dnsmasq.rs # renders dnsmasq.conf (server= + address= lines)
│ ├── ethers.rs # renders /etc/ethers (MAC → name/IP)
│ └── firewall.rs # renders UCI firewall rules (internet-only networks)
└── target/ # gitignored Cargo build output — ignore it
Usage
openwrt-gen dnsmasq < config.json # prints dnsmasq.conf
openwrt-gen ethers < config.json # prints /etc/ethers
openwrt-gen firewall < config.json # prints UCI firewall rules (uci batch format)
Input shape (config.rs)
generalMappings:[{ ip, domains[] }]— shared domain→IP records.networks:{ <network>: [ { mac, name, id?, domains?, justMac?, internetOnly? } ] }— per-network hosts. A device withinternetOnly: trueis restricted to internet-only access (matched by MAC), on any network.dnsResolvers:[{ ip, port?, name? }]— upstreamserver=entries for dnsmasq.internetOnly:[<network>, ...]— networks restricted to internet-only access.
Internet-only firewall rules (firewall.rs)
The generator emits UCI config rule blocks that DROP forwarded traffic to the rest of the
LAN (10.0.0.0/16 IPv4, fd59:de0a:bff5::/48 IPv6). Two sources of restriction:
Network-level — for each network in internetOnly:
- IPv4 — one rule per network matching the whole subnet (catches every device on it).
- IPv6 — one rule per known device MAC. All subnets share a single ULA /64
(
fd59:de0a:bff5::/48, seenetwork::ULA_PREFIX), so subnets are not distinguishable in IPv6 and blocking is done per known device. An unknown IPv6-only device on those subnets is not blocked (inherent to the flat-L2 topology).
Device-level — for any device with internetOnly: true (on any network):
- IPv4 + IPv6 — one rule each matched by that device's MAC. This lets you restrict a
single machine (e.g. a printer on
lab) without restricting its whole subnet.
Rules are named skyg_<network>_drop_lan / skyg_<network>_drop_lan_v6 (network-level) and
skyg_mac_<mac>_drop_lan / skyg_mac_<mac>_drop_lan_v6 (device-level, MAC with colons
stripped) so the deploy script owns exactly these. Reverse direction (LAN → restricted
device) stays open, and within-subnet traffic (e.g. cam→cam) is blocked.
Conventions
- This crate has no I/O of its own beyond stdin/stdout — config decryption and SSH delivery are the deploy script's job. Keep it a pure JSON → text transformer.
config.rscarries unit tests; runcargo testhere when changing the parser. Building the Nix package is fine (it doesn't touch a live system), but is not required — the user validates deploys viaopenwrt-deploy.- Keep
Cargo.lockcommitted and in sync (the Nix build pins to it).