Give every member crate its own `[lints.clippy]` (`pedantic`, `todo`, `unwrap_used`) and `[lints.rust]` block, add explicit `release` and `production` profiles to the workspace manifest, and fix the code so that `just clippy` -- `cargo clippy --workspace --all-targets --all-features -- -D warnings` -- passes with zero warnings. Declaring the lints is the easy half and on its own it was actively harmful: `unsafe_code = "forbid"` cannot be overridden from source, so it broke `lanparty-gateway`, `lanparty-client-tap`, and `lanparty-client-route`, which all need OS FFI, and the workspace no longer compiled on Linux at all. The rest of the workspace produced roughly 120 denied lints. A lint policy nobody can satisfy is worse than no policy, so this commit makes the tree actually clean. Unsafe policy: the three FFI crates use `unsafe_code = "deny"`, and the one module in each that holds the FFI (`gateway::packet`, `client_tap::windows`, `client_route::windows`) opts back in with a documented `#![allow(unsafe_code)]`. Every other crate keeps `"forbid"`. `deny` was chosen over dropping the lint so that `unsafe` outside those modules is still a hard error, and over per-block allows because the FFI is dense enough that per-block attributes would drown the code. Build profiles: `release` stays optimized but debuggable (debug info, debug assertions, overflow checks, no LTO, incremental) so that a bug reproduced at close-to-real speed still panics loudly and gives a usable backtrace. `production` inherits from it and turns all of that off, adding fat LTO and a single codegen unit, and is what ships. Note the consequence: `cargo build --release` binaries are now slower than before this commit and must not be shipped; use `just build-production`. Code changes made to satisfy the lints, grouped by kind: - `# Errors` sections on every public fallible function, and `# Panics` on `RoomRegistry::new`. The Windows-only implementations and their non-Windows `bail!` stubs are documented in parallel so the crates stay clean when built for Windows too, which CI on Linux cannot check. - Panicking paths removed rather than documented where the panic was only an unreachable invariant: `OverlayHeader::decode` and `declared_payload_len` take a fixed-size prefix via `first_chunk` instead of `try_into().expect(...)`, and `RoomRegistry::join` returns an `InternalError` reject instead of `expect`ing the room it just inserted. - Tests keep using `unwrap`/`unwrap_err`: each crate root carries `#![cfg_attr(test, allow(clippy::unwrap_used))]`. The lint is about production code; an `expect` message per assertion buys nothing. - Lossy casts replaced with `try_from` where a real conversion was happening (datagram-size negotiation, control-frame length prefix, MTU clamping). In `gateway::packet` the cast lints are allowed module-wide instead: those casts move values between libc's C types and ours after the value is already known to fit, so fallible conversions would only add unreachable branches. - Signature changes, all on private or crate-internal items except one: `RoomRegistry::join` now takes `&EndpointHello` (public, it never consumed the hello), `Room::join` and `reject_control_error` likewise. `Room`'s `room_id` field is now `id`. - `#[allow(clippy::too_many_lines)]` with a reason on three end-to-end test scenarios and on `bridge_until_shutdown`, whose `select!` loop mutates state shared by every arm; splitting it would hide that. - Mechanical fixes from `cargo clippy --fix`: `map_or_else`, `let...else`, inline format args, backticks in doc comments, `Duration::from_mins`. No runtime behavior changes: the only observable differences are the error text when a relay room lookup fails immediately after insertion (previously a panic) and the profile rename described above. Also documents the lint policy and the profile table in README.md, whose build section now points at the `just` recipes it should have used all along. Test Plan: - `just clippy` -- passes, zero warnings - `just test` -- 186 tests pass, 0 failed - `cargo check --profile production --workspace --all-features` -- passes - `just fmt` -- clean - Not verified: the Windows-only code paths and their doc comments, which need a Windows target to compile. Refs: https://doc.rust-lang.org/cargo/reference/profiles.html Refs: https://rust-lang.github.io/rust-clippy/master/index.html#missing_errors_doc Refs: https://doc.rust-lang.org/reference/attributes/diagnostics.html#lint-check-attributes
softlan-vpn
Monorepo for a Layer 2 over QUIC LAN party bridge.
Workspace crates
lanparty-proto: shared frame format, MAC validation, MTU helpers.lanparty-ctrl: control-plane messages (join/hello/role/version).lanparty-net: shared relay endpoint parsing and resolution.lanparty-obs: shared diagnostics/logging event models.lanparty-client-core: platform-agnostic client session state.lanparty-client-route: Windows relay-route inspection.lanparty-client-tap: TAP-Windows6 adapter discovery and frame I/O.lanparty-client-win: Windows TAP + route/metric handling binary.lanparty-gateway: Linux AF_PACKET gateway binary.lanparty-relay: public QUIC relay binary.
lanparty-proto
Transport-agnostic tunnel contract shared by all binaries:
- overlay datagram header encoding and decoding
- v1 overlay datagrams reject reserved nonzero flags until their semantics are defined
- negotiated QUIC datagram budget validation before send
- Ethernet frame header parsing
- MAC address parsing and identity validation
- QUIC datagram to TAP MTU budget helpers
lanparty-ctrl
Reliable control-plane schema shared by the QUIC stream handlers:
- endpoint hello messages with role, room, MAC, and datagram budget
- server welcome mode, reject, peer lifecycle, stats, and disconnect messages
- initial room gateway-presence status and gateway peer id in server welcomes
- room-code, role/MAC, peer-id, and effective-MTU validation
- length-prefixed JSON control frames for reliable QUIC streams
lanparty-obs
Shared diagnostics and structured logging vocabulary:
- client/gateway/relay frame logs with MACs, ethertype, length, peer, and action
- tunnel counters shared by control messages and runtime diagnostics
- client connectivity/TAP diagnostics and user-facing status messages
lanparty-net
Shared network address handling for tunnel binaries:
- relay DNS name, IP literal, and socket-address parsing
- UDP/443 default for bare relay hosts
- relay address resolution before tunnel interface activation
lanparty-client-core
Platform-neutral remote client relay session:
- relay QUIC connection with pinned relay certificate trust
- client hello with room, virtual MAC, and datagram budget
- welcome/reject handling with assigned peer id and effective TAP MTU
- QUIC DATAGRAM support and negotiated datagram budget diagnostics
- relay RTT diagnostics from the active QUIC connection
- reliable relay control-event reads for peer lifecycle messages
- Ethernet frame send/receive helpers over QUIC DATAGRAM with budget, source MAC, and remote-to-LAN safety checks plus local drop outcomes
- client tunnel statistics for frame/datagram rx/tx and drops
- reliable client stats snapshot sends for relay diagnostics
- best-effort graceful disconnect messages before QUIC close
lanparty-client-route
Windows route-table boundary:
- read-only best-route lookup for a relay destination IP
- selected source address, next hop, interface index/LUID, prefix, and metric
- interface index/LUID lookup from Windows network adapter GUIDs
- scoped IP interface MTU overrides with restore-on-drop behavior
- scoped IP interface metric overrides with restore-on-drop behavior
- scoped default-route suppression with restore-on-drop behavior
- unicast IP address snapshots for TAP diagnostics
- scoped host-route pinning for the relay IP on the pre-TAP interface
- host-route pin matching for relay-route verification after TAP activation
- reuse of an already-existing matching relay host route without deleting it on exit
- non-Windows builds return a clear unsupported-platform error
lanparty-client-tap
Windows TAP adapter boundary:
- TAP-Windows6 adapter discovery from the Windows network adapter registry
- TAP
NetworkAddressregistry configuration for the tunnel MAC identity \\.\Global\{NetCfgInstanceId}.tapdevice path construction- blocking Ethernet frame reads/writes through the TAP device handle
- oversized TAP read buffering so jumbo frames can be counted and dropped
- TAP driver IOCTL helpers for media status, adapter MAC, and MTU
lanparty-relay
Public relay binary and relay-owned room state:
- QUIC endpoint binding and first-stream hello/welcome admission
- room admission for clients and gateways
- one gateway per room, duplicate client MAC rejection, and room limits
- stable effective room MTU chosen before Ethernet datagrams flow
- live Ethernet datagram forwarding with no ingress reflection
- forwarding drops for Ethernet frames above the negotiated TAP MTU
- per-peer egress budget checks against the negotiated datagram size
- reliable
PeerJoined/PeerLeftnotifications plus gateway identity in welcome messages - L2 safety filters for invalid-source, jumbo, switch-control, remote VLAN tags, remote IPv6 fragments, IPv4/IPv6 DHCP-server, and IPv6-RA frames, including frames behind ordinary IPv6 extension headers
- client broadcast/multicast, unknown-unicast, and total bandwidth limiting
- malformed peer datagram disconnect threshold
- peer stats control events retained for relay diagnostics
- graceful disconnect control events propagated as peer-leave reasons
- per-peer last-seen timestamps in relay room snapshots
- peer leave cleanup for room membership and MAC indexes
Build And Local Checks
just fmt
just test
just clippy
just build-release
git diff --check
just clippy must stay completely clean: every crate turns on
clippy::pedantic, clippy::todo, and clippy::unwrap_used, and the recipe
runs with -D warnings over all targets and features. Crates that need OS FFI
(lanparty-gateway, lanparty-client-tap, lanparty-client-route) set
unsafe_code = "deny" instead of "forbid", and the single module holding
their FFI opts back in with #![allow(unsafe_code)]; every other crate forbids
unsafe outright.
These checks cover the local Rust code and the real client/relay/gateway session paths that can run without Windows TAP or LAN hardware. For the Windows client build and the manual MVP end-to-end proof, see TESTING.md.
Build Profiles
| Profile | Command | Purpose |
|---|---|---|
dev |
just build |
Everyday development. |
release |
just build-release |
Optimized but debuggable: debug info, debug assertions and overflow checks on, no LTO. |
production |
just build-production |
Shipping builds: no debug info, stripped, no debug assertions, fat LTO, one codegen unit. |
Use release when reproducing a bug at close-to-real speed, and production
for anything handed to someone else.
Relay
cargo run -p lanparty-relay -- --listen 443/udp --dev-cert-der-out relay-cert.der
--listen accepts either a socket address or a UDP port shorthand such as
443/udp. The relay binds a QUIC endpoint, accepts a control-stream hello,
replies with welcome or reject, and forwards live Ethernet QUIC datagrams
between accepted peers in the same room. It currently uses a generated
self-signed development certificate; --dev-cert-der-out writes that
certificate so the gateway and client can pin it in development. Production
certificate handling remains future work. Ethernet forwarding decisions are
logged with room, peer, MAC, ethertype, action, drop reason, and target count.
Safety-policy rejects use the filtered action so they are distinguishable from
malformed/unknown-destination drops and rate limits. Malformed peer datagrams
log their per-peer count before the relay disconnects peers that cross the
malformed-datagram threshold. Relay egress skips caused by a target peer's
smaller datagram budget are logged with the ingress peer, target peer, encoded
length, and target budget. Ingress datagrams larger than the sending peer's
negotiated datagram budget are dropped before decode/forwarding and logged with
reason=datagram_budget. Unknown unicast from a client is forwarded only to the
gateway port; unknown unicast from the gateway is dropped instead of flooded to
every remote client. When a peer joins or leaves, the relay sends a reliable
lifecycle control event to peers that are still present in the room. Newly
joined peers also receive PeerJoined events for peers that were already
present, and catch-up delivery is part of the accepted handshake rather than a
best-effort follow-up. When a client joins, the relay notifies existing peers
before the client receives its welcome, so gateways can seed client MAC state
before that client starts sending frames. When a gateway joins, the relay gives
the gateway the current client list before notifying clients that the gateway is
available.
MVP Trust Model
The MVP relay terminates QUIC for every client and gateway connection. QUIC protects traffic on the public network path, but the relay process sees plaintext Ethernet frames while forwarding them between peers in a room. That is acceptable for the first LAN-party proof, where the relay is an operator-trusted component, but it is not end-to-end encrypted.
Future room-key payload encryption should keep the relay-visible routing header small and leave only Ethernet payload bytes encrypted end-to-end between clients and the LAN gateway.
Gateway
cargo run -p lanparty-gateway -- \
--relay lanparty-relay.local \
--server-name lanparty-relay.local \
--relay-ca-cert relay-cert.der \
--room ROOM1 \
--interface eth0
The gateway first opens the wired LAN interface as an AF_PACKET socket with
promiscuous packet membership, then connects to the relay as role = gateway
and completes the control-stream hello/welcome handshake. That startup order
keeps an invalid, wireless, or unplugged interface from briefly advertising a
gateway that cannot bridge. Once both sides are ready, it bridges Ethernet
frames between the relay and wired LAN until shutdown. It captures whole LAN
frames up to the overlay payload-length ceiling before deciding whether they fit
the tunnel. It never fragments Ethernet frames; LAN frames with invalid source
MACs, L2 control-plane traffic, jumbo frames, frames above the negotiated TAP
MTU, or encoded datagrams exceeding the negotiated QUIC budget are counted,
dropped, and logged locally instead of stopping the bridge or consuming relay
bandwidth. Remote frames received from the relay are safety-checked again before
LAN injection and must use the announced virtual MAC for their source peer, so
invalid-source, forged-source, L2 control-plane, remote VLAN, DHCP-server, IPv6
Router Advertisement, IPv6 fragment, jumbo, and over-TAP-MTU frames cannot cross
the gateway's final physical-LAN boundary even if they reached the gateway over
QUIC. --relay accepts a DNS name or socket address; bare hosts default to
UDP/443. The gateway rejects Linux interfaces that sysfs identifies as Wi-Fi,
and rejects wired interfaces whose sysfs carrier state reports no link; managed
wireless NICs are not supported for the physical LAN bridge. It tracks
remote-client MACs from relay lifecycle events and periodically emits small CAM
refresh frames, logged with reason=periodic, so the physical switch keeps
those MACs associated with the gateway port. A newly observed client also
triggers an immediate CAM refresh frame logged with reason=peer_joined instead
of waiting for the first periodic refresh tick. When control events and frame
work are both ready, the bridge handles the lifecycle event first so first
packets after a client joins use the freshest remote-MAC state available
locally. Gateway frame logs include direction, peer id when present, MACs,
ethertype/length, frame length, action, and drop reason. The gateway also tracks
frame/datagram counters and periodically sends stats snapshots to the relay.
Malformed or runt LAN frames are counted and logged as dropped instead of
disappearing before accounting. It drops unrelated LAN unicast locally once the
destination is known not to be a connected remote client, so busy LAN traffic is
not sent to the public relay just to be discarded there. Relay lifecycle events
seed and retire remote-client MACs for CAM refresh and LAN-destination filtering
even before that client sends traffic. On shutdown, the gateway sends a
best-effort disconnect control message before closing QUIC so the relay can
report the intended reason.
Windows Client
cargo run -p lanparty-client-win -- \
--relay lanparty-relay.local \
--server-name lanparty-relay.local \
--relay-ca-cert relay-cert.der \
--room ROOM1
The Windows client binary is runtime-gated to Windows because the real client
path depends on TAP-Windows6 and Windows route protection. Non-Windows builds
are useful for type checking, but they fail before tunnel setup instead of
joining a room without a TAP adapter. On Windows, the binary connects to the
relay as role = client with a generated locally administered virtual MAC
persisted in lanparty-client-identity.json. Before resolving or connecting to
the relay, it writes the generated tunnel MAC to the selected TAP driver's
NetworkAddress registry setting and marks TAP media disconnected. That clears
stale connected state from a previous crashed run without letting the TAP
adapter influence relay DNS or route selection. The client then resolves the
relay endpoint, pins a host route for the resolved relay IP on the current
pre-TAP interface, verifies that Windows is using that host route, completes the
control-stream hello/welcome handshake, verifies the host route again after TAP
activation, and bridges Ethernet frames between the relay and the TAP-Windows6
adapter until shutdown. --relay accepts a DNS name or socket address; bare
hosts default to UDP/443. TAP frames whose source MAC does not match that
generated tunnel MAC are dropped locally before they can consume relay
bandwidth; the relay still enforces the same source-MAC rule. If the exact relay
host route already exists, the client uses it and leaves it alone on exit. The
startup status reports whether the relay already has a LAN gateway for the room.
--virtual-mac can still override the stored identity for manual testing. On
Windows it sets the TAP IP interface MTU to the relay-selected MTU, marks the
TAP media connected for the scoped client run, and reports the driver MAC/MTU
before forwarding frames, along with the TAP interface index/LUID. The client
applies a scoped TAP interface metric and disables TAP default routes while it
runs, periodically rechecks that the relay route remains pinned, then restores
the previous route policy and TAP media status on exit. Startup prints a warning
when TAP default routes were enabled before the scoped protection was applied.
Startup still fails before bridging if the driver-reported MAC does not match
the tunnel identity, because an already-initialized Windows TAP adapter may need
to be disabled/enabled or reinstalled before it reloads the configured
NetworkAddress. If exactly one TAP-Windows6 adapter is installed, the client
opens it automatically. If multiple TAP-Windows6 adapters are installed, startup
fails until --tap-instance-id selects the intended adapter by NetCfgInstanceId
/ InterfaceGuid. --list-tap-adapters prints the TAP adapter ids and exits
without connecting. It prints and reports client diagnostics snapshots with
relay reachability, LAN-gateway presence, route-pinning, QUIC datagram budget,
relay RTT, TAP status/IP, broadcast frame flow, frame/datagram counters, and
drops. The periodic diagnostics refresh the TAP unicast IP so DHCP results that
arrive after bridging starts become visible in later status lines, preferring a
non-link-local IPv4 address when Windows reports several TAP addresses. Each
snapshot also emits short user-facing lines such as relay/gateway connection
status, relay-route and TAP readiness warnings, DHCP address presence, relay
RTT, and broadcast-flow confirmation. One-way broadcast diagnostics distinguish
frames sent toward the LAN from broadcast frames received back from the LAN.
Malformed frames read from TAP, invalid or unauthorized source-MAC frames, L2
control-plane traffic, remote VLAN tags, DHCP server replies, IPv6 Router
Advertisements, IPv6 fragments, jumbo frames, frames above the negotiated TAP
MTU, and TAP frames whose encoded datagrams exceed the negotiated QUIC budget
are counted and dropped before relay send without stopping the bridge. Relayed
LAN frames are also safety-checked before TAP writes, so switch-control traffic,
invalid-source frames, jumbo frames, and over-TAP-MTU frames stay out of the
Windows adapter even if they reached the client over QUIC. Misdirected unicast
frames not addressed to the client's virtual MAC are also counted, skipped, and
logged with the drop reason; accepted TAP-to-relay and relay-to-TAP frames are
logged with direction, peer id, MACs, ethertype/length, frame length, action,
and drop reason. TAP device read/write errors still stop the bridge. Relay
lifecycle events are logged as they arrive, including gateway joins and peer
leaves. The client remembers peer identities from join and catch-up events and
from the initial welcome, so later leave logs can identify a disconnected LAN
gateway or client MAC when that peer was known.