enricobuehler
eea23c5647
Root-caused live on a phone at 100 Mbps (stream stuck seconds behind, then oscillating): a stack of transport defects, each amplifying the next. - MTU-safe shards: shard_payload 1452 overshot the IPv4/1500 budget (the old math forgot the 40 B header + 24 B crypto ride inside the UDP payload and counted IP+UDP as 8 B) — the kernel silently split EVERY video datagram into two IP fragments, doubling per-datagram loss on Wi-Fi. New config::mtu1500_shard_payload() = 1408 (1472 sealed = the exact ceiling), negotiated in the Welcome, pinned by a unit test. - Android batched I/O: recv/send batching was cfg(linux); Android is target_os="android" and silently fell back to a syscall per datagram. The libc crate binds neither recvmmsg/sendmmsg nor mmsghdr for Android, so a local bionic extern binding provides them (API 21+, floor is 28); cbindgen excludes them from the C header. The pump/runtime threads also get the Apple-QoS analogue on Android: nice −8 (below the decode thread's −10). - Latency-bounded receive: packets are consumed strictly in order at exactly the arrival rate, so a standing queue (Wi-Fi stall, power-save clumping) NEVER drains — observed as a stream permanently 6-7 s behind with both 32 MB socket buffers full. The pump now flushes the entire backlog (Session::flush_backlog: discard ring + kernel queue at memcpy speed, reset the reassembler) and requests a keyframe when frames keep completing > 400 ms behind the skew-corrected capture clock (30 consecutive, 2 s cooldown, logged). - Time-based loss window: the reassembler declared an incomplete frame lost a fixed 4 INDICES behind the newest — 33 ms at 120 fps, inside normal Wi-Fi retry/reorder timescales, so merely-late frames were pruned every few seconds, each costing a recovery-IDR burst + an inflated loss report. Now 120 ms of capture time (LOSS_WINDOW_NS), same fuse at every refresh rate, with a 64-index hard cap bounding memory against hostile pts. - Adaptive-FEC hysteresis: the controller was memoryless — one clean 750 ms report dropped FEC from 8 % straight back to the 1 % floor, so periodic burst loss (Wi-Fi scan / BT coexistence beats) always hit an unprotected stream and ping-ponged 1↔8 % with a frozen frame per cycle (observed in the host log as alternating loss_ppm=0/50000). Attack stays instant; decay is now one point per clean report. Verified: full core suite (incl. new flush + time-window tests) on macOS + Linux, host release build, arm64 cargo-ndk build, and a 30 s wired probe run at 2800x1260@120 — 3559/3559 frames, zero loss, capture→received p50 5.3 ms (host 5.1 + network 0.3). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
punktfunk-core
The shared protocol core — the one place where punktfunk's transport, forward error correction, and crypto live. It's linked into the host and every native client, so there's exactly one implementation of the wire format everywhere.
Written in Rust with no async on the per-frame path (native threads only). It exposes both a normal Rust API and a stable, versioned C ABI, so the Swift and Kotlin clients — and any C embedder — link the same code as the Rust ones.
What's in here
- Transport & session (
session.rs,transport/,packet.rs) — thepunktfunk/1data plane over raw UDP: packetization, reassembly (with attacker-bounded limits), pacing, and socket tuning. - FEC (
fec/) — the wall-breaker. Two codes:- GF(2⁸) classic Reed–Solomon with the Cauchy generator matrix — byte-identical to the
nanorslibrary Moonlight uses, so our parity is decodable by a stock Moonlight client. - GF(2¹⁶) Leopard-RS (SIMD, O(n log n)) — up to 65535 shards/block, which removes the ~1 Gbps
FEC ceiling.
punktfunk/1negotiates this one.
- GF(2⁸) classic Reed–Solomon with the Cauchy generator matrix — byte-identical to the
- Crypto (
crypto.rs) — AES-128-GCM session encryption with per-direction nonce salts and sequence-as-AAD; SPAKE2 PIN pairing lives behind thequicfeature. - QUIC control plane (
quic.rs,client.rs, featurequic) — the Hello/Welcome/Start handshake, cert pinning/TOFU, reverse audio, and the embeddableNativeClientconnector. This is the only placetokio/quinnare allowed; the feature is off by default so the core stays runtime-free. - C ABI (
abi.rs) — the versioned surface (punktfunk_abi_version(),PunktfunkConfigcarrying its ownstruct_size) that generatesinclude/punktfunk_core.hvia cbindgen at build time.
Build outputs
The crate builds three ways at once (crate-type = ["lib", "cdylib", "staticlib"]):
| Output | Used by |
|---|---|
lib (rlib) |
the host, probe, and tools link it as a normal Rust crate |
cdylib (.so/.dylib) |
the Swift / Kotlin clients via the C ABI |
staticlib (.a) |
the C test harness and static embedding |
Test
cargo test -p punktfunk-core # unit + proptest + loopback
cargo run -p loss-harness # FEC loss-resilience sweep (no network needed)
bash crates/punktfunk-core/tests/c/run.sh # standalone C-ABI link + round-trip proof
Design invariants (do not regress)
- One core, linked everywhere — protocol/FEC/crypto live only here, behind the stable C ABI.
- No async on the hot path — the per-frame pipeline is native threads only;
quic(tokio/quinn) is control-plane only, feature-gated, off by default. - Security hardening stays intact — the reassembler bounds attacker-controlled fields before
allocating; AES-GCM keeps per-direction nonce salts + seq-as-AAD; the ABI checks
struct_size. Regression tests exist — keep them green.
Related
punktfunk-host— the streaming host built on this core- Clients — the apps that link this core over the C ABI (or directly, in Rust)
design/implementation-plan.md— why GF(2¹⁶) FEC, the latency budget, and the architecture thesis