fix: resolve critical ARQ bugs causing Speedtest tunnel drops + docs overhaul

Critical fixes (6):
- protocol.rs: in_flight_count() now counts only retransmittable Data frames,
  not Ack/Nack control frames — eliminates false backpressure under load
- protocol.rs: NACK is now rate-limited to once per 30ms — prevents
  retransmission storm during normal UDP jitter
- protocol.rs: zombie frames exceeding max_retries+4 are evicted each tick —
  prevents unbounded memory growth and stale retransmits
- protocol.rs: Closing state now processes final in-flight packets instead
  of silently dropping them — prevents data loss at session teardown
- server/lib.rs: stream_tx changed from bounded(10000) to unbounded_channel —
  prevents TCP-reader collapse during Speedtest with 50+ streams
- bridge.rs: liveness timeout raised from 30s to 60s — prevents false
  reconnect during heavy Speedtest load

Medium fixes (8):
- protocol.rs: ACK range truncation preserves cumulative range (index 0)
- bridge.rs: Ping now uses send_datagram() for correct TURN wrapping
- dispatcher.rs: replay_cache hard-capped at 100k entries (DoS protection)
- dispatcher.rs: old addr cleaned from addr_to_session on roaming
- server/lib.rs: TCP connect_target() now has 10s timeout
- config.rs: TURN section parsed during hot-reload
- proxy.rs: HTTP header parsing uses 512-byte chunks instead of 1-byte reads
- proxy.rs: stream_id wrap-around skips active IDs to prevent collision
- runner.rs: is_essential_log matches actual log strings from bridge.rs

Other:
- kex.rs: clearly marked as dead PQ stub (not used by protocol)
- README.md + README.ru.md: complete rewrite with architecture diagram
- docs/en/specification.md: updated ARQ section with all new semantics
This commit is contained in:
ospab 2026-05-17 03:20:50 +03:00
parent a9ba941782
commit 7424ccc0ff
11 changed files with 409 additions and 108 deletions

160
README.md
View File

@ -1,4 +1,4 @@
# OSTP (Ospab Stealth Transport Protocol) # OSTP — Ospab Stealth Transport Protocol
[Русский язык](README.ru.md) [Русский язык](README.ru.md)
@ -6,30 +6,67 @@
![License: BSL 1.1](https://img.shields.io/badge/License-BSL%201.1-orange.svg?style=flat-square) ![License: BSL 1.1](https://img.shields.io/badge/License-BSL%201.1-orange.svg?style=flat-square)
![Platform: Windows | Linux | macOS | Android](https://img.shields.io/badge/Platform-Windows%20%7C%20Linux%20%7C%20macOS%20%7C%20Android-green.svg?style=flat-square) ![Platform: Windows | Linux | macOS | Android](https://img.shields.io/badge/Platform-Windows%20%7C%20Linux%20%7C%20macOS%20%7C%20Android-green.svg?style=flat-square)
OSTP is a fast and secure transport protocol designed to bypass DPI and network restrictions. It masks traffic as high-entropy data, making it difficult to detect or block. OSTP is a high-performance, censorship-resistant transport protocol designed to tunnel TCP traffic over UDP with full traffic obfuscation. It is resistant to Deep Packet Inspection (DPI), active probing, and statistical traffic analysis.
--- ---
## Features ## Key Features
- **Traffic Obfuscation**: Hides VPN/proxy signatures from network analysis. | Feature | Description |
- **High Performance**: Written in Rust using the gVisor network stack for low latency. |---------|-------------|
- **Reliable Connectivity**: Built-in keep-alive mechanism for stable operation on mobile networks. | **Traffic Obfuscation** | Every packet — including headers — is indistinguishable from random noise on the wire. Session IDs and nonces are masked with per-packet HMAC-derived keys. |
- **Flexible Modes**: Supports SOCKS5/HTTP proxying and full-system TUN (VPN) mode. | **Noise Protocol Handshake** | `Noise_NNpsk0_25519_ChaChaPoly_BLAKE2s` — pre-shared key authenticated, forward-secret key exchange with no static identity exposure. |
- **Multi-platform**: Compatible with Windows, Linux, macOS, and Android. | **Reliable UDP (ARQ)** | Selective ACK/NACK with rate-limited retransmission, configurable reorder buffer, and exponential backoff. Designed for 10 Gbps throughput. |
| **Multiplexed Streams** | Multiple logical TCP streams over a single encrypted UDP session, with per-stream flow control. |
| **Seamless Roaming** | Clients can switch networks (WiFi ↔ 4G) without session interruption — the server tracks session-ID, not IP address. |
| **TUN Mode** | Full-system VPN via `tun2socks` integration on Windows and Linux. All traffic is transparently routed through the tunnel. |
| **TURN Relay** | RFC 5766 TURN support for environments where direct UDP is blocked. |
| **Hot-Reload** | Runtime config reload without restarting the process (access keys, exclusions, mux settings, TURN). |
| **Cross-Platform** | Windows, Linux, macOS, Android. Single binary, no runtime dependencies. |
---
## Architecture
```
┌─────────────────────────────────────────────────────────────┐
│ Client │
│ ┌──────────┐ ┌──────────┐ ┌────────────────────────┐ │
│ │ Browser │──▸│ SOCKS5/ │──▸│ Bridge (Mux) │ │
│ │ / Apps │ │ HTTP │ │ ┌─────────────────┐ │ │
│ │ │ │ Proxy │ │ │ ProtocolMachine │ │ │
│ └──────────┘ └──────────┘ │ │ (Noise + AEAD) │ │ │
│ │ └────────┬────────┘ │ │
│ ┌──────────┐ │ │ │ │
│ │ TUN Mode │──────────────────┤ UDP Socket │ │
│ │tun2socks │ │ (32MB buffers, │ │
│ └──────────┘ │ obfuscated wire) │ │
│ └───────────┬────────────┘ │
└────────────────────────────────────────────┼────────────────┘
│ UDP
┌────────────────────────────────────────────┼────────────────┐
│ Server │ │
│ ┌─────────────────────────────────────────┴───────────┐ │
│ │ Dispatcher │ │
│ │ (Session lookup, roaming detection, replay guard) │ │
│ └──────────────┬──────────────────────────────────────┘ │
│ │ │
│ ┌──────────────▾──────────────────┐ │
│ │ Relay Loop (per-stream TCP) │──▸ Internet / Backend │
│ └─────────────────────────────────┘ │
└─────────────────────────────────────────────────────────────┘
```
--- ---
## Installation ## Installation
### Linux ### Linux
Run the installer script to set up OSTP as a system service:
```bash ```bash
bash <(curl -Ls https://raw.githubusercontent.com/ospab/ostp/master/scripts/install.sh) bash <(curl -Ls https://raw.githubusercontent.com/ospab/ostp/master/scripts/install.sh)
``` ```
### Windows ### Windows (PowerShell, Administrator)
Run the following in PowerShell as Administrator:
```powershell ```powershell
irm https://raw.githubusercontent.com/ospab/ostp/master/scripts/install.ps1 | iex irm https://raw.githubusercontent.com/ospab/ostp/master/scripts/install.ps1 | iex
``` ```
@ -38,23 +75,23 @@ irm https://raw.githubusercontent.com/ospab/ostp/master/scripts/install.ps1 | ie
## Configuration ## Configuration
Initialize a default config file: Generate a default config:
```bash ```bash
./ostp --init server # For VPS ./ostp --init server # VPS
./ostp --init client # For local machine ./ostp --init client # Local machine
``` ```
### Server (config.json) ### Server (`config.json`)
```jsonc ```jsonc
{ {
// OSTP Server Configuration
"mode": "server", "mode": "server",
"listen": "0.0.0.0:50000", "listen": "0.0.0.0:50000",
"access_keys": ["YOUR_KEY"], "access_keys": ["YOUR_SECRET_KEY"],
// Optional: forward traffic to another proxy "debug": false,
// Optional: forward traffic through an upstream proxy
"outbound": { "outbound": {
"enabled": false, "enabled": false,
"protocol": "socks5", "protocol": "socks5", // "socks5" or "http"
"address": "127.0.0.1", "address": "127.0.0.1",
"port": 9050, "port": 9050,
"default_action": "proxy" "default_action": "proxy"
@ -62,20 +99,35 @@ Initialize a default config file:
} }
``` ```
### Client (config.json) ### Client (`config.json`)
```jsonc ```jsonc
{ {
// OSTP Client Configuration
"mode": "client", "mode": "client",
"server": "SERVER_IP:50000", "server": "YOUR_SERVER_IP:50000",
"access_key": "YOUR_KEY", "access_key": "YOUR_SECRET_KEY",
"socks5_bind": "127.0.0.1:1088", "socks5_bind": "127.0.0.1:1088",
// Virtual network adapter settings "debug": false,
// TUN mode (full-system VPN)
"tun": { "tun": {
"enable": false, "enable": false,
"wintun_path": "./wintun.dll",
"ipv4_address": "10.1.0.2/24",
"dns": "1.1.1.1" "dns": "1.1.1.1"
},
// Multiplexing: spread traffic across multiple UDP sessions
"mux": {
"enabled": false,
"sessions": 2
},
// TURN relay for restricted networks
"turn": {
"enabled": false,
"server_addr": "turn.example.com:3478",
"username": "user",
"access_key": "pass"
},
// Traffic exclusions (bypassed directly)
"exclude": {
"domains": ["example.local"],
"ips": ["192.168.0.0/16"]
} }
} }
``` ```
@ -84,15 +136,65 @@ Initialize a default config file:
## Usage ## Usage
Start the node with your configuration:
```bash ```bash
# Start with config
./ostp --config config.json ./ostp --config config.json
# Or just run (looks for config.json in current/binary directory)
./ostp
``` ```
For TUN mode on Windows, ensure `tun2socks.exe` and `wintun.dll` are in the same directory. ### TUN Mode (Windows)
Requires `tun2socks.exe` in the same directory. Automatically requests Administrator privileges.
### TUN Mode (Linux)
Requires root. Uses `tun2socks` binary (same directory or in `$PATH`).
---
## Protocol Specification
See [docs/en/specification.md](docs/en/specification.md) for the full wire format, handshake flow, and ARQ semantics.
### Quick Summary
| Layer | Mechanism |
|-------|-----------|
| Key Exchange | Noise NNpsk0 (X25519 + ChaChaPoly + BLAKE2s) |
| Encryption | ChaCha20-Poly1305 AEAD per-packet |
| Header Obfuscation | HMAC-SHA256 derived per-packet mask over session_id + nonce |
| Reliability | Selective ACK with cumulative + SACK ranges |
| Retransmission | Rate-limited NACK (30ms cooldown) + exponential backoff RTO |
| Flow Control | In-flight window (retransmittable frames only) |
| Keepalive | Ping/Pong with RTT measurement every 5s |
| Session Timeout | 60s inactivity on client, 300s on server |
---
## Building from Source
```bash
# Prerequisites: Rust toolchain (1.75+)
cargo build --release
# Cross-compile for Linux (from Windows/macOS)
cross build --release --target x86_64-unknown-linux-gnu
```
---
## Documentation
- [Architecture Overview](docs/en/architecture.md)
- [Protocol Specification](docs/en/specification.md)
- [Obfuscation Design](docs/en/obfuscation.md)
- [Server Administration](docs/en/server.md)
- [Client Configuration](docs/en/client.md)
- [Integration Guide](docs/en/integrations.md)
--- ---
## License ## License
Business Source License 1.1. Free for personal and non-commercial use. Converts to MIT License on May 14, 2030. Business Source License 1.1. Free for personal and non-commercial use.
Converts to MIT License on May 14, 2030.

View File

@ -1,4 +1,4 @@
# OSTP (Ospab Stealth Transport Protocol) # OSTP — Ospab Stealth Transport Protocol
[English](README.md) [English](README.md)
@ -6,52 +6,89 @@
![License: BSL 1.1](https://img.shields.io/badge/License-BSL%201.1-orange.svg?style=flat-square) ![License: BSL 1.1](https://img.shields.io/badge/License-BSL%201.1-orange.svg?style=flat-square)
![Platform: Windows | Linux | macOS | Android](https://img.shields.io/badge/Platform-Windows%20%7C%20Linux%20%7C%20macOS%20%7C%20Android-green.svg?style=flat-square) ![Platform: Windows | Linux | macOS | Android](https://img.shields.io/badge/Platform-Windows%20%7C%20Linux%20%7C%20macOS%20%7C%20Android-green.svg?style=flat-square)
OSTP — это быстрый и безопасный транспортный протокол для обхода DPI и сетевых ограничений. Он маскирует трафик под высокоэнтропийные данные, что делает его труднообнаружимым для систем блокировки. OSTP — высокопроизводительный транспортный протокол, устойчивый к цензуре. Туннелирует TCP-трафик поверх UDP с полной обфускацией. Устойчив к Deep Packet Inspection (DPI), активному зондированию и статистическому анализу трафика.
--- ---
## Возможности ## Возможности
- **Обфускация трафика**: Скрывает сигнатуры VPN и прокси от сетевого анализа. | Возможность | Описание |
- **Высокая производительность**: Написан на Rust с использованием сетевого стека gVisor. |-------------|----------|
- **Стабильность**: Встроенный механизм keep-alive для надежной работы в мобильных сетях. | **Обфускация трафика** | Каждый пакет, включая заголовки, неотличим от случайного шума. Session ID и nonce маскируются HMAC-ключами, уникальными для каждого пакета. |
- **Гибкость**: Поддержка проксирования SOCKS5/HTTP и полнофункционального TUN (VPN) режима. | **Noise Protocol** | `Noise_NNpsk0_25519_ChaChaPoly_BLAKE2s` — аутентификация через PSK, forward secrecy, без раскрытия идентичности. |
- **Кроссплатформенность**: Работает на Windows, Linux, macOS и Android. | **Reliable UDP (ARQ)** | Selective ACK/NACK с rate-limited ретрансмиссией, настраиваемым reorder-буфером и exponential backoff. Разработан для 10 Гбит/с. |
| **Мультиплексирование** | Несколько логических TCP-потоков поверх одной зашифрованной UDP-сессии с per-stream flow control. |
| **Бесшовный роуминг** | Клиент может менять сети (WiFi ↔ 4G) без разрыва сессии — сервер отслеживает session-ID, а не IP-адрес. |
| **TUN-режим** | Полносистемный VPN через интеграцию с `tun2socks` на Windows и Linux. |
| **TURN Relay** | RFC 5766 TURN для окружений, где прямой UDP заблокирован. |
| **Hot-Reload** | Перезагрузка конфига в рантайме без перезапуска (ключи, исключения, mux, TURN). |
| **Кросс-платформа** | Windows, Linux, macOS, Android. Один бинарник, без зависимостей. |
---
## Архитектура
```
┌────────────────────────────────────────────────────────────┐
│ Клиент │
│ ┌──────────┐ ┌──────────┐ ┌───────────────────────┐ │
│ │ Браузер │──▸│ SOCKS5/ │──▸│ Bridge (Mux) │ │
│ │ / Прил. │ │ HTTP │ │ ┌─────────────────┐ │ │
│ │ │ │ Прокси │ │ │ ProtocolMachine │ │ │
│ └──────────┘ └──────────┘ │ │ (Noise + AEAD) │ │ │
│ │ └────────┬────────┘ │ │
│ ┌──────────┐ │ │ │ │
│ │ TUN Mode │──────────────────┤ UDP-сокет │ │
│ │tun2socks │ │ (32МБ буферы, │ │
│ └──────────┘ │ обфускация) │ │
│ └───────────┬────────────┘ │
└────────────────────────────────────────────┼────────────────┘
│ UDP
┌────────────────────────────────────────────┼────────────────┐
│ Сервер │ │
│ ┌─────────────────────────────────────────┴──────────┐ │
│ │ Dispatcher │ │
│ │ (Поиск сессий, роуминг, защита от replay) │ │
│ └──────────────┬──────────────────────────────────────┘ │
│ │ │
│ ┌──────────────▾──────────────────┐ │
│ │ Relay Loop (TCP per-stream) │──▸ Интернет / Backend │
│ └─────────────────────────────────┘ │
└──────────────────────────────────────────────────────────────┘
```
--- ---
## Установка ## Установка
### Linux ### Linux
Используйте скрипт для автоматической установки и настройки сервиса:
```bash ```bash
bash <(curl -Ls https://raw.githubusercontent.com/ospab/ostp/master/scripts/install.sh) bash <(curl -Ls https://raw.githubusercontent.com/ospab/ostp/master/scripts/install.sh)
``` ```
### Windows ### Windows (PowerShell от Администратора)
Запустите в PowerShell от имени администратора:
```powershell ```powershell
irm https://raw.githubusercontent.com/ospab/ostp/master/scripts/install.ps1 | iex irm https://raw.githubusercontent.com/ospab/ostp/master/scripts/install.ps1 | iex
``` ```
--- ---
## Настройка ## Конфигурация
Создайте файл конфигурации по умолчанию: Создать конфиг по умолчанию:
```bash ```bash
./ostp --init server # Для сервера (VPS) ./ostp --init server # VPS
./ostp --init client # Для клиента (ПК) ./ostp --init client # Локальная машина
``` ```
### Сервер (config.json) ### Сервер (`config.json`)
```jsonc ```jsonc
{ {
// Конфигурация Сервера OSTP
"mode": "server", "mode": "server",
"listen": "0.0.0.0:50000", "listen": "0.0.0.0:50000",
"access_keys": ["ВАШ_КЛЮЧ"], "access_keys": ["ВАШ_КЛЮЧ"],
// Опционально: пересылка трафика через другой прокси "debug": false,
// Опционально: проксировать трафик через upstream
"outbound": { "outbound": {
"enabled": false, "enabled": false,
"protocol": "socks5", "protocol": "socks5",
@ -62,20 +99,35 @@ irm https://raw.githubusercontent.com/ospab/ostp/master/scripts/install.ps1 | ie
} }
``` ```
### Клиент (config.json) ### Клиент (`config.json`)
```jsonc ```jsonc
{ {
// Конфигурация Клиента OSTP
"mode": "client", "mode": "client",
"server": "IP_СЕРВЕРА:50000", "server": "IP_СЕРВЕРА:50000",
"access_key": "ВАШ_КЛЮЧ", "access_key": "ВАШ_КЛЮЧ",
"socks5_bind": "127.0.0.1:1088", "socks5_bind": "127.0.0.1:1088",
// Настройки виртуального сетевого адаптера "debug": false,
// TUN-режим (полносистемный VPN)
"tun": { "tun": {
"enable": false, "enable": false,
"wintun_path": "./wintun.dll",
"ipv4_address": "10.1.0.2/24",
"dns": "1.1.1.1" "dns": "1.1.1.1"
},
// Мультиплексирование: несколько UDP-сессий
"mux": {
"enabled": false,
"sessions": 2
},
// TURN-реле для заблокированных сетей
"turn": {
"enabled": false,
"server_addr": "turn.example.com:3478",
"username": "user",
"access_key": "pass"
},
// Исключения (идут напрямую, минуя туннель)
"exclude": {
"domains": ["example.local"],
"ips": ["192.168.0.0/16"]
} }
} }
``` ```
@ -84,15 +136,61 @@ irm https://raw.githubusercontent.com/ospab/ostp/master/scripts/install.ps1 | ie
## Использование ## Использование
Запустите программу с вашим конфигом:
```bash ```bash
# Запуск с конфигом
./ostp --config config.json ./ostp --config config.json
# Или просто (ищет config.json рядом с бинарником)
./ostp
``` ```
Для работы TUN режима в Windows файлы `tun2socks.exe` и `wintun.dll` должны находиться в одной папке с бинарным файлом. ### TUN-режим (Windows)
Требуется `tun2socks.exe` в той же директории. Автоматически запрашивает права Администратора.
### TUN-режим (Linux)
Требуется root. Нужен бинарник `tun2socks` (рядом или в `$PATH`).
---
## Спецификация протокола
| Уровень | Механизм |
|---------|----------|
| Обмен ключами | Noise NNpsk0 (X25519 + ChaChaPoly + BLAKE2s) |
| Шифрование | ChaCha20-Poly1305 AEAD на каждый пакет |
| Обфускация заголовков | HMAC-SHA256 маска session_id + nonce, уникальная для каждого пакета |
| Надёжность | Selective ACK с cumulative + SACK диапазонами |
| Ретрансмиссия | Rate-limited NACK (30мс cooldown) + exponential backoff RTO |
| Flow Control | Окно in-flight (только retransmittable фреймы) |
| Keepalive | Ping/Pong с измерением RTT каждые 5с |
| Таймаут сессии | 60с на клиенте, 300с на сервере |
---
## Сборка из исходников
```bash
# Требования: Rust toolchain (1.75+)
cargo build --release
# Кросс-компиляция для Linux
cross build --release --target x86_64-unknown-linux-gnu
```
---
## Документация
- [Архитектура](docs/ru/architecture.md)
- [Спецификация протокола](docs/ru/specification.md)
- [Дизайн обфускации](docs/ru/obfuscation.md)
- [Администрирование сервера](docs/ru/server.md)
- [Настройка клиента](docs/ru/client.md)
- [Интеграции](docs/ru/integrations.md)
--- ---
## Лицензия ## Лицензия
Business Source License 1.1. Бесплатно для личного и некоммерческого использования. Переходит в MIT License 14 мая 2030 года. Business Source License 1.1. Бесплатно для личного и некоммерческого использования.
Переходит в MIT License 14 мая 2030 года.

View File

@ -102,16 +102,23 @@ The initial handshake payload includes a Unix timestamp to mitigate replay attac
### 7.1 Selective-Repeat ARQ ### 7.1 Selective-Repeat ARQ
OSTP provides reliability over UDP using a **Selective-Repeat ARQ** mechanism: OSTP provides reliability over UDP using a **Selective-Repeat ARQ** mechanism:
* The receiver maintains a reorder buffer (default: 8192 packets). * The receiver maintains a reorder buffer (default: 32768 packets) for out-of-order packet reassembly.
* Unacknowledged packets are retransmitted after an adaptive Retransmission Time Out (RTO). * Acknowledgments use a **Cumulative + SACK** scheme: the ACK payload contains a cumulative range `(0, expected_recv_nonce - 1)` confirming all contiguous packets received, plus up to 7 additional Selective ACK ranges for non-contiguous blocks in the reorder buffer.
* Acknowledgments (ACKs) are piggybacked onto outbound data frames to minimize overhead. * **Rate-limited NACK:** When a gap is detected, the receiver emits a NACK for the lowest missing nonce, but no more than once per 30ms. This prevents retransmission storms under normal UDP jitter.
* Backpressure is applied dynamically based on the number of in-flight unacknowledged frames. * **Retransmission:** Unacknowledged data frames are retransmitted after an adaptive Retransmission Timeout (RTO, default: 100ms) with exponential backoff (up to 64× base RTO).
* **Zombie Frame Eviction:** Frames exceeding `max_retries + 4` attempts are automatically dropped from the send history, preventing unbounded memory consumption and stale retransmissions.
* **In-flight Counting:** Backpressure is based only on retransmittable (data) frames; control frames (ACK/NACK) are excluded from the in-flight count to prevent false backpressure under high load.
* **Graceful Close:** The `Closing` state processes all remaining in-flight packets before transitioning to `Closed`, preventing data loss during session teardown.
### 7.2 Adaptive Padding ### 7.2 Adaptive Padding
To resist traffic analysis via Packet Length Analysis (PLA), OSTP pads plaintext payloads before AEAD encryption. Padding bytes are drawn from a cryptographically secure random source. The protocol supports dynamic padding boundaries up to the maximum MTU (e.g., 1400 bytes), smoothing out recognizable application traffic bursts into constant-bitrate-like streams. To resist traffic analysis via Packet Length Analysis (PLA), OSTP pads plaintext payloads before AEAD encryption. Padding bytes are drawn from a cryptographically secure random source. The protocol supports dynamic padding boundaries up to the maximum MTU (e.g., 1400 bytes), smoothing out recognizable application traffic bursts into constant-bitrate-like streams.
### 7.3 IP Roaming ### 7.3 IP Roaming
The server supports seamless network handoffs (e.g., transitioning from Wi-Fi to cellular networks). If a packet successfully passes AEAD authentication, the server automatically binds the Session ID to the new source IP address without requiring a session restart. The server supports seamless network handoffs (e.g., transitioning from Wi-Fi to cellular networks). If a packet successfully passes AEAD authentication, the server automatically binds the Session ID to the new source IP address without requiring a session restart. The server maintains a rate-limited roaming scanner (50 tokens/sec) to prevent CPU exhaustion from probing attacks.
### 7.4 Session Keepalive
* **Client-side:** Ping/Pong frames with RTT measurement are sent every 5 seconds. If no valid UDP packet is received for 60 seconds, the client initiates reconnection.
* **Server-side:** Sessions with no activity for 300 seconds are automatically evicted.
--- ---

View File

@ -354,7 +354,7 @@ impl Bridge {
_ = keepalive_tick.tick() => { _ = keepalive_tick.tick() => {
if self.running { if self.running {
// 1. Connection Liveness Check // 1. Connection Liveness Check
if self.last_valid_recv.elapsed().as_secs() > 30 { if self.last_valid_recv.elapsed().as_secs() > 60 {
let _ = tx.send(UiEvent::Log("Connection lost (timeout). Reconnecting...".into())).await; let _ = tx.send(UiEvent::Log("Connection lost (timeout). Reconnecting...".into())).await;
self.running = false; self.running = false;
_proxy_guard = None; _proxy_guard = None;
@ -369,11 +369,13 @@ impl Bridge {
// 2. Active Keep-Alive / Heartbeat // 2. Active Keep-Alive / Heartbeat
if let Some(sessions) = sessions_opt.as_mut() { if let Some(sessions) = sessions_opt.as_mut() {
for session in sessions.iter_mut() { for session in sessions.iter_mut() {
// Send Ping (Internal Metric) // Send Ping (Internal RTT Metric)
let ts = SystemTime::now().duration_since(std::time::UNIX_EPOCH).unwrap().as_millis() as u64; let ts = SystemTime::now().duration_since(std::time::UNIX_EPOCH).unwrap().as_millis() as u64;
let ping_payload = Bytes::from(RelayMessage::Ping(ts).encode()); let ping_payload = Bytes::from(RelayMessage::Ping(ts).encode());
if let Ok(ProtocolAction::SendDatagram(frame)) = session.machine.on_event(OstpEvent::Outbound(0, ping_payload)) { if let Ok(ProtocolAction::SendDatagram(frame)) = session.machine.on_event(OstpEvent::Outbound(0, ping_payload)) {
let _ = session.socket.send(&frame).await; // Must go through send_datagram() for TURN-mode wrapping;
// raw socket.send() bypasses the ChannelData header and breaks RTT in TURN.
let _ = send_datagram(&session.socket, &frame, self.turn_enabled).await;
self.metrics.bytes_sent.fetch_add(frame.len() as u64, Ordering::Relaxed); self.metrics.bytes_sent.fetch_add(frame.len() as u64, Ordering::Relaxed);
} }

View File

@ -129,6 +129,7 @@ struct RawUnifiedConfig {
tun: Option<RawTunSection>, tun: Option<RawTunSection>,
exclude: Option<RawExcludeSection>, exclude: Option<RawExcludeSection>,
mux: Option<RawMuxSection>, mux: Option<RawMuxSection>,
turn: Option<RawTurnSection>,
} }
#[derive(Debug, Deserialize)] #[derive(Debug, Deserialize)]
@ -150,6 +151,14 @@ struct RawMuxSection {
sessions: Option<usize>, sessions: Option<usize>,
} }
#[derive(Debug, Deserialize)]
struct RawTurnSection {
enabled: Option<bool>,
server_addr: Option<String>,
username: Option<String>,
access_key: Option<String>,
}
impl ClientConfig { impl ClientConfig {
/// Hot-reload from `config.json` placed next to the running binary. /// Hot-reload from `config.json` placed next to the running binary.
/// Returns a new `ClientConfig` built from the unified JSON format. /// Returns a new `ClientConfig` built from the unified JSON format.
@ -192,7 +201,15 @@ impl ClientConfig {
bind_addr: socks5, bind_addr: socks5,
connect_timeout_ms: 15000, connect_timeout_ms: 15000,
}, },
turn: TurnConfig::default(), turn: match raw.turn {
Some(t) => TurnConfig {
enabled: t.enabled.unwrap_or(false),
server_addr: t.server_addr.unwrap_or_default(),
username: t.username.unwrap_or_default(),
access_key: t.access_key.unwrap_or_default(),
},
None => TurnConfig::default(),
},
exclusions: ExclusionConfig { exclusions: ExclusionConfig {
domains: exclusions.domains.unwrap_or_default(), domains: exclusions.domains.unwrap_or_default(),
ips: exclusions.ips.unwrap_or_default(), ips: exclusions.ips.unwrap_or_default(),

View File

@ -279,15 +279,16 @@ fn format_bytes(bps: u64) -> String {
fn is_essential_log(text: &str) -> bool { fn is_essential_log(text: &str) -> bool {
matches!( matches!(
text, text,
"Handshaking started" "Connection established"
| "Bridge connection established"
| "TUN Tunnel established" | "TUN Tunnel established"
| "Bridge stopped" | "Bridge stopped"
| "TUN Tunnel stopped" | "TUN Tunnel stopped"
| "Runtime config reloaded" | "Runtime config reloaded"
| "Connecting to remote server..."
) || text.starts_with("Connected UDP directly to ") ) || text.starts_with("Connected UDP directly to ")
|| text.starts_with("TURN: Relay allocated") || text.starts_with("TURN: Relay allocated")
|| text.starts_with("TURN allocation failed") || text.starts_with("TURN allocation failed")
|| text.starts_with("Handshake failed") || text.starts_with("Connection failed:")
|| text.starts_with("Connection timeout") || text.starts_with("Connection lost")
|| text.starts_with("Protocol tick fatal error")
} }

View File

@ -43,8 +43,12 @@ pub async fn run_local_socks5_proxy(
accepted = listener.accept() => { accepted = listener.accept() => {
let (socket, _) = accepted?; let (socket, _) = accepted?;
let stream_id = next_stream_id; let stream_id = next_stream_id;
// Advance, skipping zero and any stream_id still in active_streams
loop {
next_stream_id = next_stream_id.wrapping_add(1); next_stream_id = next_stream_id.wrapping_add(1);
if next_stream_id == 0 { next_stream_id = 1; } if next_stream_id == 0 { next_stream_id = 1; }
if !active_streams.contains_key(&next_stream_id) { break; }
}
let (tx, rx) = mpsc::unbounded_channel(); let (tx, rx) = mpsc::unbounded_channel();
active_streams.insert(stream_id, tx); active_streams.insert(stream_id, tx);
@ -229,13 +233,19 @@ async fn handle_proxy_client(
// Read the rest of the HTTP request headers byte-by-byte // Read the rest of the HTTP request headers byte-by-byte
let mut header_bytes = Vec::with_capacity(512); let mut header_bytes = Vec::with_capacity(512);
header_bytes.push(first_byte[0]); header_bytes.push(first_byte[0]);
let mut byte = [0_u8; 1]; let mut chunk = [0_u8; 512];
loop { loop {
client.read_exact(&mut byte).await?; let n = client.read(&mut chunk).await?;
header_bytes.push(byte[0]); if n == 0 {
if header_bytes.ends_with(b"\r\n\r\n") { return Err(anyhow!("connection closed during HTTP header read"));
}
header_bytes.extend_from_slice(&chunk[..n]);
if header_bytes.len() >= 4 {
let tail = &header_bytes[header_bytes.len().saturating_sub(4)..];
if tail.ends_with(b"\r\n\r\n") {
break; break;
} }
}
if header_bytes.len() > 8192 { if header_bytes.len() > 8192 {
client.write_all(b"HTTP/1.1 431 Request Header Fields Too Large\r\n\r\n").await?; client.write_all(b"HTTP/1.1 431 Request Header Fields Too Large\r\n\r\n").await?;
return Err(anyhow!("HTTP header too large")); return Err(anyhow!("HTTP header too large"));

View File

@ -1,7 +1,26 @@
use rand::rngs::OsRng; // =============================================================================
use sha2::{Digest, Sha256}; // OSTP Hybrid Key Exchange — STUB / NOT IN USE
use x25519_dalek::{EphemeralSecret, PublicKey}; // =============================================================================
//
// This module is a placeholder for future post-quantum key exchange.
// The actual key exchange is handled by the Noise NNpsk0 handshake in noise.rs.
//
// When ML-KEM (CRYSTALS-Kyber) support is added, this module will provide:
// 1. X25519 ephemeral DH (classical security)
// 2. ML-KEM-768 encapsulation (post-quantum security)
// 3. Combined shared secret = SHA-256(x25519_secret || ml_kem_secret)
//
// Until then, DO NOT use this module in production — it provides zero
// post-quantum security. The Noise handshake in noise.rs is the only
// active key exchange mechanism.
// =============================================================================
#![allow(dead_code)]
use sha2::{Digest, Sha256};
/// Placeholder shared secret output.
/// NOT USED by the protocol — provided for future API compatibility only.
#[derive(Debug, Clone)] #[derive(Debug, Clone)]
pub struct HybridSharedSecret { pub struct HybridSharedSecret {
pub x25519_pubkey: [u8; 32], pub x25519_pubkey: [u8; 32],
@ -9,19 +28,27 @@ pub struct HybridSharedSecret {
pub combined_secret: [u8; 32], pub combined_secret: [u8; 32],
} }
pub trait KeyExchange { /// Placeholder hybrid key exchange.
fn client_kex() -> HybridSharedSecret; /// The PQ component is a no-op stub. See module-level documentation.
}
pub struct HybridKex; pub struct HybridKex;
impl HybridKex { impl HybridKex {
/// Generate a hybrid key exchange offer.
///
/// # Security Warning
/// The post-quantum component is a **stub** — `pq_ciphertext` is all zeros.
/// This function exists solely for API scaffolding. Do not rely on it for
/// post-quantum security.
pub fn client_offer() -> HybridSharedSecret { pub fn client_offer() -> HybridSharedSecret {
use rand::rngs::OsRng;
use x25519_dalek::{EphemeralSecret, PublicKey};
let secret = EphemeralSecret::random_from_rng(OsRng); let secret = EphemeralSecret::random_from_rng(OsRng);
let pubkey = PublicKey::from(&secret); let pubkey = PublicKey::from(&secret);
// Placeholder PQ ciphertext. Replace with ML-KEM encapsulation output. // TODO: Replace with ML-KEM-768 encapsulation (crate `ml-kem`)
let pq_ciphertext = vec![0_u8; 1088]; let pq_ciphertext = vec![0_u8; 1088];
let mut hasher = Sha256::new(); let mut hasher = Sha256::new();
hasher.update(pubkey.as_bytes()); hasher.update(pubkey.as_bytes());
hasher.update(&pq_ciphertext); hasher.update(&pq_ciphertext);
@ -37,9 +64,3 @@ impl HybridKex {
} }
} }
} }
impl KeyExchange for HybridKex {
fn client_kex() -> HybridSharedSecret {
Self::client_offer()
}
}

View File

@ -81,6 +81,8 @@ pub struct ProtocolMachine {
max_sent_history: usize, max_sent_history: usize,
ack_pending: bool, ack_pending: bool,
last_ack_sent: Instant, last_ack_sent: Instant,
/// Rate-limit: prevents sending a NACK more than once per 30ms to avoid storms
last_nack_sent: Instant,
} }
#[derive(Debug, Clone)] #[derive(Debug, Clone)]
@ -121,11 +123,14 @@ impl ProtocolMachine {
max_sent_history: config.max_sent_history.max(1), max_sent_history: config.max_sent_history.max(1),
ack_pending: false, ack_pending: false,
last_ack_sent: Instant::now(), last_ack_sent: Instant::now(),
last_nack_sent: Instant::now() - Duration::from_secs(1),
}) })
} }
pub fn in_flight_count(&self) -> usize { pub fn in_flight_count(&self) -> usize {
self.sent_history.len() // COUNT ONLY retransmittable Data frames — control frames (Ack/Nack) must not
// contribute to this counter or they will trigger false backpressure.
self.sent_history.iter().filter(|f| f.is_retransmittable).count()
} }
pub fn state(&self) -> OstpState { pub fn state(&self) -> OstpState {
@ -170,9 +175,12 @@ impl ProtocolMachine {
self.build_tracked_datagram(0, FrameKind::Close, Bytes::new()) self.build_tracked_datagram(0, FrameKind::Close, Bytes::new())
.map(ProtocolAction::SendDatagram) .map(ProtocolAction::SendDatagram)
} }
(OstpState::Closing, OstpEvent::Inbound(_)) => { (OstpState::Closing, OstpEvent::Inbound(raw)) => {
// Process final in-flight packets to prevent data loss during teardown.
// The remote may still have data or ACKs in transit when we initiated Close.
let result = self.handle_inbound(raw);
self.state = OstpState::Closed; self.state = OstpState::Closed;
Ok(ProtocolAction::Noop) result
} }
(OstpState::Established, OstpEvent::Tick) => self.handle_tick(), (OstpState::Established, OstpEvent::Tick) => self.handle_tick(),
(OstpState::Closed, _) => Ok(ProtocolAction::Noop), (OstpState::Closed, _) => Ok(ProtocolAction::Noop),
@ -312,17 +320,23 @@ impl ProtocolMachine {
})?; })?;
} }
} else { } else {
// Gap detected! Buffer current packet and request immediate retransmit of the gap packet. // Gap detected! Buffer current packet and request retransmit of the gap packet.
if self.reorder_buffer.len() < self.max_reorder_buffer { if self.reorder_buffer.len() < self.max_reorder_buffer {
self.reorder_buffer.insert(nonce, action); self.reorder_buffer.insert(nonce, action);
} }
// Emit a Nack frame for the lowest missing sequence // Rate-limited NACK: send at most once per 30ms to prevent retransmit storms.
// Under high load with natural UDP reordering, sending a NACK per packet
// causes exponential retransmit explosion that saturates the channel.
let nack_cooldown = Duration::from_millis(30);
if self.last_nack_sent.elapsed() >= nack_cooldown {
self.last_nack_sent = Instant::now();
let nack_payload = self.expected_recv_nonce.to_be_bytes(); let nack_payload = self.expected_recv_nonce.to_be_bytes();
if let Ok(nack_frame) = self.build_control_datagram(0, FrameKind::Nack, Bytes::copy_from_slice(&nack_payload)) { if let Ok(nack_frame) = self.build_control_datagram(0, FrameKind::Nack, Bytes::copy_from_slice(&nack_payload)) {
outbound_actions.push(ProtocolAction::SendDatagram(nack_frame)); outbound_actions.push(ProtocolAction::SendDatagram(nack_frame));
} }
} }
}
if let Some(ack_frame) = self.build_ack_if_due()? { if let Some(ack_frame) = self.build_ack_if_due()? {
outbound_actions.push(ProtocolAction::SendDatagram(ack_frame)); outbound_actions.push(ProtocolAction::SendDatagram(ack_frame));
@ -419,16 +433,22 @@ impl ProtocolMachine {
let now = Instant::now(); let now = Instant::now();
let base_rto_ms = self.rto.as_millis().max(1) as u64; let base_rto_ms = self.rto.as_millis().max(1) as u64;
// Evict zombie frames that exceeded max_retries + grace period.
// Without eviction, unacknowledged frames accumulate forever, consuming memory
// and wasting bandwidth on retransmits that will never be acknowledged.
let grace = self.max_retries.saturating_add(4);
self.sent_history.retain(|f| !f.is_retransmittable || f.retries <= grace);
for frame in self.sent_history.iter_mut() { for frame in self.sent_history.iter_mut() {
if !frame.is_retransmittable { if !frame.is_retransmittable {
continue; continue;
} }
if frame.retries == self.max_retries { if frame.retries >= self.max_retries {
tracing::warn!( tracing::warn!(
"Frame {} exceeded max retries ({}); continuing with backoff", "Frame nonce={} retry {}/{} (backoff active)",
frame.nonce, frame.nonce, frame.retries, self.max_retries
self.max_retries
); );
} }
@ -517,7 +537,12 @@ impl ProtocolMachine {
} }
if ranges.len() > MAX_RANGES { if ranges.len() > MAX_RANGES {
ranges = ranges[ranges.len() - MAX_RANGES..].to_vec(); // Always preserve the cumulative range (index 0) so the sender knows
// all frames up to expected_recv_nonce are received. Truncate SACK ranges.
let mut trimmed = vec![ranges[0]];
let tail_start = ranges.len().saturating_sub(MAX_RANGES - 1);
trimmed.extend_from_slice(&ranges[tail_start..]);
ranges = trimmed;
} }
let mut out = Vec::with_capacity(1 + ranges.len() * 16); let mut out = Vec::with_capacity(1 + ranges.len() * 16);

View File

@ -99,6 +99,10 @@ impl Dispatcher {
if let Some(session_id) = session_id_opt { if let Some(session_id) = session_id_opt {
if let Some(peer_state) = self.peer_machines.get_mut(&session_id) { if let Some(peer_state) = self.peer_machines.get_mut(&session_id) {
// Update address on seamless roaming: remove old mapping to prevent HashMap leak
if peer_state.last_addr != peer {
self.addr_to_session.remove(&peer_state.last_addr);
}
peer_state.last_addr = peer; peer_state.last_addr = peer;
peer_state.last_seen = std::time::Instant::now(); peer_state.last_seen = std::time::Instant::now();
self.addr_to_session.insert(peer, session_id); self.addr_to_session.insert(peer, session_id);
@ -204,6 +208,11 @@ impl Dispatcher {
} }
if !self.replay_cache.contains_key(&payload.to_vec()) { if !self.replay_cache.contains_key(&payload.to_vec()) {
// Hard cap: prevent OOM under DDoS — replay cache grows
// unboundedly between purge ticks without this limit.
if self.replay_cache.len() >= 100_000 {
return Ok(DispatchOutcome::Unauthorized);
}
// §4 fix: hard cap on concurrent sessions to prevent RAM exhaustion // §4 fix: hard cap on concurrent sessions to prevent RAM exhaustion
if self.peer_machines.len() >= MAX_SESSIONS { if self.peer_machines.len() >= MAX_SESSIONS {
return Ok(DispatchOutcome::Unauthorized); return Ok(DispatchOutcome::Unauthorized);

View File

@ -208,7 +208,9 @@ async fn run_server_loop(
debug: bool, debug: bool,
) -> Result<()> { ) -> Result<()> {
let mut remotes: HashMap<(u32, u16), RemoteState> = HashMap::new(); let mut remotes: HashMap<(u32, u16), RemoteState> = HashMap::new();
let (stream_tx, mut stream_rx) = mpsc::channel::<(u32, u16, Vec<u8>)>(10000); // Unbounded channel: bounded(10000) caused TCP-reader tasks to fail under Speedtest load
// when 50+ streams competed for slots. Backpressure is managed at the relay layer instead.
let (stream_tx, mut stream_rx) = mpsc::unbounded_channel::<(u32, u16, Vec<u8>)>();
let (connect_tx, mut connect_rx) = mpsc::unbounded_channel::<(u32, u16, String, Result<(tokio::net::tcp::OwnedWriteHalf, mpsc::Sender<()>), String>)>(); let (connect_tx, mut connect_rx) = mpsc::unbounded_channel::<(u32, u16, String, Result<(tokio::net::tcp::OwnedWriteHalf, mpsc::Sender<()>), String>)>();
let socket = std::sync::Arc::new(socket); let socket = std::sync::Arc::new(socket);
@ -388,7 +390,7 @@ async fn handle_relay_message(
socket: &UdpSocket, socket: &UdpSocket,
remotes: &mut HashMap<(u32, u16), RemoteState>, remotes: &mut HashMap<(u32, u16), RemoteState>,
ui_event_tx: &mpsc::UnboundedSender<UiEvent>, ui_event_tx: &mpsc::UnboundedSender<UiEvent>,
stream_tx: mpsc::Sender<(u32, u16, Vec<u8>)>, stream_tx: mpsc::UnboundedSender<(u32, u16, Vec<u8>)>,
connect_tx: mpsc::UnboundedSender<(u32, u16, String, Result<(tokio::net::tcp::OwnedWriteHalf, mpsc::Sender<()>), String>)>, connect_tx: mpsc::UnboundedSender<(u32, u16, String, Result<(tokio::net::tcp::OwnedWriteHalf, mpsc::Sender<()>), String>)>,
outbound: Option<OutboundConfig>, outbound: Option<OutboundConfig>,
debug: bool, debug: bool,
@ -414,11 +416,11 @@ async fn handle_relay_message(
read_res = reader.read(&mut buf) => { read_res = reader.read(&mut buf) => {
match read_res { match read_res {
Ok(0) | Err(_) => { Ok(0) | Err(_) => {
let _ = stream_tx_clone.send((session_id, stream_id, Vec::new())).await; let _ = stream_tx_clone.send((session_id, stream_id, Vec::new()));
break; break;
} }
Ok(n) => { Ok(n) => {
if stream_tx_clone.send((session_id, stream_id, buf[..n].to_vec())).await.is_err() { if stream_tx_clone.send((session_id, stream_id, buf[..n].to_vec())).is_err() {
break; break;
} }
} }
@ -485,6 +487,7 @@ async fn connect_target(
outbound: Option<&OutboundConfig>, outbound: Option<&OutboundConfig>,
debug: bool, debug: bool,
) -> Result<TcpStream> { ) -> Result<TcpStream> {
let connect_timeout = Duration::from_secs(10);
if let Some(outbound) = outbound { if let Some(outbound) = outbound {
if outbound.enabled { if outbound.enabled {
let action = select_outbound_action(target, outbound, debug).await; let action = select_outbound_action(target, outbound, debug).await;
@ -493,13 +496,19 @@ async fn connect_target(
return match outbound.protocol.as_str() { return match outbound.protocol.as_str() {
"socks5" => connect_via_socks5(&proxy_addr, target).await, "socks5" => connect_via_socks5(&proxy_addr, target).await,
"http" => connect_via_http(&proxy_addr, target).await, "http" => connect_via_http(&proxy_addr, target).await,
_ => TcpStream::connect(target).await.map_err(Into::into), _ => tokio::time::timeout(connect_timeout, TcpStream::connect(target))
.await
.map_err(|_| anyhow::anyhow!("connect timeout ({}s): {}", connect_timeout.as_secs(), target))?
.map_err(Into::into),
}; };
} }
} }
} }
TcpStream::connect(target).await.map_err(Into::into) tokio::time::timeout(connect_timeout, TcpStream::connect(target))
.await
.map_err(|_| anyhow::anyhow!("connect timeout ({}s): {}", connect_timeout.as_secs(), target))?
.map_err(Into::into)
} }
async fn select_outbound_action( async fn select_outbound_action(