New Hot-Reload method + TLS-F New Methods + TLS-F/TCP-S Docs: merge pull request #420 from telemt/flow

New Hot-Reload method + TLS-F New Methods + TLS-F/TCP-S Docs
This commit is contained in:
Alexey 2026-03-14 20:45:47 +03:00 committed by GitHub
commit dda31b3d2f
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
7 changed files with 977 additions and 122 deletions

View File

@ -0,0 +1,274 @@
# TLS-F и TCP-S в Telemt
## Общая архитектура
**Telemt** - это прежде всего реализация **MTProxy**, через которую проходит payload Telegram
Подсистема **TLS-Fronting / TCP-Splitting** служит **маскировочным транспортным слоем**, задача которого - сделать MTProxy-соединение внешне похожим на обычное TLS-подключение к легитимному сайту
Таким образом:
- **MTProxy** - основной функциональный слой Telemt для обработки Telegram-трафика
- **TLS-Fronting / TCP-Splitting** - подсистема маскировки транспорта
С точки зрения сети Telemt ведёт себя как **TLS-сервер**, но фактически:
- валидные MTProxy-клиенты остаются внутри контура Telemt
- любые другие TLS-клиенты проксируются на обычный HTTPS-сервер-заглушку
# Базовый сценарий / Best-practice
Предположим, у вас есть домен:
```
umweltschutz.de
```
### 1 DNS
Вы создаёте A-запись:
```
umweltschutz.de -> A-запись 198.18.88.88
```
где `198.18.88.88` - IP вашего сервера с telemt
### 2 TLS-домен
В конфигурации Telemt:
```
tls_domain = umweltschutz.de
```
Этот домен используется клиентом как SNI в ClientHello
### 3 Сервер-заглушка
Вы поднимаете обычный HTTPS-сервер, например **nginx**, с сертификатом для этого домена.
Он может работать:
- на том же сервере
- на другом сервере
- на другом порту
В конфигурации Telemt:
```
mask_host = 127.0.0.1
mask_port = 8443
```
где `127.0.0.1` - IP сервера-заглушки, а 8443 - порт, который он слушает
Этот сервер нужен **для обработки любых non-MTProxy запросов**
### 4 Работа Telemt
После запуска Telemt действует следующим образом:
1) принимает входящее TCP-соединение
2) анализирует TLS-ClientHello
3) пытается определить, является ли соединение валидным **MTProxy FakeTLS**
Далее работают два варианта логики:
---
# Сценарий 1 - MTProxy клиент с валидным ключом
Если клиент предъявил **валидный MTProxy-ключ**:
- соединение **остаётся внутри Telemt**
- TLS используется только как **транспортная маскировка**
- далее запускается обычная логика **MTProxy**
Для внешнего наблюдателя это выглядит как:
```
TLS connection -> umweltschutz.de
```
Хотя внутри передаётся **MTProto-трафик Telegram**
# Сценарий 2 - обычный TLS-клиент - crawler / scanner / browser
Если Telemt не обнаруживает валидный MTProxy-ключ:
соединение **переключается в режим TCP-Splitting / TCP-Splicing**.
В этом режиме Telemt:
1. открывает новое TCP-соединение к
```
mask_host:mask_port
```
2. начинает **проксировать TCP-трафик**
Важно:
* клиентский TLS-запрос **НЕ модифицируется**
* **ClientHello передаётся "как есть", без изменений**
* **SNI остаётся неизменным**
* Telemt **не завершает TLS-рукопожатие**, а только перенаправляет его на более низком уровне сетевого стека - L4
Таким образом upstream-сервер получает **оригинальное TLS-соединение клиента**:
- если это nginx-заглушка, он просто отдаёт обычный сайт
- для внешнего наблюдателя это выглядит как обычный HTTPS-сервер
# TCP-S / TCP-Splitting / TCP-Splicing
Ключевые свойства механизма:
**Telemt работает как TCP-переключатель:**
1) принимает соединение
2) определяет тип клиента
3) либо:
- обрабатывает MTProxy внутри
- либо проксирует TCP-поток
При проксировании:
- Telemt **разрешает `mask_host` в IP**
- устанавливает TCP-соединение
- начинает **bidirectional TCP relay**
При этом:
- TLS-рукопожатие происходит **между клиентом и `mask_host`**
- Telemt выступает только **на уровне L4 - как TCP-релей**, такой же как HAProxy в TCP-режиме
# Использование чужого домена
Можно использовать и внешний сайт.
Например:
```
tls_domain = github.com
mask_host = github.com
mask_port = 443
```
или
```
mask_host = 140.82.121.4
```
В этом случае:
- цензор видит **TLS-подключение к github.com**
- обычные клиенты/краулер действительно получают **настоящий GitHub**
Telemt просто **проксирует TCP-соединение на GitHub**
# Что видит анализатор трафика?
Для DPI это выглядит так:
```
client -> TLS -> github.com
```
или
```
client -> TLS -> umweltschutz.de
```
TLS-handshake выглядит валидным, SNI соответствует домену, сертификат корректный - от целевого `mask_host:mask_port`
# Что видит сканер / краулер?
Если сканер попытается подключиться:
```
openssl s_client -connect 198.18.88.88:443 -servername umweltschutz.de
```
он получит **обычный HTTPS-сайт-заглушку**
Потому что:
- он не предъявил MTProxy-ключ
- Telemt отправил соединение на `mask_host:mask_port`, на котором находится nginx
# Какую проблему решает TLS-Fronting / TCP-Splitting?
Эта архитектура решает сразу несколько проблем обхода цензуры.
## 1 Закрытие плоскости MTProxy от активного сканирования
Многие цензоры:
- сканируют IP-адреса
- проверяют известные сигнатуры прокси
Telemt отвечает на такие проверки **обычным HTTPS-сайтом**, поэтому прокси невозможно обнаружить простым сканированием
---
## 2 Маскировка трафика под легитимный TLS
Для DPI-систем соединение выглядит как:
```
обычный TLS-трафик к популярному домену
```
Это делает блокировку значительно сложнее и непредсказуемее
---
## 3 Устойчивость к протокольному анализу
MTProxy трафик проходит **внутри TLS-like-потока**, поэтому:
- не видны характерные сигнатуры MTProto
- соединение выглядит как обычный HTTPS
---
## 4 Правдоподобное поведение сервера
Даже если краулер:
- подключится сам
- выполнит TLS-handshake
- попытается получить HTTP-ответ
он увидит **реальный сайт**, а не telemt
Это устраняет один из главных признаков для антифрод-краулеров мобильных операторов
# Схема
```text
Client
│ TCP
V
Telemt
├── valid MTProxy key
│ │
│ V
│ MTProxy logic
└── обычный TLS клиент
V
TCP-Splitting
V
mask_host:mask_port
```

View File

@ -21,9 +21,11 @@
//! `network.*`, `use_middle_proxy`) are **not** applied; a warning is emitted.
//! Non-hot changes are never mixed into the runtime config snapshot.
use std::collections::BTreeSet;
use std::net::IpAddr;
use std::path::PathBuf;
use std::sync::Arc;
use std::path::{Path, PathBuf};
use std::sync::{Arc, RwLock as StdRwLock};
use std::time::Duration;
use notify::{EventKind, RecursiveMode, Watcher, recommended_watcher};
use tokio::sync::{mpsc, watch};
@ -33,7 +35,10 @@ use crate::config::{
LogLevel, MeBindStaleMode, MeFloorMode, MeSocksKdfPolicy, MeTelemetryLevel,
MeWriterPickMode,
};
use super::load::ProxyConfig;
use super::load::{LoadedConfig, ProxyConfig};
const HOT_RELOAD_STABLE_SNAPSHOTS: u8 = 2;
const HOT_RELOAD_DEBOUNCE: Duration = Duration::from_millis(50);
// ── Hot fields ────────────────────────────────────────────────────────────────
@ -287,6 +292,149 @@ fn listeners_equal(
})
}
#[derive(Debug, Clone, Default, PartialEq, Eq)]
struct WatchManifest {
files: BTreeSet<PathBuf>,
dirs: BTreeSet<PathBuf>,
}
impl WatchManifest {
fn from_source_files(source_files: &[PathBuf]) -> Self {
let mut files = BTreeSet::new();
let mut dirs = BTreeSet::new();
for path in source_files {
let normalized = normalize_watch_path(path);
files.insert(normalized.clone());
if let Some(parent) = normalized.parent() {
dirs.insert(parent.to_path_buf());
}
}
Self { files, dirs }
}
fn matches_event_paths(&self, event_paths: &[PathBuf]) -> bool {
event_paths
.iter()
.map(|path| normalize_watch_path(path))
.any(|path| self.files.contains(&path))
}
}
#[derive(Debug, Default)]
struct ReloadState {
applied_snapshot_hash: Option<u64>,
candidate_snapshot_hash: Option<u64>,
candidate_hits: u8,
}
impl ReloadState {
fn new(applied_snapshot_hash: Option<u64>) -> Self {
Self {
applied_snapshot_hash,
candidate_snapshot_hash: None,
candidate_hits: 0,
}
}
fn is_applied(&self, hash: u64) -> bool {
self.applied_snapshot_hash == Some(hash)
}
fn observe_candidate(&mut self, hash: u64) -> u8 {
if self.candidate_snapshot_hash == Some(hash) {
self.candidate_hits = self.candidate_hits.saturating_add(1);
} else {
self.candidate_snapshot_hash = Some(hash);
self.candidate_hits = 1;
}
self.candidate_hits
}
fn reset_candidate(&mut self) {
self.candidate_snapshot_hash = None;
self.candidate_hits = 0;
}
fn mark_applied(&mut self, hash: u64) {
self.applied_snapshot_hash = Some(hash);
self.reset_candidate();
}
}
fn normalize_watch_path(path: &Path) -> PathBuf {
path.canonicalize().unwrap_or_else(|_| {
if path.is_absolute() {
path.to_path_buf()
} else {
std::env::current_dir()
.map(|cwd| cwd.join(path))
.unwrap_or_else(|_| path.to_path_buf())
}
})
}
fn sync_watch_paths<W: Watcher>(
watcher: &mut W,
current: &BTreeSet<PathBuf>,
next: &BTreeSet<PathBuf>,
recursive_mode: RecursiveMode,
kind: &str,
) {
for path in current.difference(next) {
if let Err(e) = watcher.unwatch(path) {
warn!(path = %path.display(), error = %e, "config watcher: failed to unwatch {kind}");
}
}
for path in next.difference(current) {
if let Err(e) = watcher.watch(path, recursive_mode) {
warn!(path = %path.display(), error = %e, "config watcher: failed to watch {kind}");
}
}
}
fn apply_watch_manifest<W1: Watcher, W2: Watcher>(
notify_watcher: Option<&mut W1>,
poll_watcher: Option<&mut W2>,
manifest_state: &Arc<StdRwLock<WatchManifest>>,
next_manifest: WatchManifest,
) {
let current_manifest = manifest_state
.read()
.map(|manifest| manifest.clone())
.unwrap_or_default();
if current_manifest == next_manifest {
return;
}
if let Some(watcher) = notify_watcher {
sync_watch_paths(
watcher,
&current_manifest.dirs,
&next_manifest.dirs,
RecursiveMode::NonRecursive,
"config directory",
);
}
if let Some(watcher) = poll_watcher {
sync_watch_paths(
watcher,
&current_manifest.files,
&next_manifest.files,
RecursiveMode::NonRecursive,
"config file",
);
}
if let Ok(mut manifest) = manifest_state.write() {
*manifest = next_manifest;
}
}
fn overlay_hot_fields(old: &ProxyConfig, new: &ProxyConfig) -> ProxyConfig {
let mut cfg = old.clone();
@ -970,18 +1118,42 @@ fn reload_config(
log_tx: &watch::Sender<LogLevel>,
detected_ip_v4: Option<IpAddr>,
detected_ip_v6: Option<IpAddr>,
) {
let new_cfg = match ProxyConfig::load(config_path) {
Ok(c) => c,
reload_state: &mut ReloadState,
) -> Option<WatchManifest> {
let loaded = match ProxyConfig::load_with_metadata(config_path) {
Ok(loaded) => loaded,
Err(e) => {
reload_state.reset_candidate();
error!("config reload: failed to parse {:?}: {}", config_path, e);
return;
return None;
}
};
let LoadedConfig {
config: new_cfg,
source_files,
rendered_hash,
} = loaded;
let next_manifest = WatchManifest::from_source_files(&source_files);
if let Err(e) = new_cfg.validate() {
reload_state.reset_candidate();
error!("config reload: validation failed: {}; keeping old config", e);
return;
return Some(next_manifest);
}
if reload_state.is_applied(rendered_hash) {
return Some(next_manifest);
}
let candidate_hits = reload_state.observe_candidate(rendered_hash);
if candidate_hits < HOT_RELOAD_STABLE_SNAPSHOTS {
info!(
snapshot_hash = rendered_hash,
candidate_hits,
required_hits = HOT_RELOAD_STABLE_SNAPSHOTS,
"config reload: candidate snapshot observed but not stable yet"
);
return Some(next_manifest);
}
let old_cfg = config_tx.borrow().clone();
@ -996,17 +1168,19 @@ fn reload_config(
}
if !hot_changed {
return;
reload_state.mark_applied(rendered_hash);
return Some(next_manifest);
}
if old_hot.dns_overrides != applied_hot.dns_overrides
&& let Err(e) = crate::network::dns_overrides::install_entries(&applied_hot.dns_overrides)
{
reload_state.reset_candidate();
error!(
"config reload: invalid network.dns_overrides: {}; keeping old config",
e
);
return;
return Some(next_manifest);
}
log_changes(
@ -1018,6 +1192,8 @@ fn reload_config(
detected_ip_v6,
);
config_tx.send(Arc::new(applied_cfg)).ok();
reload_state.mark_applied(rendered_hash);
Some(next_manifest)
}
// ── Public API ────────────────────────────────────────────────────────────────
@ -1040,80 +1216,86 @@ pub fn spawn_config_watcher(
let (config_tx, config_rx) = watch::channel(initial);
let (log_tx, log_rx) = watch::channel(initial_level);
// Bridge: sync notify callbacks → async task via mpsc.
let (notify_tx, mut notify_rx) = mpsc::channel::<()>(4);
let config_path = normalize_watch_path(&config_path);
let initial_loaded = ProxyConfig::load_with_metadata(&config_path).ok();
let initial_manifest = initial_loaded
.as_ref()
.map(|loaded| WatchManifest::from_source_files(&loaded.source_files))
.unwrap_or_else(|| WatchManifest::from_source_files(std::slice::from_ref(&config_path)));
let initial_snapshot_hash = initial_loaded.as_ref().map(|loaded| loaded.rendered_hash);
// Canonicalize so path matches what notify returns (absolute) in events.
let config_path = match config_path.canonicalize() {
Ok(p) => p,
Err(_) => config_path.to_path_buf(),
};
// Watch the parent directory rather than the file itself, because many
// editors (vim, nano) and systemd write via rename, which would cause
// inotify to lose track of the original inode.
let watch_dir = config_path
.parent()
.unwrap_or_else(|| std::path::Path::new("."))
.to_path_buf();
// ── inotify watcher (instant on local fs) ────────────────────────────
let config_file = config_path.clone();
let tx_inotify = notify_tx.clone();
let inotify_ok = match recommended_watcher(move |res: notify::Result<notify::Event>| {
let Ok(event) = res else { return };
let is_our_file = event.paths.iter().any(|p| p == &config_file);
if !is_our_file { return; }
if matches!(event.kind, EventKind::Modify(_) | EventKind::Create(_) | EventKind::Remove(_)) {
let _ = tx_inotify.try_send(());
}
}) {
Ok(mut w) => match w.watch(&watch_dir, RecursiveMode::NonRecursive) {
Ok(()) => {
info!("config watcher: inotify active on {:?}", config_path);
Box::leak(Box::new(w));
true
}
Err(e) => { warn!("config watcher: inotify watch failed: {}", e); false }
},
Err(e) => { warn!("config watcher: inotify unavailable: {}", e); false }
};
// ── poll watcher (always active, fixes Docker bind mounts / NFS) ─────
// inotify does not receive events for files mounted from the host into
// a container. PollWatcher compares file contents every 3 s and fires
// on any change regardless of the underlying fs.
let config_file2 = config_path.clone();
let tx_poll = notify_tx.clone();
match notify::poll::PollWatcher::new(
move |res: notify::Result<notify::Event>| {
let Ok(event) = res else { return };
let is_our_file = event.paths.iter().any(|p| p == &config_file2);
if !is_our_file { return; }
if matches!(event.kind, EventKind::Modify(_) | EventKind::Create(_) | EventKind::Remove(_)) {
let _ = tx_poll.try_send(());
}
},
notify::Config::default()
.with_poll_interval(std::time::Duration::from_secs(3))
.with_compare_contents(true),
) {
Ok(mut w) => match w.watch(&config_path, RecursiveMode::NonRecursive) {
Ok(()) => {
if inotify_ok {
info!("config watcher: poll watcher also active (Docker/NFS safe)");
} else {
info!("config watcher: poll watcher active on {:?} (3s interval)", config_path);
}
Box::leak(Box::new(w));
}
Err(e) => warn!("config watcher: poll watch failed: {}", e),
},
Err(e) => warn!("config watcher: poll watcher unavailable: {}", e),
}
// ── event loop ───────────────────────────────────────────────────────
tokio::spawn(async move {
let (notify_tx, mut notify_rx) = mpsc::channel::<()>(4);
let manifest_state = Arc::new(StdRwLock::new(WatchManifest::default()));
let mut reload_state = ReloadState::new(initial_snapshot_hash);
let tx_inotify = notify_tx.clone();
let manifest_for_inotify = manifest_state.clone();
let mut inotify_watcher = match recommended_watcher(move |res: notify::Result<notify::Event>| {
let Ok(event) = res else { return };
if !matches!(event.kind, EventKind::Modify(_) | EventKind::Create(_) | EventKind::Remove(_)) {
return;
}
let is_our_file = manifest_for_inotify
.read()
.map(|manifest| manifest.matches_event_paths(&event.paths))
.unwrap_or(false);
if is_our_file {
let _ = tx_inotify.try_send(());
}
}) {
Ok(watcher) => Some(watcher),
Err(e) => {
warn!("config watcher: inotify unavailable: {}", e);
None
}
};
apply_watch_manifest(
inotify_watcher.as_mut(),
Option::<&mut notify::poll::PollWatcher>::None,
&manifest_state,
initial_manifest.clone(),
);
if inotify_watcher.is_some() {
info!("config watcher: inotify active on {:?}", config_path);
}
let tx_poll = notify_tx.clone();
let manifest_for_poll = manifest_state.clone();
let mut poll_watcher = match notify::poll::PollWatcher::new(
move |res: notify::Result<notify::Event>| {
let Ok(event) = res else { return };
if !matches!(event.kind, EventKind::Modify(_) | EventKind::Create(_) | EventKind::Remove(_)) {
return;
}
let is_our_file = manifest_for_poll
.read()
.map(|manifest| manifest.matches_event_paths(&event.paths))
.unwrap_or(false);
if is_our_file {
let _ = tx_poll.try_send(());
}
},
notify::Config::default()
.with_poll_interval(Duration::from_secs(3))
.with_compare_contents(true),
) {
Ok(watcher) => Some(watcher),
Err(e) => {
warn!("config watcher: poll watcher unavailable: {}", e);
None
}
};
apply_watch_manifest(
Option::<&mut notify::RecommendedWatcher>::None,
poll_watcher.as_mut(),
&manifest_state,
initial_manifest.clone(),
);
if poll_watcher.is_some() {
info!("config watcher: poll watcher active (Docker/NFS safe)");
}
#[cfg(unix)]
let mut sighup = {
use tokio::signal::unix::{SignalKind, signal};
@ -1133,11 +1315,25 @@ pub fn spawn_config_watcher(
#[cfg(not(unix))]
if notify_rx.recv().await.is_none() { break; }
// Debounce: drain extra events that arrive within 50 ms.
tokio::time::sleep(std::time::Duration::from_millis(50)).await;
// Debounce: drain extra events that arrive within a short quiet window.
tokio::time::sleep(HOT_RELOAD_DEBOUNCE).await;
while notify_rx.try_recv().is_ok() {}
reload_config(&config_path, &config_tx, &log_tx, detected_ip_v4, detected_ip_v6);
if let Some(next_manifest) = reload_config(
&config_path,
&config_tx,
&log_tx,
detected_ip_v4,
detected_ip_v6,
&mut reload_state,
) {
apply_watch_manifest(
inotify_watcher.as_mut(),
poll_watcher.as_mut(),
&manifest_state,
next_manifest,
);
}
}
});
@ -1152,6 +1348,40 @@ mod tests {
ProxyConfig::default()
}
fn write_reload_config(path: &Path, ad_tag: Option<&str>, server_port: Option<u16>) {
let mut config = String::from(
r#"
[censorship]
tls_domain = "example.com"
[access.users]
user = "00000000000000000000000000000000"
"#,
);
if ad_tag.is_some() {
config.push_str("\n[general]\n");
if let Some(tag) = ad_tag {
config.push_str(&format!("ad_tag = \"{tag}\"\n"));
}
}
if let Some(port) = server_port {
config.push_str("\n[server]\n");
config.push_str(&format!("port = {port}\n"));
}
std::fs::write(path, config).unwrap();
}
fn temp_config_path(prefix: &str) -> PathBuf {
let nonce = std::time::SystemTime::now()
.duration_since(std::time::UNIX_EPOCH)
.unwrap()
.as_nanos();
std::env::temp_dir().join(format!("{prefix}_{nonce}.toml"))
}
#[test]
fn overlay_applies_hot_and_preserves_non_hot() {
let old = sample_config();
@ -1219,4 +1449,61 @@ mod tests {
assert_eq!(applied.general.use_middle_proxy, old.general.use_middle_proxy);
assert!(!config_equal(&applied, &new));
}
#[test]
fn reload_requires_stable_snapshot_before_hot_apply() {
let initial_tag = "11111111111111111111111111111111";
let final_tag = "22222222222222222222222222222222";
let path = temp_config_path("telemt_hot_reload_stable");
write_reload_config(&path, Some(initial_tag), None);
let initial_cfg = Arc::new(ProxyConfig::load(&path).unwrap());
let initial_hash = ProxyConfig::load_with_metadata(&path).unwrap().rendered_hash;
let (config_tx, _config_rx) = watch::channel(initial_cfg.clone());
let (log_tx, _log_rx) = watch::channel(initial_cfg.general.log_level.clone());
let mut reload_state = ReloadState::new(Some(initial_hash));
write_reload_config(&path, None, None);
reload_config(&path, &config_tx, &log_tx, None, None, &mut reload_state).unwrap();
assert_eq!(
config_tx.borrow().general.ad_tag.as_deref(),
Some(initial_tag)
);
write_reload_config(&path, Some(final_tag), None);
reload_config(&path, &config_tx, &log_tx, None, None, &mut reload_state).unwrap();
assert_eq!(
config_tx.borrow().general.ad_tag.as_deref(),
Some(initial_tag)
);
reload_config(&path, &config_tx, &log_tx, None, None, &mut reload_state).unwrap();
assert_eq!(config_tx.borrow().general.ad_tag.as_deref(), Some(final_tag));
let _ = std::fs::remove_file(path);
}
#[test]
fn reload_keeps_hot_apply_when_non_hot_fields_change() {
let initial_tag = "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa";
let final_tag = "bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb";
let path = temp_config_path("telemt_hot_reload_mixed");
write_reload_config(&path, Some(initial_tag), None);
let initial_cfg = Arc::new(ProxyConfig::load(&path).unwrap());
let initial_hash = ProxyConfig::load_with_metadata(&path).unwrap().rendered_hash;
let (config_tx, _config_rx) = watch::channel(initial_cfg.clone());
let (log_tx, _log_rx) = watch::channel(initial_cfg.general.log_level.clone());
let mut reload_state = ReloadState::new(Some(initial_hash));
write_reload_config(&path, Some(final_tag), Some(initial_cfg.server.port + 1));
reload_config(&path, &config_tx, &log_tx, None, None, &mut reload_state).unwrap();
reload_config(&path, &config_tx, &log_tx, None, None, &mut reload_state).unwrap();
let applied = config_tx.borrow().clone();
assert_eq!(applied.general.ad_tag.as_deref(), Some(final_tag));
assert_eq!(applied.server.port, initial_cfg.server.port);
let _ = std::fs::remove_file(path);
}
}

View File

@ -1,8 +1,9 @@
#![allow(deprecated)]
use std::collections::HashMap;
use std::collections::{BTreeSet, HashMap};
use std::hash::{DefaultHasher, Hash, Hasher};
use std::net::{IpAddr, SocketAddr};
use std::path::Path;
use std::path::{Path, PathBuf};
use rand::Rng;
use tracing::warn;
@ -13,7 +14,37 @@ use crate::error::{ProxyError, Result};
use super::defaults::*;
use super::types::*;
fn preprocess_includes(content: &str, base_dir: &Path, depth: u8) -> Result<String> {
#[derive(Debug, Clone)]
pub(crate) struct LoadedConfig {
pub(crate) config: ProxyConfig,
pub(crate) source_files: Vec<PathBuf>,
pub(crate) rendered_hash: u64,
}
fn normalize_config_path(path: &Path) -> PathBuf {
path.canonicalize().unwrap_or_else(|_| {
if path.is_absolute() {
path.to_path_buf()
} else {
std::env::current_dir()
.map(|cwd| cwd.join(path))
.unwrap_or_else(|_| path.to_path_buf())
}
})
}
fn hash_rendered_snapshot(rendered: &str) -> u64 {
let mut hasher = DefaultHasher::new();
rendered.hash(&mut hasher);
hasher.finish()
}
fn preprocess_includes(
content: &str,
base_dir: &Path,
depth: u8,
source_files: &mut BTreeSet<PathBuf>,
) -> Result<String> {
if depth > 10 {
return Err(ProxyError::Config("Include depth > 10".into()));
}
@ -25,10 +56,16 @@ fn preprocess_includes(content: &str, base_dir: &Path, depth: u8) -> Result<Stri
if let Some(rest) = rest.strip_prefix('=') {
let path_str = rest.trim().trim_matches('"');
let resolved = base_dir.join(path_str);
source_files.insert(normalize_config_path(&resolved));
let included = std::fs::read_to_string(&resolved)
.map_err(|e| ProxyError::Config(e.to_string()))?;
let included_dir = resolved.parent().unwrap_or(base_dir);
output.push_str(&preprocess_includes(&included, included_dir, depth + 1)?);
output.push_str(&preprocess_includes(
&included,
included_dir,
depth + 1,
source_files,
)?);
output.push('\n');
continue;
}
@ -138,10 +175,16 @@ pub struct ProxyConfig {
impl ProxyConfig {
pub fn load<P: AsRef<Path>>(path: P) -> Result<Self> {
let content =
std::fs::read_to_string(&path).map_err(|e| ProxyError::Config(e.to_string()))?;
let base_dir = path.as_ref().parent().unwrap_or(Path::new("."));
let processed = preprocess_includes(&content, base_dir, 0)?;
Self::load_with_metadata(path).map(|loaded| loaded.config)
}
pub(crate) fn load_with_metadata<P: AsRef<Path>>(path: P) -> Result<LoadedConfig> {
let path = path.as_ref();
let content = std::fs::read_to_string(path).map_err(|e| ProxyError::Config(e.to_string()))?;
let base_dir = path.parent().unwrap_or(Path::new("."));
let mut source_files = BTreeSet::new();
source_files.insert(normalize_config_path(path));
let processed = preprocess_includes(&content, base_dir, 0, &mut source_files)?;
let parsed_toml: toml::Value =
toml::from_str(&processed).map_err(|e| ProxyError::Config(e.to_string()))?;
@ -786,7 +829,11 @@ impl ProxyConfig {
.entry("203".to_string())
.or_insert_with(|| vec!["91.105.192.100:443".to_string()]);
Ok(config)
Ok(LoadedConfig {
config,
source_files: source_files.into_iter().collect(),
rendered_hash: hash_rendered_snapshot(&processed),
})
}
pub fn validate(&self) -> Result<()> {
@ -1111,6 +1158,48 @@ mod tests {
);
}
#[test]
fn load_with_metadata_collects_include_files() {
let nonce = std::time::SystemTime::now()
.duration_since(std::time::UNIX_EPOCH)
.unwrap()
.as_nanos();
let dir = std::env::temp_dir().join(format!("telemt_load_metadata_{nonce}"));
std::fs::create_dir_all(&dir).unwrap();
let main_path = dir.join("config.toml");
let include_path = dir.join("included.toml");
std::fs::write(
&include_path,
r#"
[access.users]
user = "00000000000000000000000000000000"
"#,
)
.unwrap();
std::fs::write(
&main_path,
r#"
include = "included.toml"
[censorship]
tls_domain = "example.com"
"#,
)
.unwrap();
let loaded = ProxyConfig::load_with_metadata(&main_path).unwrap();
let main_normalized = normalize_config_path(&main_path);
let include_normalized = normalize_config_path(&include_path);
assert!(loaded.source_files.contains(&main_normalized));
assert!(loaded.source_files.contains(&include_normalized));
let _ = std::fs::remove_file(main_path);
let _ = std::fs::remove_file(include_path);
let _ = std::fs::remove_dir(dir);
}
#[test]
fn dc_overrides_inject_dc203_default() {
let toml = r#"

View File

@ -8,7 +8,9 @@ use tokio::sync::RwLock;
use tokio::time::sleep;
use tracing::{debug, warn, info};
use crate::tls_front::types::{CachedTlsData, ParsedServerHello, TlsFetchResult};
use crate::tls_front::types::{
CachedTlsData, ParsedServerHello, TlsBehaviorProfile, TlsFetchResult,
};
/// Lightweight in-memory + optional on-disk cache for TLS fronting data.
#[derive(Debug)]
@ -37,6 +39,7 @@ impl TlsFrontCache {
cert_payload: None,
app_data_records_sizes: vec![default_len],
total_app_data_len: default_len,
behavior_profile: TlsBehaviorProfile::default(),
fetched_at: SystemTime::now(),
domain: "default".to_string(),
});
@ -189,6 +192,7 @@ impl TlsFrontCache {
cert_payload: fetched.cert_payload,
app_data_records_sizes: fetched.app_data_records_sizes.clone(),
total_app_data_len: fetched.total_app_data_len,
behavior_profile: fetched.behavior_profile,
fetched_at: SystemTime::now(),
domain: domain.to_string(),
};

View File

@ -3,7 +3,7 @@ use crate::protocol::constants::{
TLS_RECORD_APPLICATION, TLS_RECORD_CHANGE_CIPHER, TLS_RECORD_HANDSHAKE, TLS_VERSION,
};
use crate::protocol::tls::{TLS_DIGEST_LEN, TLS_DIGEST_POS, gen_fake_x25519_key};
use crate::tls_front::types::{CachedTlsData, ParsedCertificateInfo};
use crate::tls_front::types::{CachedTlsData, ParsedCertificateInfo, TlsProfileSource};
const MIN_APP_DATA: usize = 64;
const MAX_APP_DATA: usize = 16640; // RFC 8446 §5.2 allows up to 2^14 + 256
@ -108,14 +108,12 @@ pub fn build_emulated_server_hello(
) -> Vec<u8> {
// --- ServerHello ---
let mut extensions = Vec::new();
// KeyShare (x25519)
let key = gen_fake_x25519_key(rng);
extensions.extend_from_slice(&0x0033u16.to_be_bytes()); // key_share
extensions.extend_from_slice(&(2 + 2 + 32u16).to_be_bytes()); // len
extensions.extend_from_slice(&0x001du16.to_be_bytes()); // X25519
extensions.extend_from_slice(&0x0033u16.to_be_bytes());
extensions.extend_from_slice(&(2 + 2 + 32u16).to_be_bytes());
extensions.extend_from_slice(&0x001du16.to_be_bytes());
extensions.extend_from_slice(&(32u16).to_be_bytes());
extensions.extend_from_slice(&key);
// supported_versions (TLS1.3)
extensions.extend_from_slice(&0x002bu16.to_be_bytes());
extensions.extend_from_slice(&(2u16).to_be_bytes());
extensions.extend_from_slice(&0x0304u16.to_be_bytes());
@ -128,7 +126,6 @@ pub fn build_emulated_server_hello(
extensions.push(alpn_proto.len() as u8);
extensions.extend_from_slice(alpn_proto);
}
let extensions_len = extensions.len() as u16;
let body_len = 2 + // version
@ -173,11 +170,22 @@ pub fn build_emulated_server_hello(
];
// --- ApplicationData (fake encrypted records) ---
// Use the same number and sizes of ApplicationData records as the cached server.
let mut sizes = cached.app_data_records_sizes.clone();
if sizes.is_empty() {
sizes.push(cached.total_app_data_len.max(1024));
}
let sizes = match cached.behavior_profile.source {
TlsProfileSource::Raw | TlsProfileSource::Merged => cached
.app_data_records_sizes
.first()
.copied()
.or_else(|| cached.behavior_profile.app_data_record_sizes.first().copied())
.map(|size| vec![size])
.unwrap_or_else(|| vec![cached.total_app_data_len.max(1024)]),
_ => {
let mut sizes = cached.app_data_records_sizes.clone();
if sizes.is_empty() {
sizes.push(cached.total_app_data_len.max(1024));
}
sizes
}
};
let mut sizes = jitter_and_clamp_sizes(&sizes, rng);
let compact_payload = cached
.cert_info
@ -269,7 +277,9 @@ pub fn build_emulated_server_hello(
mod tests {
use std::time::SystemTime;
use crate::tls_front::types::{CachedTlsData, ParsedServerHello, TlsCertPayload};
use crate::tls_front::types::{
CachedTlsData, ParsedServerHello, TlsBehaviorProfile, TlsCertPayload, TlsProfileSource,
};
use super::build_emulated_server_hello;
use crate::crypto::SecureRandom;
@ -300,6 +310,7 @@ mod tests {
cert_payload,
app_data_records_sizes: vec![64],
total_app_data_len: 64,
behavior_profile: TlsBehaviorProfile::default(),
fetched_at: SystemTime::now(),
domain: "example.com".to_string(),
}
@ -385,4 +396,34 @@ mod tests {
let payload = first_app_data_payload(&response);
assert!(payload.starts_with(b"CN=example.com"));
}
#[test]
fn test_build_emulated_server_hello_ignores_tail_records_for_raw_profile() {
let mut cached = make_cached(None);
cached.app_data_records_sizes = vec![27, 3905, 537, 69];
cached.total_app_data_len = 4538;
cached.behavior_profile.source = TlsProfileSource::Merged;
cached.behavior_profile.app_data_record_sizes = vec![27, 3905, 537];
cached.behavior_profile.ticket_record_sizes = vec![69];
let rng = SecureRandom::new();
let response = build_emulated_server_hello(
b"secret",
&[0x12; 32],
&[0x34; 16],
&cached,
false,
&rng,
None,
0,
);
let hello_len = u16::from_be_bytes([response[3], response[4]]) as usize;
let ccs_start = 5 + hello_len;
let app_start = ccs_start + 6;
let app_len = u16::from_be_bytes([response[app_start + 3], response[app_start + 4]]) as usize;
assert_eq!(response[app_start], TLS_RECORD_APPLICATION);
assert_eq!(app_start + 5 + app_len, response.len());
}
}

View File

@ -21,14 +21,18 @@ use x509_parser::certificate::X509Certificate;
use crate::crypto::SecureRandom;
use crate::network::dns_overrides::resolve_socket_addr;
use crate::protocol::constants::{TLS_RECORD_APPLICATION, TLS_RECORD_HANDSHAKE};
use crate::protocol::constants::{
TLS_RECORD_APPLICATION, TLS_RECORD_CHANGE_CIPHER, TLS_RECORD_HANDSHAKE,
};
use crate::transport::proxy_protocol::{ProxyProtocolV1Builder, ProxyProtocolV2Builder};
use crate::tls_front::types::{
ParsedCertificateInfo,
ParsedServerHello,
TlsBehaviorProfile,
TlsCertPayload,
TlsExtension,
TlsFetchResult,
TlsProfileSource,
};
/// No-op verifier: accept any certificate (we only need lengths and metadata).
@ -282,6 +286,41 @@ fn parse_server_hello(body: &[u8]) -> Option<ParsedServerHello> {
})
}
fn derive_behavior_profile(records: &[(u8, Vec<u8>)]) -> TlsBehaviorProfile {
let mut change_cipher_spec_count = 0u8;
let mut app_data_record_sizes = Vec::new();
for (record_type, body) in records {
match *record_type {
TLS_RECORD_CHANGE_CIPHER => {
change_cipher_spec_count = change_cipher_spec_count.saturating_add(1);
}
TLS_RECORD_APPLICATION => {
app_data_record_sizes.push(body.len());
}
_ => {}
}
}
let mut ticket_record_sizes = Vec::new();
while app_data_record_sizes
.last()
.is_some_and(|size| *size <= 256 && ticket_record_sizes.len() < 2)
{
if let Some(size) = app_data_record_sizes.pop() {
ticket_record_sizes.push(size);
}
}
ticket_record_sizes.reverse();
TlsBehaviorProfile {
change_cipher_spec_count: change_cipher_spec_count.max(1),
app_data_record_sizes,
ticket_record_sizes,
source: TlsProfileSource::Raw,
}
}
fn parse_cert_info(certs: &[CertificateDer<'static>]) -> Option<ParsedCertificateInfo> {
let first = certs.first()?;
let (_rem, cert) = X509Certificate::from_der(first.as_ref()).ok()?;
@ -443,39 +482,50 @@ where
.await??;
let mut records = Vec::new();
// Read up to 4 records: ServerHello, CCS, and up to two ApplicationData.
for _ in 0..4 {
let mut app_records_seen = 0usize;
// Read a bounded encrypted flight: ServerHello, CCS, certificate-like data,
// and a small number of ticket-like tail records.
for _ in 0..8 {
match timeout(connect_timeout, read_tls_record(&mut stream)).await {
Ok(Ok(rec)) => records.push(rec),
Ok(Ok(rec)) => {
if rec.0 == TLS_RECORD_APPLICATION {
app_records_seen += 1;
}
records.push(rec);
}
Ok(Err(e)) => return Err(e),
Err(_) => break,
}
if records.len() >= 3 && records.iter().any(|(t, _)| *t == TLS_RECORD_APPLICATION) {
if app_records_seen >= 4 {
break;
}
}
let mut app_sizes = Vec::new();
let mut server_hello = None;
for (t, body) in &records {
if *t == TLS_RECORD_HANDSHAKE && server_hello.is_none() {
server_hello = parse_server_hello(body);
} else if *t == TLS_RECORD_APPLICATION {
app_sizes.push(body.len());
}
}
let parsed = server_hello.ok_or_else(|| anyhow!("ServerHello not received"))?;
let behavior_profile = derive_behavior_profile(&records);
let mut app_sizes = behavior_profile.app_data_record_sizes.clone();
app_sizes.extend_from_slice(&behavior_profile.ticket_record_sizes);
let total_app_data_len = app_sizes.iter().sum::<usize>().max(1024);
let app_data_records_sizes = behavior_profile
.app_data_record_sizes
.first()
.copied()
.or_else(|| behavior_profile.ticket_record_sizes.first().copied())
.map(|size| vec![size])
.unwrap_or_else(|| vec![total_app_data_len]);
Ok(TlsFetchResult {
server_hello_parsed: parsed,
app_data_records_sizes: if app_sizes.is_empty() {
vec![total_app_data_len]
} else {
app_sizes
},
app_data_records_sizes,
total_app_data_len,
behavior_profile,
cert_info: None,
cert_payload: None,
})
@ -608,6 +658,12 @@ where
server_hello_parsed: parsed,
app_data_records_sizes: app_data_records_sizes.clone(),
total_app_data_len: app_data_records_sizes.iter().sum(),
behavior_profile: TlsBehaviorProfile {
change_cipher_spec_count: 1,
app_data_record_sizes: app_data_records_sizes,
ticket_record_sizes: Vec::new(),
source: TlsProfileSource::Rustls,
},
cert_info,
cert_payload,
})
@ -706,6 +762,7 @@ pub async fn fetch_real_tls(
if let Some(mut raw) = raw_result {
raw.cert_info = rustls_result.cert_info;
raw.cert_payload = rustls_result.cert_payload;
raw.behavior_profile.source = TlsProfileSource::Merged;
debug!(sni = %sni, "Fetched TLS metadata via raw probe + rustls cert chain");
Ok(raw)
} else {
@ -725,7 +782,11 @@ pub async fn fetch_real_tls(
#[cfg(test)]
mod tests {
use super::encode_tls13_certificate_message;
use super::{derive_behavior_profile, encode_tls13_certificate_message};
use crate::protocol::constants::{
TLS_RECORD_APPLICATION, TLS_RECORD_CHANGE_CIPHER, TLS_RECORD_HANDSHAKE,
};
use crate::tls_front::types::TlsProfileSource;
fn read_u24(bytes: &[u8]) -> usize {
((bytes[0] as usize) << 16) | ((bytes[1] as usize) << 8) | (bytes[2] as usize)
@ -753,4 +814,20 @@ mod tests {
fn test_encode_tls13_certificate_message_empty_chain() {
assert!(encode_tls13_certificate_message(&[]).is_none());
}
#[test]
fn test_derive_behavior_profile_splits_ticket_like_tail_records() {
let profile = derive_behavior_profile(&[
(TLS_RECORD_HANDSHAKE, vec![0u8; 90]),
(TLS_RECORD_CHANGE_CIPHER, vec![0x01]),
(TLS_RECORD_APPLICATION, vec![0u8; 1400]),
(TLS_RECORD_APPLICATION, vec![0u8; 220]),
(TLS_RECORD_APPLICATION, vec![0u8; 180]),
]);
assert_eq!(profile.change_cipher_spec_count, 1);
assert_eq!(profile.app_data_record_sizes, vec![1400]);
assert_eq!(profile.ticket_record_sizes, vec![220, 180]);
assert_eq!(profile.source, TlsProfileSource::Raw);
}
}

View File

@ -39,6 +39,53 @@ pub struct TlsCertPayload {
pub certificate_message: Vec<u8>,
}
/// Provenance of the cached TLS behavior profile.
#[derive(Debug, Clone, Copy, Serialize, Deserialize, PartialEq, Eq, Default)]
#[serde(rename_all = "snake_case")]
pub enum TlsProfileSource {
/// Built from hardcoded defaults or legacy cache entries.
#[default]
Default,
/// Derived from raw TLS record capture only.
Raw,
/// Derived from rustls-only metadata fallback.
Rustls,
/// Merged from raw TLS capture and rustls certificate metadata.
Merged,
}
/// Coarse-grained TLS response behavior captured per SNI.
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct TlsBehaviorProfile {
/// Number of ChangeCipherSpec records observed before encrypted flight.
#[serde(default = "default_change_cipher_spec_count")]
pub change_cipher_spec_count: u8,
/// Sizes of the primary encrypted flight records carrying cert-like payload.
#[serde(default)]
pub app_data_record_sizes: Vec<usize>,
/// Sizes of small tail ApplicationData records that look like tickets.
#[serde(default)]
pub ticket_record_sizes: Vec<usize>,
/// Source of this behavior profile.
#[serde(default)]
pub source: TlsProfileSource,
}
fn default_change_cipher_spec_count() -> u8 {
1
}
impl Default for TlsBehaviorProfile {
fn default() -> Self {
Self {
change_cipher_spec_count: default_change_cipher_spec_count(),
app_data_record_sizes: Vec::new(),
ticket_record_sizes: Vec::new(),
source: TlsProfileSource::Default,
}
}
}
/// Cached data per SNI used by the emulator.
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct CachedTlsData {
@ -48,6 +95,8 @@ pub struct CachedTlsData {
pub cert_payload: Option<TlsCertPayload>,
pub app_data_records_sizes: Vec<usize>,
pub total_app_data_len: usize,
#[serde(default)]
pub behavior_profile: TlsBehaviorProfile,
#[serde(default = "now_system_time", skip_serializing, skip_deserializing)]
pub fetched_at: SystemTime,
pub domain: String,
@ -63,6 +112,40 @@ pub struct TlsFetchResult {
pub server_hello_parsed: ParsedServerHello,
pub app_data_records_sizes: Vec<usize>,
pub total_app_data_len: usize,
#[serde(default)]
pub behavior_profile: TlsBehaviorProfile,
pub cert_info: Option<ParsedCertificateInfo>,
pub cert_payload: Option<TlsCertPayload>,
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn cached_tls_data_deserializes_without_behavior_profile() {
let json = r#"
{
"server_hello_template": {
"version": [3, 3],
"random": [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
"session_id": [],
"cipher_suite": [19, 1],
"compression": 0,
"extensions": []
},
"cert_info": null,
"cert_payload": null,
"app_data_records_sizes": [1024],
"total_app_data_len": 1024,
"domain": "example.com"
}
"#;
let cached: CachedTlsData = serde_json::from_str(json).unwrap();
assert_eq!(cached.behavior_profile.change_cipher_spec_count, 1);
assert!(cached.behavior_profile.app_data_record_sizes.is_empty());
assert!(cached.behavior_profile.ticket_record_sizes.is_empty());
assert_eq!(cached.behavior_profile.source, TlsProfileSource::Default);
}
}