Middle Proxy is so real

src/transport/middle_proxy/codec.rs

@@ -0,0 +1,178 @@
+use tokio::io::{AsyncReadExt, AsyncWriteExt};
+
+use crate::crypto::{AesCbc, crc32};
+use crate::error::{ProxyError, Result};
+use crate::protocol::constants::*;
+
+pub(crate) fn build_rpc_frame(seq_no: i32, payload: &[u8]) -> Vec<u8> {
+    let total_len = (4 + 4 + payload.len() + 4) as u32;
+    let mut frame = Vec::with_capacity(total_len as usize);
+    frame.extend_from_slice(&total_len.to_le_bytes());
+    frame.extend_from_slice(&seq_no.to_le_bytes());
+    frame.extend_from_slice(payload);
+    let c = crc32(&frame);
+    frame.extend_from_slice(&c.to_le_bytes());
+    frame
+}
+
+pub(crate) async fn read_rpc_frame_plaintext(
+    rd: &mut (impl AsyncReadExt + Unpin),
+) -> Result<(i32, Vec<u8>)> {
+    let mut len_buf = [0u8; 4];
+    rd.read_exact(&mut len_buf).await.map_err(ProxyError::Io)?;
+    let total_len = u32::from_le_bytes(len_buf) as usize;
+
+    if !(12..=(1 << 24)).contains(&total_len) {
+        return Err(ProxyError::InvalidHandshake(format!(
+            "Bad RPC frame length: {total_len}"
+        )));
+    }
+
+    let mut rest = vec![0u8; total_len - 4];
+    rd.read_exact(&mut rest).await.map_err(ProxyError::Io)?;
+
+    let mut full = Vec::with_capacity(total_len);
+    full.extend_from_slice(&len_buf);
+    full.extend_from_slice(&rest);
+
+    let crc_offset = total_len - 4;
+    let expected_crc = u32::from_le_bytes(full[crc_offset..crc_offset + 4].try_into().unwrap());
+    let actual_crc = crc32(&full[..crc_offset]);
+    if expected_crc != actual_crc {
+        return Err(ProxyError::InvalidHandshake(format!(
+            "CRC mismatch: 0x{expected_crc:08x} vs 0x{actual_crc:08x}"
+        )));
+    }
+
+    let seq_no = i32::from_le_bytes(full[4..8].try_into().unwrap());
+    let payload = full[8..crc_offset].to_vec();
+    Ok((seq_no, payload))
+}
+
+pub(crate) fn build_nonce_payload(key_selector: u32, crypto_ts: u32, nonce: &[u8; 16]) -> [u8; 32] {
+    let mut p = [0u8; 32];
+    p[0..4].copy_from_slice(&RPC_NONCE_U32.to_le_bytes());
+    p[4..8].copy_from_slice(&key_selector.to_le_bytes());
+    p[8..12].copy_from_slice(&RPC_CRYPTO_AES_U32.to_le_bytes());
+    p[12..16].copy_from_slice(&crypto_ts.to_le_bytes());
+    p[16..32].copy_from_slice(nonce);
+    p
+}
+
+pub(crate) fn parse_nonce_payload(d: &[u8]) -> Result<(u32, u32, [u8; 16])> {
+    if d.len() < 32 {
+        return Err(ProxyError::InvalidHandshake(format!(
+            "Nonce payload too short: {} bytes",
+            d.len()
+        )));
+    }
+
+    let t = u32::from_le_bytes(d[0..4].try_into().unwrap());
+    if t != RPC_NONCE_U32 {
+        return Err(ProxyError::InvalidHandshake(format!(
+            "Expected RPC_NONCE 0x{RPC_NONCE_U32:08x}, got 0x{t:08x}"
+        )));
+    }
+
+    let schema = u32::from_le_bytes(d[8..12].try_into().unwrap());
+    let ts = u32::from_le_bytes(d[12..16].try_into().unwrap());
+    let mut nonce = [0u8; 16];
+    nonce.copy_from_slice(&d[16..32]);
+    Ok((schema, ts, nonce))
+}
+
+pub(crate) fn build_handshake_payload(
+    our_ip: [u8; 4],
+    our_port: u16,
+    peer_ip: [u8; 4],
+    peer_port: u16,
+) -> [u8; 32] {
+    let mut p = [0u8; 32];
+    p[0..4].copy_from_slice(&RPC_HANDSHAKE_U32.to_le_bytes());
+
+    // Keep C memory layout compatibility for PID IPv4 bytes.
+    p[8..12].copy_from_slice(&our_ip);
+    p[12..14].copy_from_slice(&our_port.to_le_bytes());
+    let pid = (std::process::id() & 0xffff) as u16;
+    p[14..16].copy_from_slice(&pid.to_le_bytes());
+    let utime = std::time::SystemTime::now()
+        .duration_since(std::time::UNIX_EPOCH)
+        .unwrap_or_default()
+        .as_secs() as u32;
+    p[16..20].copy_from_slice(&utime.to_le_bytes());
+
+    p[20..24].copy_from_slice(&peer_ip);
+    p[24..26].copy_from_slice(&peer_port.to_le_bytes());
+    p
+}
+
+pub(crate) fn cbc_encrypt_padded(
+    key: &[u8; 32],
+    iv: &[u8; 16],
+    plaintext: &[u8],
+) -> Result<(Vec<u8>, [u8; 16])> {
+    let pad = (16 - (plaintext.len() % 16)) % 16;
+    let mut buf = plaintext.to_vec();
+    let pad_pattern: [u8; 4] = [0x04, 0x00, 0x00, 0x00];
+    for i in 0..pad {
+        buf.push(pad_pattern[i % 4]);
+    }
+
+    let cipher = AesCbc::new(*key, *iv);
+    cipher
+        .encrypt_in_place(&mut buf)
+        .map_err(|e| ProxyError::Crypto(format!("CBC encrypt: {e}")))?;
+
+    let mut new_iv = [0u8; 16];
+    if buf.len() >= 16 {
+        new_iv.copy_from_slice(&buf[buf.len() - 16..]);
+    }
+    Ok((buf, new_iv))
+}
+
+pub(crate) fn cbc_decrypt_inplace(
+    key: &[u8; 32],
+    iv: &[u8; 16],
+    data: &mut [u8],
+) -> Result<[u8; 16]> {
+    let mut new_iv = [0u8; 16];
+    if data.len() >= 16 {
+        new_iv.copy_from_slice(&data[data.len() - 16..]);
+    }
+
+    AesCbc::new(*key, *iv)
+        .decrypt_in_place(data)
+        .map_err(|e| ProxyError::Crypto(format!("CBC decrypt: {e}")))?;
+    Ok(new_iv)
+}
+
+pub(crate) struct RpcWriter {
+    pub(crate) writer: tokio::io::WriteHalf<tokio::net::TcpStream>,
+    pub(crate) key: [u8; 32],
+    pub(crate) iv: [u8; 16],
+    pub(crate) seq_no: i32,
+}
+
+impl RpcWriter {
+    pub(crate) async fn send(&mut self, payload: &[u8]) -> Result<()> {
+        let frame = build_rpc_frame(self.seq_no, payload);
+        self.seq_no += 1;
+
+        let pad = (16 - (frame.len() % 16)) % 16;
+        let mut buf = frame;
+        let pad_pattern: [u8; 4] = [0x04, 0x00, 0x00, 0x00];
+        for i in 0..pad {
+            buf.push(pad_pattern[i % 4]);
+        }
+
+        let cipher = AesCbc::new(self.key, self.iv);
+        cipher
+            .encrypt_in_place(&mut buf)
+            .map_err(|e| ProxyError::Crypto(format!("{e}")))?;
+
+        if buf.len() >= 16 {
+            self.iv.copy_from_slice(&buf[buf.len() - 16..]);
+        }
+        self.writer.write_all(&buf).await.map_err(ProxyError::Io)
+    }
+}

src/transport/middle_proxy/health.rs

@@ -0,0 +1,38 @@
+use std::net::SocketAddr;
+use std::sync::Arc;
+use std::time::Duration;
+
+use tracing::{debug, info, warn};
+
+use crate::crypto::SecureRandom;
+use crate::protocol::constants::TG_MIDDLE_PROXIES_FLAT_V4;
+
+use super::MePool;
+
+pub async fn me_health_monitor(pool: Arc<MePool>, rng: Arc<SecureRandom>, min_connections: usize) {
+    loop {
+        tokio::time::sleep(Duration::from_secs(30)).await;
+        let current = pool.connection_count();
+        if current < min_connections {
+            warn!(
+                current,
+                min = min_connections,
+                "ME pool below minimum, reconnecting..."
+            );
+            let addrs = TG_MIDDLE_PROXIES_FLAT_V4.clone();
+            for &(ip, port) in addrs.iter() {
+                let needed = min_connections.saturating_sub(pool.connection_count());
+                if needed == 0 {
+                    break;
+                }
+                for _ in 0..needed {
+                    let addr = SocketAddr::new(ip, port);
+                    match pool.connect_one(addr, &rng).await {
+                        Ok(()) => info!(%addr, "ME reconnected"),
+                        Err(e) => debug!(%addr, error = %e, "ME reconnect failed"),
+                    }
+                }
+            }
+        }
+    }
+}

src/transport/middle_proxy/mod.rs

@@ -0,0 +1,24 @@
+//! Middle Proxy RPC transport.
+
+mod codec;
+mod health;
+mod pool;
+mod reader;
+mod registry;
+mod secret;
+mod wire;
+
+use bytes::Bytes;
+
+pub use health::me_health_monitor;
+pub use pool::MePool;
+pub use registry::ConnRegistry;
+pub use secret::fetch_proxy_secret;
+pub use wire::proto_flags_for_tag;
+
+#[derive(Debug)]
+pub enum MeResponse {
+    Data { flags: u32, data: Bytes },
+    Ack(u32),
+    Close,
+}

src/transport/middle_proxy/pool.rs

@@ -0,0 +1,431 @@
+use std::net::{IpAddr, SocketAddr};
+use std::sync::Arc;
+use std::sync::atomic::{AtomicU64, Ordering};
+use std::time::Duration;
+
+use bytes::BytesMut;
+use tokio::io::{AsyncReadExt, AsyncWriteExt};
+use tokio::net::TcpStream;
+use tokio::sync::{Mutex, RwLock};
+use tokio::time::{Instant, timeout};
+use tracing::{debug, info, warn};
+
+use crate::crypto::{SecureRandom, derive_middleproxy_keys};
+use crate::error::{ProxyError, Result};
+use crate::protocol::constants::*;
+
+use super::ConnRegistry;
+use super::codec::{
+    RpcWriter, build_handshake_payload, build_nonce_payload, build_rpc_frame, cbc_decrypt_inplace,
+    cbc_encrypt_padded, parse_nonce_payload, read_rpc_frame_plaintext,
+};
+use super::reader::reader_loop;
+use super::wire::{IpMaterial, build_proxy_req_payload, extract_ip_material};
+
+pub struct MePool {
+    registry: Arc<ConnRegistry>,
+    writers: Arc<RwLock<Vec<Arc<Mutex<RpcWriter>>>>>,
+    rr: AtomicU64,
+    proxy_tag: Option<Vec<u8>>,
+    proxy_secret: Vec<u8>,
+    nat_ip: Option<IpAddr>,
+    pool_size: usize,
+}
+
+impl MePool {
+    pub fn new(
+        proxy_tag: Option<Vec<u8>>,
+        proxy_secret: Vec<u8>,
+        nat_ip: Option<IpAddr>,
+    ) -> Arc<Self> {
+        Arc::new(Self {
+            registry: Arc::new(ConnRegistry::new()),
+            writers: Arc::new(RwLock::new(Vec::new())),
+            rr: AtomicU64::new(0),
+            proxy_tag,
+            proxy_secret,
+            nat_ip,
+            pool_size: 2,
+        })
+    }
+
+    pub fn has_proxy_tag(&self) -> bool {
+        self.proxy_tag.is_some()
+    }
+
+    pub fn translate_our_addr(&self, addr: SocketAddr) -> SocketAddr {
+        let ip = self.translate_ip_for_nat(addr.ip());
+        SocketAddr::new(ip, addr.port())
+    }
+
+    pub fn registry(&self) -> &Arc<ConnRegistry> {
+        &self.registry
+    }
+
+    fn translate_ip_for_nat(&self, ip: IpAddr) -> IpAddr {
+        let Some(nat_ip) = self.nat_ip else {
+            return ip;
+        };
+
+        match (ip, nat_ip) {
+            (IpAddr::V4(src), IpAddr::V4(dst))
+                if src.is_private() || src.is_loopback() || src.is_unspecified() =>
+            {
+                IpAddr::V4(dst)
+            }
+            (IpAddr::V6(src), IpAddr::V6(dst)) if src.is_loopback() || src.is_unspecified() => {
+                IpAddr::V6(dst)
+            }
+            (orig, _) => orig,
+        }
+    }
+
+    fn writers_arc(&self) -> Arc<RwLock<Vec<Arc<Mutex<RpcWriter>>>>> {
+        self.writers.clone()
+    }
+
+    fn key_selector(&self) -> u32 {
+        if self.proxy_secret.len() >= 4 {
+            u32::from_le_bytes([
+                self.proxy_secret[0],
+                self.proxy_secret[1],
+                self.proxy_secret[2],
+                self.proxy_secret[3],
+            ])
+        } else {
+            0
+        }
+    }
+
+    pub async fn init(self: &Arc<Self>, pool_size: usize, rng: &SecureRandom) -> Result<()> {
+        let addrs = &*TG_MIDDLE_PROXIES_FLAT_V4;
+        let ks = self.key_selector();
+        info!(
+            me_servers = addrs.len(),
+            pool_size,
+            key_selector = format_args!("0x{ks:08x}"),
+            secret_len = self.proxy_secret.len(),
+            "Initializing ME pool"
+        );
+
+        for &(ip, port) in addrs.iter() {
+            for i in 0..pool_size {
+                let addr = SocketAddr::new(ip, port);
+                match self.connect_one(addr, rng).await {
+                    Ok(()) => info!(%addr, idx = i, "ME connected"),
+                    Err(e) => warn!(%addr, idx = i, error = %e, "ME connect failed"),
+                }
+            }
+            if self.writers.read().await.len() >= pool_size {
+                break;
+            }
+        }
+
+        if self.writers.read().await.is_empty() {
+            return Err(ProxyError::Proxy("No ME connections".into()));
+        }
+        Ok(())
+    }
+
+    pub(crate) async fn connect_one(
+        self: &Arc<Self>,
+        addr: SocketAddr,
+        rng: &SecureRandom,
+    ) -> Result<()> {
+        let secret = &self.proxy_secret;
+        if secret.len() < 32 {
+            return Err(ProxyError::Proxy(
+                "proxy-secret too short for ME auth".into(),
+            ));
+        }
+
+        let stream = timeout(
+            Duration::from_secs(ME_CONNECT_TIMEOUT_SECS),
+            TcpStream::connect(addr),
+        )
+        .await
+        .map_err(|_| ProxyError::ConnectionTimeout {
+            addr: addr.to_string(),
+        })?
+        .map_err(ProxyError::Io)?;
+        stream.set_nodelay(true).ok();
+
+        let local_addr = stream.local_addr().map_err(ProxyError::Io)?;
+        let peer_addr = stream.peer_addr().map_err(ProxyError::Io)?;
+        let local_addr_nat = self.translate_our_addr(local_addr);
+        let peer_addr_nat =
+            SocketAddr::new(self.translate_ip_for_nat(peer_addr.ip()), peer_addr.port());
+        let (mut rd, mut wr) = tokio::io::split(stream);
+
+        let my_nonce: [u8; 16] = rng.bytes(16).try_into().unwrap();
+        let crypto_ts = std::time::SystemTime::now()
+            .duration_since(std::time::UNIX_EPOCH)
+            .unwrap_or_default()
+            .as_secs() as u32;
+
+        let ks = self.key_selector();
+        let nonce_payload = build_nonce_payload(ks, crypto_ts, &my_nonce);
+        let nonce_frame = build_rpc_frame(-2, &nonce_payload);
+        wr.write_all(&nonce_frame).await.map_err(ProxyError::Io)?;
+        wr.flush().await.map_err(ProxyError::Io)?;
+
+        let (srv_seq, srv_nonce_payload) = timeout(
+            Duration::from_secs(ME_HANDSHAKE_TIMEOUT_SECS),
+            read_rpc_frame_plaintext(&mut rd),
+        )
+        .await
+        .map_err(|_| ProxyError::TgHandshakeTimeout)??;
+
+        if srv_seq != -2 {
+            return Err(ProxyError::InvalidHandshake(format!(
+                "Expected seq=-2, got {srv_seq}"
+            )));
+        }
+
+        let (schema, srv_ts, srv_nonce) = parse_nonce_payload(&srv_nonce_payload)?;
+        if schema != RPC_CRYPTO_AES_U32 {
+            return Err(ProxyError::InvalidHandshake(format!(
+                "Unsupported crypto schema: 0x{schema:x}"
+            )));
+        }
+
+        let skew = crypto_ts.abs_diff(srv_ts);
+        if skew > 30 {
+            return Err(ProxyError::InvalidHandshake(format!(
+                "nonce crypto_ts skew too large: client={crypto_ts}, server={srv_ts}, skew={skew}s"
+            )));
+        }
+
+        let ts_bytes = crypto_ts.to_le_bytes();
+        let server_port_bytes = peer_addr_nat.port().to_le_bytes();
+        let client_port_bytes = local_addr_nat.port().to_le_bytes();
+
+        let server_ip = extract_ip_material(peer_addr_nat);
+        let client_ip = extract_ip_material(local_addr_nat);
+
+        let (srv_ip_opt, clt_ip_opt, clt_v6_opt, srv_v6_opt, hs_our_ip, hs_peer_ip) =
+            match (server_ip, client_ip) {
+                (IpMaterial::V4(srv), IpMaterial::V4(clt)) => {
+                    (Some(srv), Some(clt), None, None, clt, srv)
+                }
+                (IpMaterial::V6(srv), IpMaterial::V6(clt)) => {
+                    let zero = [0u8; 4];
+                    (None, None, Some(clt), Some(srv), zero, zero)
+                }
+                _ => {
+                    return Err(ProxyError::InvalidHandshake(
+                        "mixed IPv4/IPv6 endpoints are not supported for ME key derivation"
+                            .to_string(),
+                    ));
+                }
+            };
+
+        let (wk, wi) = derive_middleproxy_keys(
+            &srv_nonce,
+            &my_nonce,
+            &ts_bytes,
+            srv_ip_opt.as_ref().map(|x| &x[..]),
+            &client_port_bytes,
+            b"CLIENT",
+            clt_ip_opt.as_ref().map(|x| &x[..]),
+            &server_port_bytes,
+            secret,
+            clt_v6_opt.as_ref(),
+            srv_v6_opt.as_ref(),
+        );
+        let (rk, ri) = derive_middleproxy_keys(
+            &srv_nonce,
+            &my_nonce,
+            &ts_bytes,
+            srv_ip_opt.as_ref().map(|x| &x[..]),
+            &client_port_bytes,
+            b"SERVER",
+            clt_ip_opt.as_ref().map(|x| &x[..]),
+            &server_port_bytes,
+            secret,
+            clt_v6_opt.as_ref(),
+            srv_v6_opt.as_ref(),
+        );
+
+        let hs_payload =
+            build_handshake_payload(hs_our_ip, local_addr.port(), hs_peer_ip, peer_addr.port());
+        let hs_frame = build_rpc_frame(-1, &hs_payload);
+        let (encrypted_hs, write_iv) = cbc_encrypt_padded(&wk, &wi, &hs_frame)?;
+        wr.write_all(&encrypted_hs).await.map_err(ProxyError::Io)?;
+        wr.flush().await.map_err(ProxyError::Io)?;
+
+        let deadline = Instant::now() + Duration::from_secs(ME_HANDSHAKE_TIMEOUT_SECS);
+        let mut enc_buf = BytesMut::with_capacity(256);
+        let mut dec_buf = BytesMut::with_capacity(256);
+        let mut read_iv = ri;
+        let mut handshake_ok = false;
+
+        while Instant::now() < deadline && !handshake_ok {
+            let remaining = deadline - Instant::now();
+            let mut tmp = [0u8; 256];
+            let n = match timeout(remaining, rd.read(&mut tmp)).await {
+                Ok(Ok(0)) => {
+                    return Err(ProxyError::Io(std::io::Error::new(
+                        std::io::ErrorKind::UnexpectedEof,
+                        "ME closed during handshake",
+                    )));
+                }
+                Ok(Ok(n)) => n,
+                Ok(Err(e)) => return Err(ProxyError::Io(e)),
+                Err(_) => return Err(ProxyError::TgHandshakeTimeout),
+            };
+
+            enc_buf.extend_from_slice(&tmp[..n]);
+
+            let blocks = enc_buf.len() / 16 * 16;
+            if blocks > 0 {
+                let mut chunk = vec![0u8; blocks];
+                chunk.copy_from_slice(&enc_buf[..blocks]);
+                read_iv = cbc_decrypt_inplace(&rk, &read_iv, &mut chunk)?;
+                dec_buf.extend_from_slice(&chunk);
+                let _ = enc_buf.split_to(blocks);
+            }
+
+            while dec_buf.len() >= 4 {
+                let fl = u32::from_le_bytes(dec_buf[0..4].try_into().unwrap()) as usize;
+
+                if fl == 4 {
+                    let _ = dec_buf.split_to(4);
+                    continue;
+                }
+                if !(12..=(1 << 24)).contains(&fl) {
+                    return Err(ProxyError::InvalidHandshake(format!(
+                        "Bad HS response frame len: {fl}"
+                    )));
+                }
+                if dec_buf.len() < fl {
+                    break;
+                }
+
+                let frame = dec_buf.split_to(fl);
+                let pe = fl - 4;
+                let ec = u32::from_le_bytes(frame[pe..pe + 4].try_into().unwrap());
+                let ac = crate::crypto::crc32(&frame[..pe]);
+                if ec != ac {
+                    return Err(ProxyError::InvalidHandshake(format!(
+                        "HS CRC mismatch: 0x{ec:08x} vs 0x{ac:08x}"
+                    )));
+                }
+
+                let hs_type = u32::from_le_bytes(frame[8..12].try_into().unwrap());
+                if hs_type == RPC_HANDSHAKE_ERROR_U32 {
+                    let err_code = if frame.len() >= 16 {
+                        i32::from_le_bytes(frame[12..16].try_into().unwrap())
+                    } else {
+                        -1
+                    };
+                    return Err(ProxyError::InvalidHandshake(format!(
+                        "ME rejected handshake (error={err_code})"
+                    )));
+                }
+                if hs_type != RPC_HANDSHAKE_U32 {
+                    return Err(ProxyError::InvalidHandshake(format!(
+                        "Expected HANDSHAKE 0x{RPC_HANDSHAKE_U32:08x}, got 0x{hs_type:08x}"
+                    )));
+                }
+
+                handshake_ok = true;
+                break;
+            }
+        }
+
+        if !handshake_ok {
+            return Err(ProxyError::TgHandshakeTimeout);
+        }
+
+        info!(%addr, "RPC handshake OK");
+
+        let rpc_w = Arc::new(Mutex::new(RpcWriter {
+            writer: wr,
+            key: wk,
+            iv: write_iv,
+            seq_no: 0,
+        }));
+        self.writers.write().await.push(rpc_w.clone());
+
+        let reg = self.registry.clone();
+        let w_pong = rpc_w.clone();
+        let w_pool = self.writers_arc();
+        tokio::spawn(async move {
+            if let Err(e) =
+                reader_loop(rd, rk, read_iv, reg, enc_buf, dec_buf, w_pong.clone()).await
+            {
+                warn!(error = %e, "ME reader ended");
+            }
+            let mut ws = w_pool.write().await;
+            ws.retain(|w| !Arc::ptr_eq(w, &w_pong));
+            info!(remaining = ws.len(), "Dead ME writer removed from pool");
+        });
+
+        Ok(())
+    }
+
+    pub async fn send_proxy_req(
+        &self,
+        conn_id: u64,
+        client_addr: SocketAddr,
+        our_addr: SocketAddr,
+        data: &[u8],
+        proto_flags: u32,
+    ) -> Result<()> {
+        let payload = build_proxy_req_payload(
+            conn_id,
+            client_addr,
+            our_addr,
+            data,
+            self.proxy_tag.as_deref(),
+            proto_flags,
+        );
+
+        loop {
+            let ws = self.writers.read().await;
+            if ws.is_empty() {
+                return Err(ProxyError::Proxy("All ME connections dead".into()));
+            }
+
+            let idx = self.rr.fetch_add(1, Ordering::Relaxed) as usize % ws.len();
+            let w = ws[idx].clone();
+            drop(ws);
+
+            match w.lock().await.send(&payload).await {
+                Ok(()) => return Ok(()),
+                Err(e) => {
+                    warn!(error = %e, "ME write failed, removing dead conn");
+                    let mut ws = self.writers.write().await;
+                    ws.retain(|o| !Arc::ptr_eq(o, &w));
+                    if ws.is_empty() {
+                        return Err(ProxyError::Proxy("All ME connections dead".into()));
+                    }
+                }
+            }
+        }
+    }
+
+    pub async fn send_close(&self, conn_id: u64) -> Result<()> {
+        let ws = self.writers.read().await;
+        if !ws.is_empty() {
+            let w = ws[0].clone();
+            drop(ws);
+            let mut p = Vec::with_capacity(12);
+            p.extend_from_slice(&RPC_CLOSE_EXT_U32.to_le_bytes());
+            p.extend_from_slice(&conn_id.to_le_bytes());
+            if let Err(e) = w.lock().await.send(&p).await {
+                debug!(error = %e, "ME close write failed");
+                let mut ws = self.writers.write().await;
+                ws.retain(|o| !Arc::ptr_eq(o, &w));
+            }
+        }
+
+        self.registry.unregister(conn_id).await;
+        Ok(())
+    }
+
+    pub fn connection_count(&self) -> usize {
+        self.writers.try_read().map(|w| w.len()).unwrap_or(0)
+    }
+}

src/transport/middle_proxy/reader.rs

@@ -0,0 +1,141 @@
+use std::sync::Arc;
+
+use bytes::{Bytes, BytesMut};
+use tokio::io::AsyncReadExt;
+use tokio::net::TcpStream;
+use tokio::sync::Mutex;
+use tracing::{debug, trace, warn};
+
+use crate::crypto::{AesCbc, crc32};
+use crate::error::{ProxyError, Result};
+use crate::protocol::constants::*;
+
+use super::codec::RpcWriter;
+use super::{ConnRegistry, MeResponse};
+
+pub(crate) async fn reader_loop(
+    mut rd: tokio::io::ReadHalf<TcpStream>,
+    dk: [u8; 32],
+    mut div: [u8; 16],
+    reg: Arc<ConnRegistry>,
+    enc_leftover: BytesMut,
+    mut dec: BytesMut,
+    writer: Arc<Mutex<RpcWriter>>,
+) -> Result<()> {
+    let mut raw = enc_leftover;
+
+    loop {
+        let mut tmp = [0u8; 16_384];
+        let n = rd.read(&mut tmp).await.map_err(ProxyError::Io)?;
+        if n == 0 {
+            return Ok(());
+        }
+        raw.extend_from_slice(&tmp[..n]);
+
+        let blocks = raw.len() / 16 * 16;
+        if blocks > 0 {
+            let mut new_iv = [0u8; 16];
+            new_iv.copy_from_slice(&raw[blocks - 16..blocks]);
+
+            let mut chunk = vec![0u8; blocks];
+            chunk.copy_from_slice(&raw[..blocks]);
+            AesCbc::new(dk, div)
+                .decrypt_in_place(&mut chunk)
+                .map_err(|e| ProxyError::Crypto(format!("{e}")))?;
+            div = new_iv;
+            dec.extend_from_slice(&chunk);
+            let _ = raw.split_to(blocks);
+        }
+
+        while dec.len() >= 12 {
+            let fl = u32::from_le_bytes(dec[0..4].try_into().unwrap()) as usize;
+            if fl == 4 {
+                let _ = dec.split_to(4);
+                continue;
+            }
+            if !(12..=(1 << 24)).contains(&fl) {
+                warn!(frame_len = fl, "Invalid RPC frame len");
+                dec.clear();
+                break;
+            }
+            if dec.len() < fl {
+                break;
+            }
+
+            let frame = dec.split_to(fl);
+            let pe = fl - 4;
+            let ec = u32::from_le_bytes(frame[pe..pe + 4].try_into().unwrap());
+            if crc32(&frame[..pe]) != ec {
+                warn!("CRC mismatch in data frame");
+                continue;
+            }
+
+            let payload = &frame[8..pe];
+            if payload.len() < 4 {
+                continue;
+            }
+
+            let pt = u32::from_le_bytes(payload[0..4].try_into().unwrap());
+            let body = &payload[4..];
+
+            if pt == RPC_PROXY_ANS_U32 && body.len() >= 12 {
+                let flags = u32::from_le_bytes(body[0..4].try_into().unwrap());
+                let cid = u64::from_le_bytes(body[4..12].try_into().unwrap());
+                let data = Bytes::copy_from_slice(&body[12..]);
+                trace!(cid, flags, len = data.len(), "RPC_PROXY_ANS");
+
+                let routed = reg.route(cid, MeResponse::Data { flags, data }).await;
+                if !routed {
+                    reg.unregister(cid).await;
+                    send_close_conn(&writer, cid).await;
+                }
+            } else if pt == RPC_SIMPLE_ACK_U32 && body.len() >= 12 {
+                let cid = u64::from_le_bytes(body[0..8].try_into().unwrap());
+                let cfm = u32::from_le_bytes(body[8..12].try_into().unwrap());
+                trace!(cid, cfm, "RPC_SIMPLE_ACK");
+
+                let routed = reg.route(cid, MeResponse::Ack(cfm)).await;
+                if !routed {
+                    reg.unregister(cid).await;
+                    send_close_conn(&writer, cid).await;
+                }
+            } else if pt == RPC_CLOSE_EXT_U32 && body.len() >= 8 {
+                let cid = u64::from_le_bytes(body[0..8].try_into().unwrap());
+                debug!(cid, "RPC_CLOSE_EXT from ME");
+                reg.route(cid, MeResponse::Close).await;
+                reg.unregister(cid).await;
+            } else if pt == RPC_CLOSE_CONN_U32 && body.len() >= 8 {
+                let cid = u64::from_le_bytes(body[0..8].try_into().unwrap());
+                debug!(cid, "RPC_CLOSE_CONN from ME");
+                reg.route(cid, MeResponse::Close).await;
+                reg.unregister(cid).await;
+            } else if pt == RPC_PING_U32 && body.len() >= 8 {
+                let ping_id = i64::from_le_bytes(body[0..8].try_into().unwrap());
+                trace!(ping_id, "RPC_PING -> RPC_PONG");
+                let mut pong = Vec::with_capacity(12);
+                pong.extend_from_slice(&RPC_PONG_U32.to_le_bytes());
+                pong.extend_from_slice(&ping_id.to_le_bytes());
+                if let Err(e) = writer.lock().await.send(&pong).await {
+                    warn!(error = %e, "PONG send failed");
+                    break;
+                }
+            } else {
+                debug!(
+                    rpc_type = format_args!("0x{pt:08x}"),
+                    len = body.len(),
+                    "Unknown RPC"
+                );
+            }
+        }
+    }
+}
+
+async fn send_close_conn(writer: &Arc<Mutex<RpcWriter>>, conn_id: u64) {
+    let mut p = Vec::with_capacity(12);
+    p.extend_from_slice(&RPC_CLOSE_CONN_U32.to_le_bytes());
+    p.extend_from_slice(&conn_id.to_le_bytes());
+
+    if let Err(e) = writer.lock().await.send(&p).await {
+        debug!(conn_id, error = %e, "Failed to send RPC_CLOSE_CONN");
+    }
+}

src/transport/middle_proxy/registry.rs

@@ -0,0 +1,40 @@
+use std::collections::HashMap;
+use std::sync::atomic::{AtomicU64, Ordering};
+
+use tokio::sync::{RwLock, mpsc};
+
+use super::MeResponse;
+
+pub struct ConnRegistry {
+    map: RwLock<HashMap<u64, mpsc::Sender<MeResponse>>>,
+    next_id: AtomicU64,
+}
+
+impl ConnRegistry {
+    pub fn new() -> Self {
+        Self {
+            map: RwLock::new(HashMap::new()),
+            next_id: AtomicU64::new(1),
+        }
+    }
+
+    pub async fn register(&self) -> (u64, mpsc::Receiver<MeResponse>) {
+        let id = self.next_id.fetch_add(1, Ordering::Relaxed);
+        let (tx, rx) = mpsc::channel(256);
+        self.map.write().await.insert(id, tx);
+        (id, rx)
+    }
+
+    pub async fn unregister(&self, id: u64) {
+        self.map.write().await.remove(&id);
+    }
+
+    pub async fn route(&self, id: u64, resp: MeResponse) -> bool {
+        let m = self.map.read().await;
+        if let Some(tx) = m.get(&id) {
+            tx.send(resp).await.is_ok()
+        } else {
+            false
+        }
+    }
+}

src/transport/middle_proxy/secret.rs

@@ -0,0 +1,76 @@
+use std::time::Duration;
+
+use tracing::{debug, info, warn};
+
+use crate::error::{ProxyError, Result};
+
+/// Fetch Telegram proxy-secret binary.
+pub async fn fetch_proxy_secret(cache_path: Option<&str>) -> Result<Vec<u8>> {
+    let cache = cache_path.unwrap_or("proxy-secret");
+
+    if let Ok(metadata) = tokio::fs::metadata(cache).await {
+        if let Ok(modified) = metadata.modified() {
+            let age = std::time::SystemTime::now()
+                .duration_since(modified)
+                .unwrap_or(Duration::from_secs(u64::MAX));
+            if age < Duration::from_secs(86_400) {
+                if let Ok(data) = tokio::fs::read(cache).await {
+                    if data.len() >= 32 {
+                        info!(
+                            path = cache,
+                            len = data.len(),
+                            age_hours = age.as_secs() / 3600,
+                            "Loaded proxy-secret from cache"
+                        );
+                        return Ok(data);
+                    }
+                    warn!(
+                        path = cache,
+                        len = data.len(),
+                        "Cached proxy-secret too short"
+                    );
+                }
+            }
+        }
+    }
+
+    info!("Downloading proxy-secret from core.telegram.org...");
+    let data = download_proxy_secret().await?;
+
+    if let Err(e) = tokio::fs::write(cache, &data).await {
+        warn!(error = %e, "Failed to cache proxy-secret (non-fatal)");
+    } else {
+        debug!(path = cache, len = data.len(), "Cached proxy-secret");
+    }
+
+    Ok(data)
+}
+
+async fn download_proxy_secret() -> Result<Vec<u8>> {
+    let resp = reqwest::get("https://core.telegram.org/getProxySecret")
+        .await
+        .map_err(|e| ProxyError::Proxy(format!("Failed to download proxy-secret: {e}")))?;
+
+    if !resp.status().is_success() {
+        return Err(ProxyError::Proxy(format!(
+            "proxy-secret download HTTP {}",
+            resp.status()
+        )));
+    }
+
+    let data = resp
+        .bytes()
+        .await
+        .map_err(|e| ProxyError::Proxy(format!("Read proxy-secret body: {e}")))?
+        .to_vec();
+
+    if data.len() < 32 {
+        return Err(ProxyError::Proxy(format!(
+            "proxy-secret too short: {} bytes (need >= 32)",
+            data.len()
+        )));
+    }
+
+    info!(len = data.len(), "Downloaded proxy-secret OK");
+    Ok(data)
+}

src/transport/middle_proxy/wire.rs

@@ -0,0 +1,106 @@
+use std::net::{IpAddr, Ipv4Addr, SocketAddr};
+
+use crate::protocol::constants::*;
+
+#[derive(Clone, Copy)]
+pub(crate) enum IpMaterial {
+    V4([u8; 4]),
+    V6([u8; 16]),
+}
+
+pub(crate) fn extract_ip_material(addr: SocketAddr) -> IpMaterial {
+    match addr.ip() {
+        IpAddr::V4(v4) => IpMaterial::V4(v4.octets()),
+        IpAddr::V6(v6) => {
+            if let Some(v4) = v6.to_ipv4_mapped() {
+                IpMaterial::V4(v4.octets())
+            } else {
+                IpMaterial::V6(v6.octets())
+            }
+        }
+    }
+}
+
+fn ipv4_to_mapped_v6_c_compat(ip: Ipv4Addr) -> [u8; 16] {
+    let mut buf = [0u8; 16];
+
+    // Matches tl_store_long(0) + tl_store_int(-0x10000).
+    buf[8..12].copy_from_slice(&(-0x10000i32).to_le_bytes());
+
+    // Matches tl_store_int(htonl(remote_ip_host_order)).
+    let host_order = u32::from_ne_bytes(ip.octets());
+    let network_order = host_order.to_be();
+    buf[12..16].copy_from_slice(&network_order.to_le_bytes());
+
+    buf
+}
+
+fn append_mapped_addr_and_port(buf: &mut Vec<u8>, addr: SocketAddr) {
+    match addr.ip() {
+        IpAddr::V4(v4) => buf.extend_from_slice(&ipv4_to_mapped_v6_c_compat(v4)),
+        IpAddr::V6(v6) => buf.extend_from_slice(&v6.octets()),
+    }
+    buf.extend_from_slice(&(addr.port() as u32).to_le_bytes());
+}
+
+pub(crate) fn build_proxy_req_payload(
+    conn_id: u64,
+    client_addr: SocketAddr,
+    our_addr: SocketAddr,
+    data: &[u8],
+    proxy_tag: Option<&[u8]>,
+    proto_flags: u32,
+) -> Vec<u8> {
+    let mut b = Vec::with_capacity(128 + data.len());
+
+    b.extend_from_slice(&RPC_PROXY_REQ_U32.to_le_bytes());
+    b.extend_from_slice(&proto_flags.to_le_bytes());
+    b.extend_from_slice(&conn_id.to_le_bytes());
+
+    append_mapped_addr_and_port(&mut b, client_addr);
+    append_mapped_addr_and_port(&mut b, our_addr);
+
+    if proto_flags & 12 != 0 {
+        let extra_start = b.len();
+        b.extend_from_slice(&0u32.to_le_bytes());
+
+        if let Some(tag) = proxy_tag {
+            b.extend_from_slice(&TL_PROXY_TAG_U32.to_le_bytes());
+
+            if tag.len() < 254 {
+                b.push(tag.len() as u8);
+                b.extend_from_slice(tag);
+                let pad = (4 - ((1 + tag.len()) % 4)) % 4;
+                b.extend(std::iter::repeat_n(0u8, pad));
+            } else {
+                b.push(0xfe);
+                let len_bytes = (tag.len() as u32).to_le_bytes();
+                b.extend_from_slice(&len_bytes[..3]);
+                b.extend_from_slice(tag);
+                let pad = (4 - (tag.len() % 4)) % 4;
+                b.extend(std::iter::repeat_n(0u8, pad));
+            }
+        }
+
+        let extra_bytes = (b.len() - extra_start - 4) as u32;
+        b[extra_start..extra_start + 4].copy_from_slice(&extra_bytes.to_le_bytes());
+    }
+
+    b.extend_from_slice(data);
+    b
+}
+
+pub fn proto_flags_for_tag(tag: crate::protocol::constants::ProtoTag, has_proxy_tag: bool) -> u32 {
+    use crate::protocol::constants::ProtoTag;
+
+    let mut flags = RPC_FLAG_MAGIC | RPC_FLAG_EXTMODE2;
+    if has_proxy_tag {
+        flags |= RPC_FLAG_HAS_AD_TAG;
+    }
+
+    match tag {
+        ProtoTag::Abridged => flags | RPC_FLAG_ABRIDGED,
+        ProtoTag::Intermediate => flags | RPC_FLAG_INTERMEDIATE,
+        ProtoTag::Secure => flags | RPC_FLAG_PAD | RPC_FLAG_INTERMEDIATE,
+    }
+}