feat: add block_quic config option (#213)

w0l4i has been asking for client-side QUIC block since #213. Now
implemented as a small config flag.

When `block_quic = true`, the SOCKS5 UDP relay drops any datagram
destined for port 443 — that's HTTP/3-over-UDP. The client's QUIC
stack retries a couple of times and then falls back to TCP/HTTPS
through the regular CONNECT path (which goes through the relay
normally).

Why client-side rather than server-side udpgw block: the udpgw
block in #222 is bound to Full mode + Android tun2proxy. This
covers everyone — apps_script users, desktop, Full mode, all the
same path. Skipping at the SOCKS5 layer rather than the tunnel-node
layer also avoids paying 200–500 ms tunnel-node round-trip per
QUIC datagram drop, which compounds during browser retries.

Silent drop is the contractually correct shape: SOCKS5 UDP wire
has no `host unreachable` reply (RFC 1928 §6 only defines that for
TCP CONNECT). Browsers' QUIC stacks have a "no response → fall
back" timeout, so silent drop matches what the protocol expects.

Default false (opt-in) — udpgw mitigates QUIC partly via persistent
sockets, and a tiny minority of sites only support HTTP/3.

Will ship in v1.7.5.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

src/config.rs

@@ -163,6 +163,33 @@ pub struct Config {
     /// Issues #39, #127.
     #[serde(default)]
     pub passthrough_hosts: Vec<String>,
+
+    /// Block outbound QUIC (UDP/443) at the SOCKS5 listener.
+    ///
+    /// QUIC is HTTP/3-over-UDP. In `apps_script` mode it's hopeless —
+    /// Apps Script is HTTP-only, so QUIC datagrams either get refused
+    /// outright (UDP ASSOCIATE rejected) or silently fall through to
+    /// `raw-tcp direct` and fail in interesting ways. In `full` mode
+    /// the tunnel-node CAN carry UDP, but QUIC's congestion control
+    /// stacked on top of TCP-encapsulated transport produces TCP
+    /// meltdown for any non-trivial bandwidth — browsers see <1 Mbps
+    /// where the same site over plain HTTPS would do >50.
+    ///
+    /// With `block_quic = true`, the SOCKS5 UDP relay drops any
+    /// datagram destined for port 443 (silent UDP — caller's stack
+    /// retries a few times then falls back). Browsers then re-issue
+    /// the same request as TCP/HTTPS through the regular CONNECT
+    /// path, which goes through the relay normally.
+    ///
+    /// Why this is opt-in rather than always-on: for users on Full
+    /// mode + udpgw (a recent path; v1.7.0+) the QUIC TCP-meltdown
+    /// is partially mitigated by udpgw's persistent-socket reuse,
+    /// and a tiny minority of sites only support HTTP/3 (rare). The
+    /// flag lets users who care about consistency over peak speed
+    /// opt out of QUIC at the source rather than discovering its
+    /// failure modes later. Issue #213.
+    #[serde(default)]
+    pub block_quic: bool,
 }
 
 fn default_fetch_ips_from_api() -> bool { false }

src/proxy_server.rs

@@ -195,6 +195,10 @@ pub struct RewriteCtx {
     /// and pass through as plain TCP (optionally via upstream_socks5).
     /// See config.rs `passthrough_hosts` for matching rules. Issues #39, #127.
     pub passthrough_hosts: Vec<String>,
+    /// If true, drop SOCKS5 UDP datagrams destined for port 443 so
+    /// callers fall back to TCP/HTTPS. See config.rs `block_quic` for
+    /// the trade-off. Issue #213.
+    pub block_quic: bool,
 }
 
 /// True if `host` matches any entry in the user's passthrough list.
@@ -263,6 +267,7 @@ impl ProxyServer {
             mode,
             youtube_via_relay: config.youtube_via_relay,
             passthrough_hosts: config.passthrough_hosts.clone(),
+            block_quic: config.block_quic,
         });
 
         let socks5_port = config.socks5_port.unwrap_or(config.listen_port + 1);
@@ -864,6 +869,30 @@ async fn handle_socks5_udp_associate(
                     continue;
                 };
 
+                // Issue #213: client-side QUIC block. UDP/443 is
+                // HTTP/3 — drop the datagram silently so the client
+                // stack retries a couple of times and then falls back
+                // to TCP/HTTPS, which goes through the regular CONNECT
+                // path. Skipping this at the SOCKS5 layer (rather than
+                // letting it hit the tunnel-node) avoids paying the
+                // 200–500 ms tunnel-node round-trip per dropped QUIC
+                // datagram, which would otherwise compound during the
+                // 1–3 retries before the browser falls back.
+                //
+                // Silent drop instead of an explicit error reply: the
+                // SOCKS5 UDP wire has no "destination unreachable"
+                // datagram — `0x04` only exists in TCP CONNECT replies
+                // (RFC 1928 §6). The browser's QUIC stack already has
+                // a "no response → fall back" timeout, so silent drop
+                // is the contractually correct shape.
+                if rewrite_ctx.block_quic && target.port == 443 {
+                    tracing::debug!(
+                        "udp dropped: block_quic=true, target {}:443",
+                        target.host
+                    );
+                    continue;
+                }
+
                 // RFC 1928 §6: lock to the first VALID datagram's source
                 // port. Subsequent datagrams must come from the same
                 // (ip, port) pair.