Keep server alive on transient UDP send errors

src/bin/dosh-server.rs

@@ -980,7 +980,7 @@ async fn handle_native_client_hello(
     };
     let body = protocol::to_body(&NativeServerHelloBody { hello })?;
     let out = protocol::encode_plain(PacketKind::NativeServerHello, pending_id, 1, 0, &body)?;
-    socket.send_to(&out, peer).await?;
+    let _ = send_udp(socket, &out, peer).await?;
     Ok(())
 }
 
@@ -1058,7 +1058,7 @@ async fn handle_native_user_auth(
             SERVER_TO_CLIENT,
             &body,
         )?;
-        socket.send_to(&out, peer).await?;
+        let _ = send_udp(socket, &out, peer).await?;
         return Ok(());
     }
 
@@ -1251,7 +1251,7 @@ async fn handle_native_user_auth(
         SERVER_TO_CLIENT,
         &body,
     )?;
-    socket.send_to(&out, peer).await?;
+    let _ = send_udp(socket, &out, peer).await?;
     Ok(())
 }
 
@@ -1359,7 +1359,7 @@ async fn handle_bootstrap_attach(
         SERVER_TO_CLIENT,
         &body,
     )?;
-    socket.send_to(&out, peer).await?;
+    let _ = send_udp(socket, &out, peer).await?;
     Ok(())
 }
 
@@ -1494,7 +1494,7 @@ async fn handle_ticket_attach(
     };
     let body = protocol::to_body(&envelope)?;
     let out = protocol::encode_plain(PacketKind::AttachOk, client_id, 1, 0, &body)?;
-    socket.send_to(&out, peer).await?;
+    let _ = send_udp(socket, &out, peer).await?;
     Ok(())
 }
 
@@ -1512,7 +1512,7 @@ async fn send_reject_to_client(
         reason: reason.to_string(),
     })?;
     let out = protocol::encode_plain(PacketKind::AttachReject, client_id, 0, 0, &body)?;
-    socket.send_to(&out, peer).await?;
+    let _ = send_udp(socket, &out, peer).await?;
     Ok(())
 }
 
@@ -1578,7 +1578,7 @@ async fn handle_resume(
         SERVER_TO_CLIENT,
         &body,
     )?;
-    socket.send_to(&out, peer).await?;
+    let _ = send_udp(socket, &out, peer).await?;
     Ok(())
 }
 
@@ -1690,7 +1690,7 @@ async fn handle_ping(
         SERVER_TO_CLIENT,
         b"",
     )?;
-    socket.send_to(&out, peer).await?;
+    let _ = send_udp(socket, &out, peer).await?;
     Ok(())
 }
 
@@ -1705,6 +1705,60 @@ fn remove_client(state: &Arc<Mutex<ServerState>>, client_id: [u8; 16]) {
     locked.remove_client_everywhere(&client_id);
 }
 
+async fn send_udp(socket: &UdpSocket, packet: &[u8], peer: SocketAddr) -> Result<bool> {
+    match socket.send_to(packet, peer).await {
+        Ok(_) => Ok(true),
+        Err(err) if is_transient_udp_send_error(&err) => Ok(false),
+        Err(err) => Err(err).with_context(|| format!("send UDP packet to {peer}")),
+    }
+}
+
+fn is_transient_udp_send_error(err: &std::io::Error) -> bool {
+    matches!(
+        err.kind(),
+        std::io::ErrorKind::Interrupted
+            | std::io::ErrorKind::TimedOut
+            | std::io::ErrorKind::WouldBlock
+    ) || err.raw_os_error().is_some_and(is_transient_udp_os_error)
+}
+
+#[cfg(unix)]
+fn is_transient_udp_os_error(code: i32) -> bool {
+    matches!(
+        code,
+        libc::EADDRNOTAVAIL
+            | libc::ECONNREFUSED
+            | libc::ECONNRESET
+            | libc::EHOSTDOWN
+            | libc::EHOSTUNREACH
+            | libc::ENETDOWN
+            | libc::ENETRESET
+            | libc::ENETUNREACH
+            | libc::ETIMEDOUT
+    )
+}
+
+#[cfg(windows)]
+fn is_transient_udp_os_error(code: i32) -> bool {
+    matches!(
+        code,
+        10049  // WSAEADDRNOTAVAIL
+            | 10050 // WSAENETDOWN
+            | 10051 // WSAENETUNREACH
+            | 10052 // WSAENETRESET
+            | 10054 // WSAECONNRESET
+            | 10060 // WSAETIMEDOUT
+            | 10061 // WSAECONNREFUSED
+            | 10064 // WSAEHOSTDOWN
+            | 10065 // WSAEHOSTUNREACH
+    )
+}
+
+#[cfg(not(any(unix, windows)))]
+fn is_transient_udp_os_error(_code: i32) -> bool {
+    false
+}
+
 async fn handle_ack(
     state: &Arc<Mutex<ServerState>>,
     peer: SocketAddr,
@@ -3020,7 +3074,7 @@ async fn send_stream_open_to_client(
         found
     };
     if let Some((endpoint, packet)) = send {
-        socket.send_to(&packet, endpoint).await?;
+        let _ = send_udp(socket, &packet, endpoint).await?;
     }
     Ok(())
 }
@@ -3114,7 +3168,7 @@ async fn queue_or_send_stream_data_to_client(
         send
     };
     if let Some((endpoint, packet)) = send {
-        socket.send_to(&packet, endpoint).await?;
+        let _ = send_udp(socket, &packet, endpoint).await?;
     }
     Ok(())
 }
@@ -3189,7 +3243,7 @@ async fn flush_stream_pending_data_to_client(
         let Some((endpoint, packet)) = send else {
             return Ok(());
         };
-        socket.send_to(&packet, endpoint).await?;
+        let _ = send_udp(socket, &packet, endpoint).await?;
     }
 }
 
@@ -3333,7 +3387,7 @@ async fn send_stream_packet_to_client(
         found
     };
     if let Some((endpoint, packet)) = send {
-        socket.send_to(&packet, endpoint).await?;
+        let _ = send_udp(socket, &packet, endpoint).await?;
     }
     Ok(())
 }
@@ -3435,7 +3489,7 @@ async fn broadcast_output(
         }
     }
     for (endpoint, packet) in sends {
-        socket.send_to(&packet, endpoint).await?;
+        let _ = send_udp(socket, &packet, endpoint).await?;
     }
     Ok(())
 }
@@ -3488,7 +3542,7 @@ async fn broadcast_session_exit(
         persist::remove_runtime_dir(&sessions_dir, &session_name);
     }
     for (endpoint, packet) in sends {
-        socket.send_to(&packet, endpoint).await?;
+        let _ = send_udp(socket, &packet, endpoint).await?;
     }
     Ok(())
 }
@@ -3595,7 +3649,7 @@ async fn retransmit_pending(
         sends
     };
     for (endpoint, packet) in sends {
-        socket.send_to(&packet, endpoint).await?;
+        let _ = send_udp(socket, &packet, endpoint).await?;
     }
     Ok(())
 }
@@ -3691,7 +3745,7 @@ async fn maybe_rekey_clients(
         sends
     };
     for (endpoint, packet) in sends {
-        socket.send_to(&packet, endpoint).await?;
+        let _ = send_udp(socket, &packet, endpoint).await?;
     }
     Ok(())
 }
@@ -4536,4 +4590,25 @@ mod tests {
         };
         remote_bind_allowed(&config, "0.0.0.0").unwrap();
     }
+
+    #[cfg(unix)]
+    #[test]
+    fn server_udp_send_classifier_treats_network_roaming_errors_as_transient() {
+        for code in [
+            libc::EADDRNOTAVAIL,
+            libc::ECONNREFUSED,
+            libc::ECONNRESET,
+            libc::EHOSTDOWN,
+            libc::EHOSTUNREACH,
+            libc::ENETDOWN,
+            libc::ENETRESET,
+            libc::ENETUNREACH,
+            libc::ETIMEDOUT,
+        ] {
+            assert!(
+                is_transient_udp_send_error(&std::io::Error::from_raw_os_error(code)),
+                "expected errno {code} to be transient"
+            );
+        }
+    }
 }