Harden reconnect paths under UDP churn

src/bin/dosh-client.rs

@@ -36,7 +36,10 @@ use dosh::transport::{
     DoshTransport, SessionEvent, SessionRole, SessionTransportConfig, TransportConfig,
     TransportEvent,
 };
-use dosh::udp::{is_transient_udp_error, is_transient_udp_send_error};
+use dosh::udp::{
+    is_transient_udp_error, is_transient_udp_send_error, recv_udp_retrying_transient,
+    send_udp_retrying_transient,
+};
 use serde::{Deserialize, Serialize};
 use sha2::{Digest, Sha256};
 use std::collections::BTreeMap;
@@ -3789,14 +3792,11 @@ async fn try_native_auth(
             hello: hello.clone(),
         })?,
     )?;
-    socket.send_to(&packet, addr).await?;
+    let auth_timeout = Duration::from_millis(config.native_auth_timeout_ms.max(1));
+    send_udp_retrying_transient(socket, &packet, addr, auth_timeout).await?;
 
     let mut buf = vec![0u8; 65535];
-    let (n, _) = tokio::time::timeout(
-        Duration::from_millis(config.native_auth_timeout_ms.max(1)),
-        socket.recv_from(&mut buf),
-    )
-    .await??;
+    let (n, _) = recv_udp_retrying_transient(socket, &mut buf, auth_timeout).await?;
     let packet = protocol::decode(&buf[..n])?;
     if packet.header.kind != PacketKind::NativeServerHello {
         if packet.header.kind == PacketKind::AttachReject {
@@ -3859,13 +3859,9 @@ async fn try_native_auth(
         CLIENT_TO_SERVER,
         &auth_body,
     )?;
-    socket.send_to(&packet, addr).await?;
+    send_udp_retrying_transient(socket, &packet, addr, auth_timeout).await?;
 
-    let (n, _) = tokio::time::timeout(
-        Duration::from_millis(config.native_auth_timeout_ms.max(1)),
-        socket.recv_from(&mut buf),
-    )
-    .await??;
+    let (n, _) = recv_udp_retrying_transient(socket, &mut buf, auth_timeout).await?;
     let packet = protocol::decode(&buf[..n])?;
     if packet.header.kind != PacketKind::NativeAuthOk {
         if packet.header.kind == PacketKind::AttachReject {
@@ -3947,14 +3943,11 @@ async fn try_native_auth_check(
             hello: hello.clone(),
         })?,
     )?;
-    socket.send_to(&packet, addr).await?;
+    let auth_timeout = Duration::from_millis(config.native_auth_timeout_ms.max(1));
+    send_udp_retrying_transient(socket, &packet, addr, auth_timeout).await?;
 
     let mut buf = vec![0u8; 65535];
-    let (n, _) = tokio::time::timeout(
-        Duration::from_millis(config.native_auth_timeout_ms.max(1)),
-        socket.recv_from(&mut buf),
-    )
-    .await??;
+    let (n, _) = recv_udp_retrying_transient(socket, &mut buf, auth_timeout).await?;
     let packet = protocol::decode(&buf[..n])?;
     if packet.header.kind != PacketKind::NativeServerHello {
         if packet.header.kind == PacketKind::AttachReject {
@@ -4016,13 +4009,9 @@ async fn try_native_auth_check(
         CLIENT_TO_SERVER,
         &protocol::to_body(&NativeUserAuthBody { auth })?,
     )?;
-    socket.send_to(&packet, addr).await?;
+    send_udp_retrying_transient(socket, &packet, addr, auth_timeout).await?;
 
-    let (n, _) = tokio::time::timeout(
-        Duration::from_millis(config.native_auth_timeout_ms.max(1)),
-        socket.recv_from(&mut buf),
-    )
-    .await??;
+    let (n, _) = recv_udp_retrying_transient(socket, &mut buf, auth_timeout).await?;
     let packet = protocol::decode(&buf[..n])?;
     if packet.header.kind != PacketKind::NativeAuthCheckOk {
         if packet.header.kind == PacketKind::AttachReject {
@@ -4182,7 +4171,7 @@ async fn bootstrap_attach(
     let body = protocol::to_body(&req)?;
     let packet =
         protocol::encode_plain(PacketKind::BootstrapAttachRequest, [0u8; 16], 1, 0, &body)?;
-    socket.send_to(&packet, addr).await?;
+    send_udp_retrying_transient(socket, &packet, addr, Duration::from_secs(5)).await?;
     let expected_key_id = protocol::session_key_id(&bootstrap.session_key);
     let ok = recv_response_until(socket, Duration::from_secs(5), |packet| {
         if packet.header.kind == PacketKind::AttachReject && packet.header.conn_id == [0u8; 16] {
@@ -4262,7 +4251,7 @@ async fn try_ticket_attach(
         0,
         &protocol::to_body(&envelope)?,
     )?;
-    socket.send_to(&packet, addr).await?;
+    send_udp_retrying_transient(socket, &packet, addr, Duration::from_millis(700)).await?;
 
     let ok = recv_response_until(socket, Duration::from_millis(700), |packet| {
         if packet.header.kind == PacketKind::AttachReject && packet.header.conn_id == [0u8; 16] {
@@ -4338,7 +4327,7 @@ async fn try_resume_fresh_process(
         CLIENT_TO_SERVER,
         &body,
     )?;
-    socket.send_to(&packet, addr).await?;
+    send_udp_retrying_transient(socket, &packet, addr, Duration::from_millis(700)).await?;
     let frame = recv_response_until(socket, Duration::from_millis(700), |packet| {
         if !is_resume_response_for_client(&packet, cached.client_id) {
             return Ok(None);
@@ -4393,7 +4382,7 @@ async fn try_live_resume(
         &body,
     )?;
     *send_seq += 1;
-    socket.send_to(&packet, addr).await?;
+    send_udp_retrying_transient(socket, &packet, addr, Duration::from_millis(700)).await?;
     let frame = recv_response_until(socket, Duration::from_millis(700), |packet| {
         if !is_resume_response_for_client(&packet, cached.client_id) {
             return Ok(None);
@@ -4440,14 +4429,9 @@ where
             return Err(anyhow!("timed out waiting for server response"));
         }
         let remaining = wait - elapsed;
-        let (n, _) = match tokio::time::timeout(remaining, socket.recv_from(&mut buf))
+        let (n, _) = recv_udp_retrying_transient(socket, &mut buf, remaining)
             .await
-            .context("timed out waiting for server response")?
-        {
-            Ok(value) => value,
-            Err(err) if is_transient_udp_error(&err) => continue,
-            Err(err) => return Err(err.into()),
-        };
+            .context("timed out waiting for server response")?;
         let Ok(packet) = protocol::decode(&buf[..n]) else {
             continue;
         };