Fix sending of leftover packets.

There was a logic error in the code. The nextP variable was used to
store the packet that was too big to pack into the most recent DNS
response. But nextP was not tied to any particular ClientID; instead it
would be sent to whatever client happened to be the recipient of the
next response.

The confusion didn't cause connections to fail completely; any
misdirected packets were treated as out-of-sequence garbage by KCP and
dropped. But it hurt performance a lot: I saw a download go from 300
KB/s to 50 KB/s just by connecting a second client with a different
ClientID (not even sending or receiving with the second client). The
reason is that a fraction of the packets intended for the downloading
client were instead sent to the idle client, which to the downloading
client looks like a packet drop, requiring a retransmission by the
server.

We fix it by placing the leftover packet in a per-ClientID stash, rather
than a variable shared by all ClientIDs.

dnstt-server/main.go

@@ -541,7 +541,6 @@ func recvLoop(domain dns.Name, dnsConn net.PacketConn, ttConn *turbotunnel.Queue
 // fit while keeping the total size under maxEncodedPayload, then sends it.
 func sendLoop(dnsConn net.PacketConn, ttConn *turbotunnel.QueuePacketConn, ch <-chan *record, maxEncodedPayload int) error {
 	var nextRec *record
-	var nextP []byte
 	for {
 		rec := nextRec
 		nextRec = nil
@@ -571,67 +570,62 @@ func sendLoop(dnsConn net.PacketConn, ttConn *turbotunnel.QueuePacketConn, ch <-
 			}
 
 			var payload bytes.Buffer
-
 			limit := maxEncodedPayload
-			if len(nextP) > 0 {
-				// No length check on any packet left over from
-				// the previous bundle -- if it's too large, we
-				// allow it to be truncated and dropped.
-				limit -= 2 + len(nextP)
-				binary.Write(&payload, binary.BigEndian, uint16(len(nextP)))
-				payload.Write(nextP)
-			}
-			nextP = nil
-
-			// We loop and write as many packets from OutgoingQueue
+			// We loop and bundle as many packets from OutgoingQueue
 			// into the response as will fit. Any packet that would
-			// overflow the capacity of the DNS response, we save in
-			// nextP to be included in a future response.
+			// overflow the capacity of the DNS response, we stash
+			// to be bundled into a future response.
 			timer := time.NewTimer(maxResponseDelay)
 		loop:
 			for {
+				var p []byte
 				select {
-				// Prioritize the first two cases over the
-				// OutgoingQueue case. The first two cases are
-				// duplicated under the default case.
+				// Check the nextRec, timer, and stash cases
+				// before considering the OutgoingQueue case.
+				// Only if all these cases fail do we enter the
+				// default arm, where they are checked again in
+				// addition to OutgoingQueue.
 				case nextRec = <-ch:
-					// If there's another response
-					// waiting to be sent, wait no
-					// longer for a payload for this
-					// one.
+					// If there's another response waiting
+					// to be sent, wait no longer for a
+					// payload for this one.
 					break loop
 				case <-timer.C:
 					break loop
+				case p = <-ttConn.Unstash(rec.ClientID):
 				default:
 					select {
 					case nextRec = <-ch:
 						break loop
 					case <-timer.C:
 						break loop
-					case p := <-ttConn.OutgoingQueue(rec.ClientID):
-						// We wait for the first packet
-						// in a bundle only. The second
-						// and later packets must be
-						// immediately available or they
-						// will be omitted from this
-						// send.
-						timer.Reset(0)
-
-						if int(uint16(len(p))) != len(p) {
-							panic(len(p))
-						}
-						if 2+len(p) > limit {
-							// Save this packet to
-							// send in the next
-							// response.
-							nextP = p
-							break loop
-						}
-						limit -= 2 + len(p)
-						binary.Write(&payload, binary.BigEndian, uint16(len(p)))
-						payload.Write(p)
+					case p = <-ttConn.Unstash(rec.ClientID):
+					case p = <-ttConn.OutgoingQueue(rec.ClientID):
 					}
 				}
+				// We wait for the first packet in a bundle
+				// only. The second and later packets must be
+				// immediately available or they will be omitted
+				// from this bundle.
+				timer.Reset(0)
+
+				limit -= 2 + len(p)
+				if payload.Len() == 0 {
+					// No packet length check for the first
+					// packet; if it's too large, we allow
+					// it to be truncated and dropped by the
+					// receiver.
+				} else if limit < 0 {
+					// Stash this packet to send in the next
+					// response.
+					ttConn.Stash(p, rec.ClientID)
+					break loop
+				}
+				if int(uint16(len(p))) != len(p) {
+					panic(len(p))
+				}
+				binary.Write(&payload, binary.BigEndian, uint16(len(p)))
+				payload.Write(p)
 			}
 			timer.Stop()