refactor: introduce reliable transport layer (#50)

This is the third commit in the series of incremental refactoring of the
current minivpn tree.

In this commit, we introduce the reliabletransport package, which is the
layer between packetmuxer and control channel.

Reliable transport adds a reliability layer on top of UDP (but it's also
used in TCP mode).

This first implementation of the reliability layer implements only a
naive strategy to get the TLS handshake working on an optimal
environment (i.e., no packet loss). We ACK any incoming packet, and we
assume all packets arrive in order.

After merging all the components in the new architecture, we will
revisit the reliability layer to follow the OpenVPN design more closely.

Reference issue: #47

---------

Co-authored-by: Simone Basso <[email protected]>

internal/reliabletransport/reliabletransport.go

@@ -0,0 +1,187 @@
+// Package reliabletransport implements the reliable transport.
+package reliabletransport
+
+import (
+	"bytes"
+
+	"github.com/ooni/minivpn/internal/model"
+	"github.com/ooni/minivpn/internal/session"
+	"github.com/ooni/minivpn/internal/workers"
+)
+
+// Service is the reliable service. Make sure you initialize
+// the channels before invoking [Service.StartWorkers].
+type Service struct {
+	// DataOrControlToMuxer is a shared channel that moves packets down to the muxer
+	DataOrControlToMuxer *chan *model.Packet
+
+	// ControlToReliable moves packets down to us
+	ControlToReliable chan *model.Packet
+
+	// MuxerToReliable moves packets up to us
+	MuxerToReliable chan *model.Packet
+
+	// ReliableToControl moves packets up from us to the control layer above
+	ReliableToControl *chan *model.Packet
+}
+
+// StartWorkers starts the reliable-transport workers. See the [ARCHITECTURE]
+// file for more information about the reliable-transport workers.
+//
+// [ARCHITECTURE]: https://github.com/ooni/minivpn/blob/main/ARCHITECTURE.md
+func (s *Service) StartWorkers(
+	logger model.Logger,
+	workersManager *workers.Manager,
+	sessionManager *session.Manager,
+) {
+	ws := &workersState{
+		logger:               logger,
+		dataOrControlToMuxer: *s.DataOrControlToMuxer,
+		controlToReliable:    s.ControlToReliable,
+		muxerToReliable:      s.MuxerToReliable,
+		reliableToControl:    *s.ReliableToControl,
+		sessionManager:       sessionManager,
+		workersManager:       workersManager,
+	}
+	workersManager.StartWorker(ws.moveUpWorker)
+	workersManager.StartWorker(ws.moveDownWorker)
+}
+
+// workersState contains the reliable workers state
+type workersState struct {
+	// logger is the logger to use
+	logger model.Logger
+
+	// dataOrControlToMuxer is the channel where we write packets going down the stack.
+	dataOrControlToMuxer chan<- *model.Packet
+
+	// controlToReliable is the channel from which we read packets going down the stack.
+	controlToReliable <-chan *model.Packet
+
+	// muxerToReliable is the channel from which we read packets going up the stack.
+	muxerToReliable <-chan *model.Packet
+
+	// reliableToControl is the channel where we write packets going up the stack.
+	reliableToControl chan<- *model.Packet
+
+	// sessionManager manages the OpenVPN session.
+	sessionManager *session.Manager
+
+	// workersManager controls the workers lifecycle.
+	workersManager *workers.Manager
+}
+
+// moveUpWorker moves packets up the stack
+func (ws *workersState) moveUpWorker() {
+	defer func() {
+		ws.workersManager.OnWorkerDone()
+		ws.workersManager.StartShutdown()
+		ws.logger.Debug("reliable: moveUpWorker: done")
+	}()
+
+	ws.logger.Debug("reliable: moveUpWorker: started")
+
+	// TODO: do we need to have notifications from the control channel
+	// to reset state or can we do this implicitly?
+
+	for {
+		// POSSIBLY BLOCK reading a packet to move up the stack
+		// or POSSIBLY BLOCK waiting for notifications
+		select {
+		case packet := <-ws.muxerToReliable:
+			ws.logger.Infof(
+				"< %s localID=%x remoteID=%x [%d bytes]",
+				packet.Opcode,
+				packet.LocalSessionID,
+				packet.RemoteSessionID,
+				len(packet.Payload),
+			)
+
+			// drop a packet that is not for our session
+			if !bytes.Equal(packet.LocalSessionID[:], ws.sessionManager.RemoteSessionID()) {
+				ws.logger.Warnf(
+					"reliable: moveUpWorker: packet with invalid RemoteSessionID: expected %x; got %x",
+					ws.sessionManager.LocalSessionID(),
+					packet.RemoteSessionID,
+				)
+				continue
+			}
+
+			// possibly ACK the incoming packet
+			if err := ws.maybeACK(packet); err != nil {
+				ws.logger.Warnf("reliable: moveUpWorker: cannot ACK packet: %s", err.Error())
+				continue
+			}
+
+			// TODO: here we should track ACKs
+
+			// POSSIBLY BLOCK delivering to the upper layer
+			select {
+			case ws.reliableToControl <- packet:
+			case <-ws.workersManager.ShouldShutdown():
+				return
+			}
+
+		case <-ws.workersManager.ShouldShutdown():
+			return
+		}
+	}
+}
+
+// moveDownWorker moves packets down the stack
+func (ws *workersState) moveDownWorker() {
+	defer func() {
+		ws.workersManager.OnWorkerDone()
+		ws.workersManager.StartShutdown()
+		ws.logger.Debug("reliable: moveDownWorker: done")
+	}()
+
+	ws.logger.Debug("reliable: moveDownWorker: started")
+
+	// TODO: we should have timer for retransmission
+	for {
+		// POSSIBLY BLOCK reading the next packet we should move down the stack
+		select {
+		case packet := <-ws.controlToReliable:
+			// TODO: here we should treat control packets specially
+
+			ws.logger.Infof(
+				"> %s localID=%x remoteID=%x [%d bytes]",
+				packet.Opcode,
+				packet.LocalSessionID,
+				packet.RemoteSessionID,
+				len(packet.Payload),
+			)
+
+			// POSSIBLY BLOCK delivering this packet to the lower layer
+			select {
+			case ws.dataOrControlToMuxer <- packet:
+			case <-ws.workersManager.ShouldShutdown():
+				return
+			}
+
+		case <-ws.workersManager.ShouldShutdown():
+			return
+		}
+	}
+}
+
+// maybeACK sends an ACK when needed.
+func (ws *workersState) maybeACK(packet *model.Packet) error {
+	// currently we are ACKing every packet
+	// TODO: implement better ACKing strategy
+
+	// this function will fail if we don't know the remote session ID
+	ACK, err := ws.sessionManager.NewACKForPacket(packet)
+	if err != nil {
+		return err
+	}
+
+	// move the packet down
+	select {
+	case ws.dataOrControlToMuxer <- ACK:
+		return nil
+	case <-ws.workersManager.ShouldShutdown():
+		return workers.ErrShutdown
+	}
+}