Merge pull request 'exchanges' (#20) from exchanges into develop

Reviewed-on: #20

config/builder.go

@@ -11,6 +11,7 @@ import (
 	"mpbl3p/tcp"
 	"mpbl3p/udp"
 	"mpbl3p/udp/congestion"
+	"mpbl3p/udp/congestion/newreno"
 	"time"
 )
 
@@ -127,7 +128,7 @@ func buildUdp(
 	default:
 		fallthrough
 	case "NewReno":
-		c = func() udp.Congestion { return congestion.NewNewReno() }
+		c = func() udp.Congestion { return newreno.NewNewReno() }
 	}
 
 	if peer.RemoteHost != "" {

flags/flags.go

@@ -1,14 +1,12 @@
 package flags
 
 import (
-	"errors"
 	"fmt"
 	goflags "github.com/jessevdk/go-flags"
 	"os"
 )
 
 var PrintedHelpErr = goflags.ErrHelp
-var NotEnoughArgs = errors.New("not enough arguments")
 
 type Options struct {
 	Foreground bool   `short:"f" long:"foreground" description:"Run in the foreground"`

flags/locs_darwin.go

@@ -0,0 +1,4 @@
+package flags
+
+const DefaultConfigFile = "/usr/local/etc/netcombiner/%s"
+const DefaultPidFile = "/var/run/netcombiner/%s.pid"

mocks/packetconn.go

@@ -1,6 +1,9 @@
 package mocks
 
-import "net"
+import (
+	"net"
+	"time"
+)
 
 type MockPerfectBiPacketConn struct {
 	directionA chan []byte
@@ -44,6 +47,10 @@ func (c MockPerfectPacketConn) LocalAddr() net.Addr {
 	}
 }
 
+func (c MockPerfectPacketConn) SetReadDeadline(time.Time) error {
+	return nil
+}
+
 func (c MockPerfectPacketConn) ReadFromUDP(b []byte) (int, *net.UDPAddr, error) {
 	p := <-c.inbound
 	return copy(b, p), &net.UDPAddr{

proxy/exchange.go

@@ -0,0 +1,9 @@
+package proxy
+
+import "context"
+
+type Exchange interface {
+	Initial(ctx context.Context) (out []byte, err error)
+	Handle(ctx context.Context, in []byte) (out []byte, data []byte, err error)
+	Complete() bool
+}

proxy/proxy.go

@@ -75,12 +75,13 @@ func (p Proxy) AddConsumer(ctx context.Context, c Consumer) {
 			if reconnectable {
 				var err error
 				for once := true; err != nil || once; once = false {
-					log.Printf("attempting to connect consumer `%v`\n", c)
+					if err := ctx.Err(); err != nil {
-					err = c.(Reconnectable).Reconnect(ctx)
-					if errors.Is(err, context.Canceled) || errors.Is(err, context.DeadlineExceeded) {
 						log.Printf("closed consumer `%v` (context)\n", c)
 						return
 					}
+
+					log.Printf("attempting to connect consumer `%v`\n", c)
+					err = c.(Reconnectable).Reconnect(ctx)
 					if !once {
 						time.Sleep(time.Second)
 					}
@@ -118,12 +119,13 @@ func (p Proxy) AddProducer(ctx context.Context, pr Producer) {
 			if reconnectable {
 				var err error
 				for once := true; err != nil || once; once = false {
-					log.Printf("attempting to connect producer `%v`\n", pr)
+					if err := ctx.Err(); err != nil {
-					err = pr.(Reconnectable).Reconnect(ctx)
-					if errors.Is(err, context.Canceled) || errors.Is(err, context.DeadlineExceeded) {
 						log.Printf("closed producer `%v` (context)\n", pr)
 						return
 					}
+
+					log.Printf("attempting to connect producer `%v`\n", pr)
+					err = pr.(Reconnectable).Reconnect(ctx)
 					if !once {
 						time.Sleep(time.Second)
 					}

shared/errors.go

@@ -5,3 +5,4 @@ import "errors"
 var ErrBadChecksum = errors.New("the packet had a bad checksum")
 var ErrDeadConnection = errors.New("the connection is dead")
 var ErrNotEnoughBytes = errors.New("not enough bytes")
+var ErrBadExchange = errors.New("bad exchange")

udp/congestion/newreno/exchange.go

@@ -0,0 +1,119 @@
+package newreno
+
+import (
+	"context"
+	"encoding/binary"
+	"math/rand"
+	"mpbl3p/shared"
+	"time"
+)
+
+func (c *NewReno) Initial(ctx context.Context) (out []byte, err error) {
+	c.alive = false
+	c.wasInitial = true
+	c.startSequenceLoop(ctx)
+
+	var s uint32
+	select {
+	case s = <-c.sequence:
+	case <-ctx.Done():
+		return nil, ctx.Err()
+	}
+
+	b := make([]byte, 12)
+	binary.LittleEndian.PutUint32(b[8:12], s)
+
+	c.inFlight = []flightInfo{{time.Now(), s}}
+
+	return b, nil
+}
+
+func (c *NewReno) Handle(ctx context.Context, in []byte) (out []byte, data []byte, err error) {
+	if c.alive || c.stopSequence == nil {
+		// reset
+		c.alive = false
+		c.startSequenceLoop(ctx)
+	}
+
+	// receive
+	if len(in) != 12 {
+		return nil, nil, shared.ErrBadExchange
+	}
+
+	rcvAck := binary.LittleEndian.Uint32(in[0:4])
+	rcvNack := binary.LittleEndian.Uint32(in[4:8])
+	rcvSeq := binary.LittleEndian.Uint32(in[8:12])
+
+	// verify
+	if rcvNack != 0 {
+		return nil, nil, shared.ErrBadExchange
+	}
+
+	var seq uint32
+
+	if c.wasInitial {
+		if rcvAck == c.inFlight[0].sequence {
+			c.ack, c.lastAck = rcvSeq, rcvSeq
+			c.alive, c.inFlight = true, nil
+		} else {
+			return nil, nil, shared.ErrBadExchange
+		}
+	} else { // if !c.wasInitial
+		if rcvAck == 0 {
+			// theirs is a syn packet
+			c.ack, c.lastAck = rcvSeq, rcvSeq
+
+			select {
+			case seq = <-c.sequence:
+			case <-ctx.Done():
+				return nil, nil, ctx.Err()
+			}
+
+			c.inFlight = []flightInfo{{time.Now(), seq}}
+		} else if len(c.inFlight) == 1 && rcvAck == c.inFlight[0].sequence {
+			c.alive, c.inFlight = true, nil
+		} else {
+			return nil, nil, shared.ErrBadExchange
+		}
+	}
+
+	// respond
+	b := make([]byte, 12)
+	binary.LittleEndian.PutUint32(b[0:4], c.ack)
+	binary.LittleEndian.PutUint32(b[8:12], seq)
+
+	return b, nil, nil
+}
+
+func (c *NewReno) Complete() bool {
+	return c.alive
+}
+
+func (c *NewReno) startSequenceLoop(ctx context.Context) {
+	if c.stopSequence != nil {
+		c.stopSequence()
+	}
+
+	var s uint32
+	for s == 0 {
+		s = rand.Uint32()
+	}
+
+	ctx, c.stopSequence = context.WithCancel(ctx)
+	go func() {
+		s := s
+		for {
+			if s == 0 {
+				s++
+				continue
+			}
+
+			select {
+			case c.sequence <- s:
+			case <-ctx.Done():
+				return
+			}
+			s++
+		}
+	}()
+}

udp/congestion/newreno/exchange_test.go

@@ -0,0 +1,57 @@
+package newreno
+
+import (
+	"context"
+	"encoding/binary"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"testing"
+)
+
+func TestNewReno_InitialHandle(t *testing.T) {
+	t.Run("InitialAckNackAreZero", func(t *testing.T) {
+		// ASSIGN
+		a := NewNewReno()
+
+		ctx := context.Background()
+
+		// ACT
+		initial, err := a.Initial(ctx)
+
+		ack := binary.LittleEndian.Uint32(initial[0:4])
+		nack := binary.LittleEndian.Uint32(initial[4:8])
+
+		// ASSERT
+		require.Nil(t, err)
+
+		assert.Zero(t, ack)
+		assert.Zero(t, nack)
+	})
+
+	t.Run("InitialHandledWithAck", func(t *testing.T) {
+		// ASSIGN
+		a := NewNewReno()
+		b := NewNewReno()
+
+		ctx := context.Background()
+
+		// ACT
+		initial, err := a.Initial(ctx)
+		require.Nil(t, err)
+
+		initialSeq := binary.LittleEndian.Uint32(initial[8:12])
+
+		response, data, err := b.Handle(ctx, initial)
+
+		ack := binary.LittleEndian.Uint32(response[0:4])
+		nack := binary.LittleEndian.Uint32(response[4:8])
+
+		// ASSERT
+		require.Nil(t, err)
+
+		assert.Equal(t, initialSeq, ack)
+		assert.Zero(t, nack)
+
+		assert.Nil(t, data)
+	})
+}

udp/congestion/newreno/newreno.go

@@ -1,4 +1,4 @@
-package congestion
+package newreno
 
 import (
 	"context"
@@ -11,10 +11,12 @@ import (
 )
 
 const RttExponentialFactor = 0.1
-const RttLossDelay = 1.5
+const RttLossDelay = 0.5
 
 type NewReno struct {
-	sequence chan uint32
+	sequence          chan uint32
+	stopSequence      context.CancelFunc
+	wasInitial, alive bool
 
 	inFlight   []flightInfo
 	lastSent   time.Time
@@ -64,19 +66,6 @@ func NewNewReno() *NewReno {
 		slowStart:  true,
 	}
 
-	go func() {
-		var s uint32
-		for {
-			if s == 0 {
-				s++
-				continue
-			}
-
-			c.sequence <- s
-			s++
-		}
-	}()
-
 	return &c
 }
 

udp/congestion/newreno/newreno_test.go

@@ -1,4 +1,4 @@
-package congestion
+package newreno
 
 import (
 	"context"
@@ -21,8 +21,8 @@ type newRenoTest struct {
 	halfRtt time.Duration
 }
 
-func newNewRenoTest(rtt time.Duration) *newRenoTest {
+func newNewRenoTest(ctx context.Context, rtt time.Duration) *newRenoTest {
-	return &newRenoTest{
+	nr := &newRenoTest{
 		sideA: NewNewReno(),
 		sideB: NewNewReno(),
 
@@ -34,6 +34,15 @@ func newNewRenoTest(rtt time.Duration) *newRenoTest {
 
 		halfRtt: rtt / 2,
 	}
+
+	p, _ := nr.sideA.Initial(ctx)
+	p, _, _ = nr.sideB.Handle(ctx, p)
+	p, _, _ = nr.sideA.Handle(ctx, p)
+
+	nr.sideB.ReceivedPacket(0, nr.sideA.NextAck(), nr.sideA.NextNack())
+	nr.sideA.ReceivedPacket(0, nr.sideB.NextAck(), nr.sideB.NextNack())
+
+	return nr
 }
 
 func (n *newRenoTest) Start(ctx context.Context) {
@@ -151,11 +160,14 @@ func TestNewReno_Congestion(t *testing.T) {
 			ctx, cancel := context.WithCancel(context.Background())
 			defer cancel()
 
-			c := newNewRenoTest(rtt)
+			c := newNewRenoTest(ctx, rtt)
 			c.Start(ctx)
 			c.RunSideA(ctx)
 			c.RunSideB(ctx)
 
+			sideAinitialAck := c.sideA.ack
+			sideBinitialAck := c.sideB.ack
+
 			// ACT
 			for i := 0; i < numPackets; i++ {
 				// sleep to simulate preparing packet
@@ -175,10 +187,10 @@ func TestNewReno_Congestion(t *testing.T) {
 			// ASSERT
 
 			assert.Equal(t, uint32(0), c.sideA.nack)
-			assert.Equal(t, uint32(0), c.sideA.ack)
+			assert.Equal(t, sideAinitialAck, c.sideA.ack)
 
 			assert.Equal(t, uint32(0), c.sideB.nack)
-			assert.Equal(t, uint32(numPackets), c.sideB.ack)
+			assert.Equal(t, sideBinitialAck+uint32(numPackets), c.sideB.ack)
 		})
 
 		t.Run("SequenceLoss", func(t *testing.T) {
@@ -189,18 +201,21 @@ func TestNewReno_Congestion(t *testing.T) {
 			ctx, cancel := context.WithCancel(context.Background())
 			defer cancel()
 
-			c := newNewRenoTest(rtt)
+			c := newNewRenoTest(ctx, rtt)
 			c.Start(ctx)
 			c.RunSideA(ctx)
 			c.RunSideB(ctx)
 
+			sideAinitialAck := c.sideA.ack
+			sideBinitialAck := c.sideB.ack
+
 			// ACT
-			for i := 0; i < numPackets; i++ {
+			for i := 1; i <= numPackets; i++ {
 				// sleep to simulate preparing packet
 				time.Sleep(1 * time.Millisecond)
 				seq, _ := c.sideA.Sequence(ctx)
 
-				if seq == 20 {
+				if i == 20 {
 					// Simulate packet loss of sequence 20
 					continue
 				}
@@ -217,10 +232,10 @@ func TestNewReno_Congestion(t *testing.T) {
 			// ASSERT
 
 			assert.Equal(t, uint32(0), c.sideA.nack)
-			assert.Equal(t, uint32(0), c.sideA.ack)
+			assert.Equal(t, sideAinitialAck, c.sideA.ack)
 
-			assert.Equal(t, uint32(20), c.sideB.nack)
+			assert.Equal(t, sideBinitialAck+uint32(20), c.sideB.nack)
-			assert.Equal(t, uint32(numPackets), c.sideB.ack)
+			assert.Equal(t, sideBinitialAck+uint32(numPackets), c.sideB.ack)
 		})
 	})
 
@@ -233,16 +248,19 @@ func TestNewReno_Congestion(t *testing.T) {
 			ctx, cancel := context.WithCancel(context.Background())
 			defer cancel()
 
-			c := newNewRenoTest(rtt)
+			c := newNewRenoTest(ctx, rtt)
 			c.Start(ctx)
 			c.RunSideA(ctx)
 			c.RunSideB(ctx)
 
+			sideAinitialAck := c.sideA.ack
+			sideBinitialAck := c.sideB.ack
+
 			// ACT
 			done := make(chan struct{})
 
 			go func() {
-				for i := 0; i < numPackets; i++ {
+				for i := 1; i <= numPackets; i++ {
 					time.Sleep(1 * time.Millisecond)
 					seq, _ := c.sideA.Sequence(ctx)
 
@@ -257,7 +275,7 @@ func TestNewReno_Congestion(t *testing.T) {
 			}()
 
 			go func() {
-				for i := 0; i < numPackets; i++ {
+				for i := 1; i <= numPackets; i++ {
 					time.Sleep(1 * time.Millisecond)
 					seq, _ := c.sideB.Sequence(ctx)
 
@@ -279,10 +297,10 @@ func TestNewReno_Congestion(t *testing.T) {
 			// ASSERT
 
 			assert.Equal(t, uint32(0), c.sideA.nack)
-			assert.Equal(t, uint32(numPackets), c.sideA.ack)
+			assert.Equal(t, sideAinitialAck+uint32(numPackets), c.sideA.ack)
 
 			assert.Equal(t, uint32(0), c.sideB.nack)
-			assert.Equal(t, uint32(numPackets), c.sideB.ack)
+			assert.Equal(t, sideBinitialAck+uint32(numPackets), c.sideB.ack)
 		})
 
 		t.Run("SequenceLoss", func(t *testing.T) {
@@ -293,20 +311,23 @@ func TestNewReno_Congestion(t *testing.T) {
 			ctx, cancel := context.WithCancel(context.Background())
 			defer cancel()
 
-			c := newNewRenoTest(rtt)
+			c := newNewRenoTest(ctx, rtt)
 			c.Start(ctx)
 			c.RunSideA(ctx)
 			c.RunSideB(ctx)
 
+			sideAinitialAck := c.sideA.ack
+			sideBinitialAck := c.sideB.ack
+
 			// ACT
 			done := make(chan struct{})
 
 			go func() {
-				for i := 0; i < numPackets; i++ {
+				for i := 1; i <= numPackets; i++ {
 					time.Sleep(1 * time.Millisecond)
 					seq, _ := c.sideA.Sequence(ctx)
 
-					if seq == 9 {
+					if i == 9 {
 						// Simulate packet loss of sequence 9
 						continue
 					}
@@ -322,11 +343,11 @@ func TestNewReno_Congestion(t *testing.T) {
 			}()
 
 			go func() {
-				for i := 0; i < numPackets; i++ {
+				for i := 1; i <= numPackets; i++ {
 					time.Sleep(1 * time.Millisecond)
 					seq, _ := c.sideB.Sequence(ctx)
 
-					if seq == 13 {
+					if i == 13 {
 						// Simulate packet loss of sequence 13
 						continue
 					}
@@ -348,11 +369,11 @@ func TestNewReno_Congestion(t *testing.T) {
 
 			// ASSERT
 
-			assert.Equal(t, uint32(13), c.sideA.nack)
+			assert.Equal(t, sideAinitialAck+uint32(13), c.sideA.nack)
-			assert.Equal(t, uint32(numPackets), c.sideA.ack)
+			assert.Equal(t, sideAinitialAck+uint32(numPackets), c.sideA.ack)
 
-			assert.Equal(t, uint32(9), c.sideB.nack)
+			assert.Equal(t, sideBinitialAck+uint32(9), c.sideB.nack)
-			assert.Equal(t, uint32(numPackets), c.sideB.ack)
+			assert.Equal(t, sideBinitialAck+uint32(numPackets), c.sideB.ack)
 		})
 	})
 }

udp/flow.go

@@ -7,7 +7,6 @@ import (
 	"mpbl3p/proxy"
 	"mpbl3p/shared"
 	"net"
-	"sync"
 	"time"
 )
 
@@ -19,29 +18,15 @@ type PacketWriter interface {
 
 type PacketConn interface {
 	PacketWriter
+	SetReadDeadline(t time.Time) error
 	ReadFromUDP(b []byte) (int, *net.UDPAddr, error)
 }
 
-type InitiatedFlow struct {
-	Local  func() string
-	Remote string
-
-	keepalive time.Duration
-
-	mu sync.RWMutex
-	Flow
-}
-
-func (f *InitiatedFlow) String() string {
-	return fmt.Sprintf("UdpOutbound{%v -> %v}", f.Local(), f.Remote)
-}
-
 type Flow struct {
 	writer PacketWriter
 	raddr  *net.UDPAddr
 
 	isAlive    bool
-	startup    bool
 	congestion Congestion
 
 	verifiers  []proxy.MacVerifier
@@ -54,24 +39,6 @@ func (f Flow) String() string {
 	return fmt.Sprintf("UdpInbound{%v -> %v}", f.raddr, f.writer.LocalAddr())
 }
 
-func InitiateFlow(
-	local func() string,
-	remote string,
-	vs []proxy.MacVerifier,
-	gs []proxy.MacGenerator,
-	c Congestion,
-	keepalive time.Duration,
-) (*InitiatedFlow, error) {
-	f := InitiatedFlow{
-		Local:     local,
-		Remote:    remote,
-		Flow:      newFlow(c, vs, gs),
-		keepalive: keepalive,
-	}
-
-	return &f, nil
-}
-
 func newFlow(c Congestion, vs []proxy.MacVerifier, gs []proxy.MacGenerator) Flow {
 	return Flow{
 		inboundDatagrams: make(chan []byte),
@@ -81,102 +48,6 @@ func newFlow(c Congestion, vs []proxy.MacVerifier, gs []proxy.MacGenerator) Flow
 	}
 }
 
-func (f *InitiatedFlow) Reconnect(ctx context.Context) error {
-	f.mu.Lock()
-	defer f.mu.Unlock()
-
-	if f.isAlive {
-		return nil
-	}
-
-	localAddr, err := net.ResolveUDPAddr("udp", f.Local())
-	if err != nil {
-		return err
-	}
-
-	remoteAddr, err := net.ResolveUDPAddr("udp", f.Remote)
-	if err != nil {
-		return err
-	}
-
-	conn, err := net.DialUDP("udp", localAddr, remoteAddr)
-	if err != nil {
-		return err
-	}
-
-	f.writer = conn
-	f.startup = true
-
-	// prod the connection once a second until we get an ack, then consider it alive
-	go func() {
-		seq, err := f.congestion.Sequence(ctx)
-		if err != nil {
-			return
-		}
-
-		for !f.isAlive {
-			if ctx.Err() != nil {
-				return
-			}
-
-			p := Packet{
-				ack:  0,
-				nack: 0,
-				seq:  seq,
-				data: proxy.SimplePacket(nil),
-			}
-
-			_ = f.sendPacket(p)
-			time.Sleep(1 * time.Second)
-		}
-	}()
-
-	go func() {
-		_, _ = f.produceInternal(ctx, false)
-	}()
-	go f.earlyUpdateLoop(ctx, f.keepalive)
-
-	if err := f.readQueuePacket(ctx, conn); err != nil {
-		return err
-	}
-
-	f.isAlive = true
-	f.startup = false
-
-	go func() {
-		lockedAccept := func() {
-			f.mu.RLock()
-			defer f.mu.RUnlock()
-
-			if err := f.readQueuePacket(ctx, conn); err != nil {
-				log.Println(err)
-			}
-		}
-
-		for f.isAlive {
-			log.Println("alive and listening for packets")
-			lockedAccept()
-		}
-		log.Println("no longer alive")
-	}()
-
-	return nil
-}
-
-func (f *InitiatedFlow) Consume(ctx context.Context, p proxy.Packet) error {
-	f.mu.RLock()
-	defer f.mu.RUnlock()
-
-	return f.Flow.Consume(ctx, p)
-}
-
-func (f *InitiatedFlow) Produce(ctx context.Context) (proxy.Packet, error) {
-	f.mu.RLock()
-	defer f.mu.RUnlock()
-
-	return f.Flow.Produce(ctx)
-}
-
 func (f *Flow) IsAlive() bool {
 	return f.isAlive
 }
@@ -265,7 +136,7 @@ func (f *Flow) queueDatagram(ctx context.Context, p []byte) error {
 	}
 }
 
-func (f *Flow) sendPacket(p Packet) error {
+func (f *Flow) sendPacket(p proxy.Packet) error {
 	b := p.Marshal()
 
 	for _, g := range f.generators {
@@ -302,13 +173,24 @@ func (f *Flow) earlyUpdateLoop(ctx context.Context, keepalive time.Duration) {
 	}
 }
 
-func (f *Flow) readQueuePacket(ctx context.Context, c PacketConn) error {
+func (f *Flow) readPacket(ctx context.Context, c PacketConn) ([]byte, error) {
-	// TODO: Replace 6000 with MTU+header size
 	buf := make([]byte, 6000)
-	n, _, err := c.ReadFromUDP(buf)
+
-	if err != nil {
+	if d, ok := ctx.Deadline(); ok {
-		return err
+		if err := c.SetReadDeadline(d); err != nil {
+			return nil, err
+		}
 	}
 
-	return f.queueDatagram(ctx, buf[:n])
+	n, _, err := c.ReadFromUDP(buf)
+	if err != nil {
+		if err, ok := err.(net.Error); ok && err.Timeout() {
+			if ctx.Err() != nil {
+				return nil, ctx.Err()
+			}
+		}
+		return nil, err
+	}
+
+	return buf[:n], nil
 }

udp/flow_test.go

@@ -107,7 +107,9 @@ func TestFlow_Produce(t *testing.T) {
 			flowA.isAlive = true
 
 			go func() {
-				err := flowA.readQueuePacket(context.Background(), testConn.SideB())
+				p, err := flowA.readPacket(context.Background(), testConn.SideB())
+				assert.Nil(t, err)
+				err = flowA.queueDatagram(context.Background(), p)
 				assert.Nil(t, err)
 			}()
 			p, err := flowA.Produce(context.Background())
@@ -143,7 +145,9 @@ func TestFlow_Produce(t *testing.T) {
 			flowA.isAlive = true
 
 			go func() {
-				err := flowA.readQueuePacket(context.Background(), testConn.SideB())
+				p, err := flowA.readPacket(context.Background(), testConn.SideB())
+				assert.Nil(t, err)
+				err = flowA.queueDatagram(context.Background(), p)
 				assert.Nil(t, err)
 			}()
 			p, err := flowA.Produce(context.Background())

udp/inbound_flow.go

@@ -0,0 +1,142 @@
+package udp
+
+import (
+	"context"
+	"log"
+	"mpbl3p/proxy"
+	"sync"
+	"time"
+)
+
+type InboundFlow struct {
+	inboundDatagrams chan []byte
+
+	mu sync.RWMutex
+	Flow
+}
+
+func newInboundFlow(f Flow) (*InboundFlow, error) {
+	fi := InboundFlow{
+		inboundDatagrams: make(chan []byte),
+		Flow:             f,
+	}
+
+	return &fi, nil
+}
+
+func (f *InboundFlow) queueDatagram(ctx context.Context, p []byte) error {
+	select {
+	case f.inboundDatagrams <- p:
+		return nil
+	case <-ctx.Done():
+		return ctx.Err()
+	}
+}
+
+func (f *InboundFlow) processPackets(ctx context.Context) {
+	for {
+		f.mu.Lock()
+
+		var err error
+		for once := true; err != nil || once; once = false {
+			if ctx.Err() != nil {
+				return
+			}
+
+			err = f.handleExchanges(ctx)
+			if err != nil {
+				log.Println(err)
+			}
+		}
+
+		f.mu.Unlock()
+
+		var p []byte
+		select {
+		case p = <-f.inboundDatagrams:
+		case <-ctx.Done():
+			return
+		}
+
+		// TODO: Check if p means redo exchanges
+		if false {
+			continue
+		}
+
+		select {
+		case f.Flow.inboundDatagrams <- p:
+		case <-ctx.Done():
+			return
+		}
+	}
+}
+
+func (f *InboundFlow) handleExchanges(ctx context.Context) error {
+	var exchanges []proxy.Exchange
+
+	if e, ok := f.congestion.(proxy.Exchange); ok {
+		exchanges = append(exchanges, e)
+	}
+
+	var exchangeData [][]byte
+
+	for _, e := range exchanges {
+		for once := true; !e.Complete() || once; once = false {
+			if err := func() (err error) {
+				ctx, cancel := context.WithTimeout(ctx, 5*time.Second)
+				defer cancel()
+
+				var recv []byte
+				select {
+				case recv = <-f.inboundDatagrams:
+				case <-ctx.Done():
+					return ctx.Err()
+				}
+
+				for i := range f.verifiers {
+					v := f.verifiers[len(f.verifiers)-i-1]
+
+					recv, err = proxy.StripMac(recv, v)
+					if err != nil {
+						return err
+					}
+				}
+
+				var resp, data []byte
+				if resp, data, err = e.Handle(ctx, recv); err != nil {
+					return err
+				}
+
+				if data != nil {
+					exchangeData = append(exchangeData, data)
+				}
+
+				if resp != nil {
+					if err = f.sendPacket(proxy.SimplePacket(resp)); err != nil {
+						return err
+					}
+				}
+
+				return nil
+			}(); err != nil {
+				return err
+			}
+		}
+	}
+
+	return nil
+}
+
+func (f *InboundFlow) Consume(ctx context.Context, p proxy.Packet) error {
+	f.mu.RLock()
+	defer f.mu.RUnlock()
+
+	return f.Flow.Consume(ctx, p)
+}
+
+func (f *InboundFlow) Produce(ctx context.Context) (proxy.Packet, error) {
+	f.mu.RLock()
+	defer f.mu.RUnlock()
+
+	return f.Flow.Produce(ctx)
+}

udp/listener.go

@@ -16,9 +16,7 @@ type ComparableUdpAddress struct {
 
 func fromUdpAddress(address net.UDPAddr) ComparableUdpAddress {
 	var ip [16]byte
-	for i, b := range []byte(address.IP) {
+	copy(ip[:], address.IP)
-		ip[i] = b
-	}
 
 	return ComparableUdpAddress{
 		IP:   ip,
@@ -47,12 +45,7 @@ func NewListener(
 		return err
 	}
 
-	err = pconn.SetWriteBuffer(0)
+	receivedConnections := make(map[ComparableUdpAddress]*InboundFlow)
-	if err != nil {
-		panic(err)
-	}
-
-	receivedConnections := make(map[ComparableUdpAddress]*Flow)
 
 	go func() {
 		for ctx.Err() == nil {
@@ -70,10 +63,11 @@ func NewListener(
 			}
 
 			raddr := fromUdpAddress(*addr)
-			if f, exists := receivedConnections[raddr]; exists {
+			if fi, exists := receivedConnections[raddr]; exists {
 				log.Println("existing flow. queuing...")
-				if err := f.queueDatagram(ctx, buf[:n]); err != nil {
+				if err := fi.queueDatagram(ctx, buf[:n]); err != nil {
-
+					log.Println("error")
+					continue
 				}
 				log.Println("queued")
 				continue
@@ -94,21 +88,28 @@ func NewListener(
 			f.raddr = addr
 			f.isAlive = true
 
+			fi, err := newInboundFlow(f)
+			if err != nil {
+				log.Println(err)
+				continue
+			}
+
 			log.Printf("received new udp connection: %v\n", f)
 
-			go f.earlyUpdateLoop(ctx, 0)
+			go fi.processPackets(ctx)
+			go fi.earlyUpdateLoop(ctx, 0)
 
-			receivedConnections[raddr] = &f
+			receivedConnections[raddr] = fi
 
 			if enableConsumers {
-				p.AddConsumer(ctx, &f)
+				p.AddConsumer(ctx, fi)
 			}
 			if enableProducers {
-				p.AddProducer(ctx, &f)
+				p.AddProducer(ctx, fi)
 			}
 
 			log.Println("handling...")
-			if err := f.queueDatagram(ctx, buf[:n]); err != nil {
+			if err := fi.queueDatagram(ctx, buf[:n]); err != nil {
 				return
 			}
 			log.Println("handled")

udp/outbound_flow.go

@@ -0,0 +1,181 @@
+package udp
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"log"
+	"mpbl3p/proxy"
+	"net"
+	"sync"
+	"time"
+)
+
+type OutboundFlow struct {
+	Local  func() string
+	Remote string
+
+	g         proxy.MacGenerator
+	keepalive time.Duration
+
+	mu sync.RWMutex
+	Flow
+}
+
+func InitiateFlow(
+	local func() string,
+	remote string,
+	vs []proxy.MacVerifier,
+	gs []proxy.MacGenerator,
+	c Congestion,
+	keepalive time.Duration,
+) (*OutboundFlow, error) {
+	f := OutboundFlow{
+		Local:     local,
+		Remote:    remote,
+		Flow:      newFlow(c, vs, gs),
+		keepalive: keepalive,
+	}
+
+	return &f, nil
+}
+
+func (f *OutboundFlow) String() string {
+	return fmt.Sprintf("UdpOutbound{%v -> %v}", f.Local(), f.Remote)
+}
+
+func (f *OutboundFlow) Reconnect(ctx context.Context) error {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+
+	if f.isAlive {
+		return nil
+	}
+
+	localAddr, err := net.ResolveUDPAddr("udp", f.Local())
+	if err != nil {
+		return err
+	}
+
+	remoteAddr, err := net.ResolveUDPAddr("udp", f.Remote)
+	if err != nil {
+		return err
+	}
+
+	conn, err := net.DialUDP("udp", localAddr, remoteAddr)
+	if err != nil {
+		return err
+	}
+
+	f.writer = conn
+
+	// prod the connection once a second until we get an ack, then consider it alive
+	var exchanges []proxy.Exchange
+
+	if e, ok := f.congestion.(proxy.Exchange); ok {
+		exchanges = append(exchanges, e)
+	}
+
+	var exchangeData [][]byte
+
+	for _, e := range exchanges {
+		i, err := e.Initial(ctx)
+		if err != nil {
+			return err
+		}
+
+		if err = f.sendPacket(proxy.SimplePacket(i)); err != nil {
+			return err
+		}
+
+		for once := true; !e.Complete() || once; once = false {
+			if err := func() error {
+				ctx, cancel := context.WithTimeout(ctx, 5*time.Second)
+				defer cancel()
+
+				var recv []byte
+				if recv, err = f.readPacket(ctx, conn); err != nil {
+					return err
+				}
+
+				for i := range f.verifiers {
+					v := f.verifiers[len(f.verifiers)-i-1]
+
+					recv, err = proxy.StripMac(recv, v)
+					if err != nil {
+						return err
+					}
+				}
+
+				var resp, data []byte
+				if resp, data, err = e.Handle(ctx, recv); err != nil {
+					return err
+				}
+
+				if data != nil {
+					exchangeData = append(exchangeData, data)
+				}
+
+				if resp != nil {
+					if err = f.sendPacket(proxy.SimplePacket(resp)); err != nil {
+						return err
+					}
+				}
+
+				return nil
+			}(); err != nil {
+				return err
+			}
+		}
+	}
+
+	go func() {
+		for _, d := range exchangeData {
+			if err := f.queueDatagram(ctx, d); err != nil {
+				return
+			}
+		}
+
+		lockedAccept := func() {
+			f.mu.RLock()
+			defer f.mu.RUnlock()
+
+			var p []byte
+			if p, err = f.readPacket(ctx, conn); err != nil {
+				if errors.Is(err, context.DeadlineExceeded) || errors.Is(err, context.Canceled) {
+					return
+				}
+				log.Println(err)
+				return
+			}
+
+			if err := f.queueDatagram(ctx, p); err != nil {
+				return
+			}
+
+		}
+
+		for f.isAlive {
+			log.Println("alive and listening for packets")
+			lockedAccept()
+		}
+		log.Println("no longer alive")
+	}()
+
+	f.isAlive = true
+	return nil
+}
+
+func (f *OutboundFlow) Consume(ctx context.Context, p proxy.Packet) error {
+	f.mu.RLock()
+	defer f.mu.RUnlock()
+
+	return f.Flow.Consume(ctx, p)
+}
+
+func (f *OutboundFlow) Produce(ctx context.Context) (proxy.Packet, error) {
+	f.mu.RLock()
+	defer f.mu.RUnlock()
+
+	return f.Flow.Produce(ctx)
+}