More reliable QUIC session handling

build/gobind/monolith.go

@@ -98,7 +98,7 @@ func (m *DendriteMonolith) Start() {
 	cfg.Database.SyncAPI = config.DataSource(fmt.Sprintf("file:%s/dendrite-syncapi.db", m.StorageDirectory))
 	cfg.Database.RoomServer = config.DataSource(fmt.Sprintf("file:%s/dendrite-roomserver.db", m.StorageDirectory))
 	cfg.Database.ServerKey = config.DataSource(fmt.Sprintf("file:%s/dendrite-serverkey.db", m.StorageDirectory))
-	cfg.Database.E2EKey = config.DataSource(fmt.Sprintf("file:%s/dendrite-e2ekey.db", m.StorageDirectory))
+	cfg.Database.E2EKey = config.DataSource(fmt.Sprintf("file:%s/dendrite-keyserver.db", m.StorageDirectory))
 	cfg.Database.FederationSender = config.DataSource(fmt.Sprintf("file:%s/dendrite-federationsender.db", m.StorageDirectory))
 	cfg.Database.AppService = config.DataSource(fmt.Sprintf("file:%s/dendrite-appservice.db", m.StorageDirectory))
 	cfg.Database.CurrentState = config.DataSource(fmt.Sprintf("file:%s/dendrite-currentstate.db", m.StorageDirectory))

cmd/dendrite-demo-yggdrasil/main.go

@@ -83,7 +83,7 @@ func main() {
 	cfg.Database.SyncAPI = config.DataSource(fmt.Sprintf("file:%s-syncapi.db", *instanceName))
 	cfg.Database.RoomServer = config.DataSource(fmt.Sprintf("file:%s-roomserver.db", *instanceName))
 	cfg.Database.ServerKey = config.DataSource(fmt.Sprintf("file:%s-serverkey.db", *instanceName))
-	cfg.Database.E2EKey = config.DataSource(fmt.Sprintf("file:%s-e2ekey.db", *instanceName))
+	cfg.Database.E2EKey = config.DataSource(fmt.Sprintf("file:%s-keyserver.db", *instanceName))
 	cfg.Database.FederationSender = config.DataSource(fmt.Sprintf("file:%s-federationsender.db", *instanceName))
 	cfg.Database.AppService = config.DataSource(fmt.Sprintf("file:%s-appservice.db", *instanceName))
 	cfg.Database.CurrentState = config.DataSource(fmt.Sprintf("file:%s-currentstate.db", *instanceName))

cmd/dendrite-demo-yggdrasil/yggconn/node.go

@@ -50,6 +50,7 @@ type Node struct {
 	tlsConfig  *tls.Config
 	quicConfig *quic.Config
 	sessions   sync.Map // string -> quic.Session
+	coords     sync.Map // string -> yggdrasil.Coords
 	incoming   chan QUICStream
 	NewSession func(remote gomatrixserverlib.ServerName)
 }

cmd/dendrite-demo-yggdrasil/yggconn/session.go

@@ -31,6 +31,7 @@ import (
 
 	"github.com/lucas-clemente/quic-go"
 	"github.com/yggdrasil-network/yggdrasil-go/src/crypto"
+	"github.com/yggdrasil-network/yggdrasil-go/src/yggdrasil"
 )
 
 func (n *Node) listenFromYgg() {
@@ -95,10 +96,34 @@ func (n *Node) Dial(network, address string) (net.Conn, error) {
 }
 
 // Implements http.Transport.DialContext
+// nolint:gocyclo
 func (n *Node) DialContext(ctx context.Context, network, address string) (net.Conn, error) {
 	s, ok1 := n.sessions.Load(address)
 	session, ok2 := s.(quic.Session)
-	if !ok1 || !ok2 || (ok1 && ok2 && session.ConnectionState().HandshakeComplete) {
+	if !ok1 || !ok2 {
+		// First of all, check if we think we know the coords of this
+		// node. If we do then we'll try to dial to it directly. This
+		// will either succeed or fail.
+		if v, ok := n.coords.Load(address); ok {
+			coords, ok := v.(yggdrasil.Coords)
+			if !ok {
+				return nil, errors.New("should have found yggdrasil.Coords but didn't")
+			}
+			n.log.Infof("Coords %s for %q cached, trying to dial", coords.String(), address)
+			var err error
+			// We think we know the coords. Try to dial the node.
+			if session, err = n.tryDial(address, coords); err != nil {
+				// We thought we knew the coords but it didn't result
+				// in a successful dial. Nuke them from the cache.
+				n.coords.Delete(address)
+				n.log.Infof("Cached coords %s for %q failed", coords.String(), address)
+			}
+		}
+
+		// We either don't know the coords for the node, or we failed
+		// to dial it before, in which case try to resolve the coords.
+		if _, ok := n.coords.Load(address); !ok {
+			n.log.Infof("Searching for coords for %q", address)
 			dest, err := hex.DecodeString(address)
 			if err != nil {
 				return nil, err
@@ -120,9 +145,31 @@ func (n *Node) DialContext(ctx context.Context, network, address string) (net.Co
 				return nil, fmt.Errorf("n.core.Resolve: %w", err)
 			}
 			fmt.Println("Found coords:", coords)
-		fmt.Println("Dialling")
+			n.coords.Store(address, coords)
 
-		session, err = quic.Dial(
+			// We now know the coords in theory. Let's try dialling the
+			// node again.
+			if session, err = n.tryDial(address, coords); err != nil {
+				return nil, fmt.Errorf("n.tryDial: %w", err)
+			}
+		}
+	}
+
+	if session == nil {
+		return nil, fmt.Errorf("should have found session but didn't")
+	}
+
+	st, err := session.OpenStream()
+	if err != nil {
+		n.log.Println("session.OpenStream:", err)
+		_ = session.CloseWithError(0, "expected to be able to open session")
+		return nil, err
+	}
+	return QUICStream{st, session}, nil
+}
+
+func (n *Node) tryDial(address string, coords yggdrasil.Coords) (quic.Session, error) {
+	session, err := quic.Dial(
 		n.core,       // yggdrasil.PacketConn
 		coords,       // dial address
 		address,      // dial SNI
@@ -130,18 +177,18 @@ func (n *Node) DialContext(ctx context.Context, network, address string) (net.Co
 		n.quicConfig, // QUIC config
 	)
 	if err != nil {
-			n.log.Println("n.dialer.DialContext:", err)
 		return nil, err
 	}
-		fmt.Println("Dial OK")
+	if len(session.ConnectionState().PeerCertificates) != 1 {
+		_ = session.CloseWithError(0, "expected a peer certificate")
+		return nil, errors.New("didn't receive a peer certificate")
+	}
+	if gotAddress := session.ConnectionState().PeerCertificates[0].DNSNames[0]; address != gotAddress {
+		_ = session.CloseWithError(0, "you aren't the host I was hoping for")
+		return nil, fmt.Errorf("expected %q but dialled %q", address, gotAddress)
+	}
 	go n.listenFromQUIC(session, address)
-	}
-	st, err := session.OpenStream()
-	if err != nil {
-		n.log.Println("session.OpenStream:", err)
-		return nil, err
-	}
-	return QUICStream{st, session}, nil
+	return session, nil
 }
 
 func (n *Node) generateTLSConfig() *tls.Config {