feat: updated gateway to quic

This commit is contained in:
=
2025-03-04 22:10:38 +05:30
parent f376eaae13
commit 9bfc2a5dd2
10 changed files with 610 additions and 280 deletions

View File

@@ -3,20 +3,49 @@ package gateway
import (
"bufio"
"bytes"
"context"
"errors"
"io"
"net"
"strings"
"sync"
"github.com/quic-go/quic-go"
"github.com/rs/zerolog/log"
)
func handleConnection(conn net.Conn) {
defer conn.Close()
log.Info().Msgf("New connection from: %s", conn.RemoteAddr().String())
func handleConnection(ctx context.Context, quicConn quic.Connection) {
log.Info().Msgf("New connection from: %s", quicConn.RemoteAddr().String())
// Use WaitGroup to track all streams
var wg sync.WaitGroup
for {
// Accept the first stream, which we'll use for commands
stream, err := quicConn.AcceptStream(ctx)
if err != nil {
log.Printf("Failed to accept QUIC stream: %v", err)
break
}
wg.Add(1)
go func(stream quic.Stream) {
defer wg.Done()
defer stream.Close()
handleStream(stream, quicConn)
}(stream)
}
wg.Wait()
log.Printf("All streams closed for connection: %s", quicConn.RemoteAddr().String())
}
func handleStream(stream quic.Stream, quicConn quic.Connection) {
streamID := stream.StreamID()
log.Printf("New stream %d from: %s", streamID, quicConn.RemoteAddr().String())
// Use buffered reader for better handling of fragmented data
reader := bufio.NewReader(conn)
reader := bufio.NewReader(stream)
defer stream.Close()
for {
msg, err := reader.ReadBytes('\n')
if err != nil {
@@ -32,6 +61,7 @@ func handleConnection(conn net.Conn) {
switch string(cmd) {
case "FORWARD-TCP":
log.Info().Msg("Starting secure connector proxy...")
proxyAddress := string(bytes.Split(args, []byte(" "))[0])
destTarget, err := net.Dial("tcp", proxyAddress)
if err != nil {
@@ -56,10 +86,10 @@ func handleConnection(conn net.Conn) {
}
}
CopyData(conn, destTarget)
CopyDataFromQuicToTcp(stream, destTarget)
return
case "PING":
if _, err := conn.Write([]byte("PONG")); err != nil {
if _, err := stream.Write([]byte("PONG\n")); err != nil {
log.Error().Msgf("Error writing PONG response: %v", err)
}
return
@@ -74,34 +104,38 @@ type CloseWrite interface {
CloseWrite() error
}
func CopyData(src, dst net.Conn) {
func CopyDataFromQuicToTcp(quicStream quic.Stream, tcpConn net.Conn) {
// Create a WaitGroup to wait for both copy operations
var wg sync.WaitGroup
wg.Add(2)
copyAndClose := func(dst, src net.Conn, done chan<- bool) {
// Start copying from QUIC stream to TCP
go func() {
defer wg.Done()
_, err := io.Copy(dst, src)
if err != nil && !errors.Is(err, io.EOF) {
log.Error().Msgf("Copy error: %v", err)
if _, err := io.Copy(tcpConn, quicStream); err != nil {
log.Error().Msgf("Error copying quic->postgres: %v", err)
}
// Signal we're done writing
done <- true
// Half close the connection if possible
if c, ok := dst.(CloseWrite); ok {
c.CloseWrite()
if e, ok := tcpConn.(CloseWrite); ok {
log.Debug().Msg("Closing TCP write end")
e.CloseWrite()
} else {
log.Debug().Msg("TCP connection does not support CloseWrite")
}
}
}()
done1 := make(chan bool, 1)
done2 := make(chan bool, 1)
go copyAndClose(dst, src, done1)
go copyAndClose(src, dst, done2)
// Start copying from TCP to QUIC stream
go func() {
defer wg.Done()
if _, err := io.Copy(quicStream, tcpConn); err != nil {
log.Debug().Msgf("Error copying postgres->quic: %v", err)
}
// Close the write side of the QUIC stream
if err := quicStream.Close(); err != nil && !strings.Contains(err.Error(), "close called for canceled stream") {
log.Error().Msgf("Error closing QUIC stream write: %v", err)
}
}()
// Wait for both copies to complete
<-done1
<-done2
wg.Wait()
}

View File

@@ -15,6 +15,8 @@ import (
"github.com/pion/logging"
"github.com/pion/turn/v4"
"github.com/rs/zerolog/log"
"github.com/quic-go/quic-go"
)
type GatewayConfig struct {
@@ -56,13 +58,12 @@ func (g *Gateway) ConnectWithRelay() error {
if relayPort == "5349" {
log.Info().Msgf("Provided relay port %s. Using TLS", relayPort)
conn, err = tls.Dial("tcp", relayDetails.TurnServerAddress, &tls.Config{
InsecureSkipVerify: false,
ServerName: relayAddress,
ServerName: relayAddress,
})
} else {
log.Info().Msgf("Provided relay port %s. Using non TLS connection.", relayPort)
peerAddr, err := net.ResolveTCPAddr("tcp", relayDetails.TurnServerAddress)
if err != nil {
peerAddr, errPeer := net.ResolveTCPAddr("tcp", relayDetails.TurnServerAddress)
if errPeer != nil {
return fmt.Errorf("Failed to parse turn server address: %w", err)
}
conn, err = net.DialTCP("tcp", nil, peerAddr)
@@ -116,20 +117,20 @@ func (g *Gateway) Listen(ctx context.Context) error {
// Allocate a relay socket on the TURN server. On success, it
// will return a net.PacketConn which represents the remote
// socket.
relayNonTlsConn, err := g.client.AllocateTCP()
relayUdpConnection, err := g.client.Allocate()
if err != nil {
return fmt.Errorf("Failed to allocate relay connection: %w", err)
}
log.Info().Msg(relayNonTlsConn.Addr().String())
log.Info().Msg(relayUdpConnection.LocalAddr().String())
defer func() {
if closeErr := relayNonTlsConn.Close(); closeErr != nil {
if closeErr := relayUdpConnection.Close(); closeErr != nil {
log.Error().Msgf("Failed to close connection: %s", closeErr)
}
}()
gatewayCert, err := api.CallExchangeRelayCertV1(g.httpClient, api.ExchangeRelayCertRequestV1{
RelayAddress: relayNonTlsConn.Addr().String(),
RelayAddress: relayUdpConnection.LocalAddr().String(),
})
if err != nil {
return err
@@ -144,45 +145,58 @@ func (g *Gateway) Listen(ctx context.Context) error {
if g.config.InfisicalStaticIp != "" {
log.Info().Msgf("Found static ip from Infisical: %s. Creating permission IP lifecycle", g.config.InfisicalStaticIp)
peerAddr, err := net.ResolveTCPAddr("tcp", g.config.InfisicalStaticIp)
peerAddr, err := net.ResolveUDPAddr("udp", g.config.InfisicalStaticIp)
if err != nil {
return fmt.Errorf("Failed to parse infisical static ip: %w", err)
}
g.registerPermissionLifecycle(func() error {
err := relayNonTlsConn.CreatePermissions(peerAddr)
return err
}, shutdownCh)
err = g.client.CreatePermission(peerAddr)
if err != nil {
return fmt.Errorf("Failed to set permission: %w", err)
}
}
cert, err := tls.X509KeyPair([]byte(gatewayCert.Certificate), []byte(gatewayCert.PrivateKey))
if err != nil {
return fmt.Errorf("failed to parse cert: %s", err)
return fmt.Errorf("failed to parse cert: %w", err)
}
caCertPool := x509.NewCertPool()
caCertPool.AppendCertsFromPEM([]byte(gatewayCert.CertificateChain))
relayConn := tls.NewListener(relayNonTlsConn, &tls.Config{
// Setup QUIC server
tlsConfig := &tls.Config{
Certificates: []tls.Certificate{cert},
MinVersion: tls.VersionTLS12,
ClientCAs: caCertPool,
ClientAuth: tls.RequireAndVerifyClientCert,
})
NextProtos: []string{"infisical-gateway"},
}
// Setup QUIC listener on the relayConn
quicConfig := &quic.Config{
EnableDatagrams: true,
MaxIdleTimeout: 30 * time.Second,
KeepAlivePeriod: 15 * time.Second,
}
quicListener, err := quic.Listen(relayUdpConnection, tlsConfig, quicConfig)
if err != nil {
return fmt.Errorf("Failed to listen for QUIC: %w", err)
}
defer quicListener.Close()
log.Printf("Listener started on %s", quicListener.Addr())
errCh := make(chan error, 1)
log.Info().Msg("Gateway started successfully")
g.registerHeartBeat(errCh, shutdownCh)
g.registerRelayIsActive(relayNonTlsConn.Addr().String(), errCh, shutdownCh)
g.registerRelayIsActive(relayUdpConnection.LocalAddr().String(), tlsConfig, quicConfig, errCh, shutdownCh)
// Create a WaitGroup to track active connections
var wg sync.WaitGroup
go func() {
for {
if relayDeadlineConn, ok := relayConn.(*net.TCPListener); ok {
relayDeadlineConn.SetDeadline(time.Now().Add(1 * time.Second))
}
select {
case <-ctx.Done():
return
@@ -190,67 +204,42 @@ func (g *Gateway) Listen(ctx context.Context) error {
return
default:
// Accept new relay connection
conn, err := relayConn.Accept()
quicConn, err := quicListener.Accept(context.Background())
if err != nil {
// Check if it's a timeout error (which we expect due to our deadline)
if netErr, ok := err.(net.Error); ok && netErr.Timeout() {
continue
}
if !strings.Contains(err.Error(), "data contains incomplete STUN or TURN frame") {
log.Error().Msgf("Failed to accept connection: %v", err)
}
log.Printf("Failed to accept QUIC connection: %v", err)
continue
}
tlsConn, ok := conn.(*tls.Conn)
if !ok {
log.Error().Msg("Failed to convert to TLS connection")
conn.Close()
continue
}
// Set a deadline for the handshake to prevent hanging
tlsConn.SetDeadline(time.Now().Add(10 * time.Second))
err = tlsConn.Handshake()
// Clear the deadline after handshake
tlsConn.SetDeadline(time.Time{})
if err != nil {
log.Error().Msgf("TLS handshake failed: %v", err)
conn.Close()
continue
}
// Get connection state which contains certificate information
state := tlsConn.ConnectionState()
if len(state.PeerCertificates) > 0 {
organizationUnit := state.PeerCertificates[0].Subject.OrganizationalUnit
commonName := state.PeerCertificates[0].Subject.CommonName
tlsState := quicConn.ConnectionState().TLS
if len(tlsState.PeerCertificates) > 0 {
organizationUnit := tlsState.PeerCertificates[0].Subject.OrganizationalUnit
commonName := tlsState.PeerCertificates[0].Subject.CommonName
if organizationUnit[0] != "gateway-client" || commonName != "cloud" {
log.Error().Msgf("Client certificate verification failed. Received %s, %s", organizationUnit, commonName)
conn.Close()
errMsg := fmt.Sprintf("Client certificate verification failed. Received %s, %s", organizationUnit, commonName)
log.Error().Msg(errMsg)
quicConn.CloseWithError(1, errMsg)
continue
}
}
// Handle the connection in a goroutine
wg.Add(1)
go func(c net.Conn) {
go func(c quic.Connection) {
defer wg.Done()
defer c.Close()
defer c.CloseWithError(0, "connection closed")
// Monitor parent context to close this connection when needed
go func() {
select {
case <-ctx.Done():
c.Close() // Force close connection when context is canceled
c.CloseWithError(0, "connection closed") // Force close connection when context is canceled
case <-shutdownCh:
c.Close() // Force close connection when accepting loop is done
c.CloseWithError(0, "connection closed") // Force close connection when accepting loop is done
}
}()
handleConnection(c)
}(conn)
handleConnection(ctx, c)
}(quicConn)
}
}
}()
@@ -282,7 +271,7 @@ func (g *Gateway) Listen(ctx context.Context) error {
}
func (g *Gateway) registerHeartBeat(errCh chan error, done chan bool) {
ticker := time.NewTicker(1 * time.Hour)
ticker := time.NewTicker(30 * time.Minute)
go func() {
time.Sleep(10 * time.Second)
@@ -306,26 +295,7 @@ func (g *Gateway) registerHeartBeat(errCh chan error, done chan bool) {
}()
}
func (g *Gateway) registerPermissionLifecycle(permissionFn func() error, done chan bool) {
ticker := time.NewTicker(3 * time.Minute)
go func() {
// wait for 5 mins
permissionFn()
log.Printf("Created permission for incoming connections")
for {
select {
case <-done:
ticker.Stop()
return
case <-ticker.C:
permissionFn()
}
}
}()
}
func (g *Gateway) registerRelayIsActive(serverAddr string, errCh chan error, done chan bool) {
func (g *Gateway) registerRelayIsActive(serverAddr string, tlsConf *tls.Config, quicConf *quic.Config, errCh chan error, done chan bool) {
ticker := time.NewTicker(10 * time.Second)
go func() {
@@ -336,14 +306,19 @@ func (g *Gateway) registerRelayIsActive(serverAddr string, errCh chan error, don
ticker.Stop()
return
case <-ticker.C:
conn, err := net.Dial("tcp", serverAddr)
if err != nil {
errCh <- err
return
}
if conn != nil {
conn.Close()
}
func() {
ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second) // 3s handshake timeout
defer cancel()
conn, err := quic.DialAddr(ctx, serverAddr, tlsConf, quicConf)
if conn != nil {
conn.CloseWithError(0, "connection closed")
}
// this error means quic connection is alive
if err != nil && !strings.Contains(err.Error(), "tls: failed to verify certificate") {
errCh <- err
return
}
}()
}
}
}()

View File

@@ -0,0 +1,186 @@
package gateway
import (
"context"
"crypto/tls"
"errors"
"fmt"
"net"
"os"
"os/signal"
"runtime"
"strconv"
"syscall"
"github.com/pion/logging"
"github.com/pion/turn/v4"
"github.com/rs/zerolog/log"
"golang.org/x/sys/unix"
"gopkg.in/yaml.v2"
)
var (
errMissingTlsCert = errors.New("Missing TLS files")
)
type GatewayRelay struct {
Config *GatewayRelayConfig
}
type GatewayRelayConfig struct {
PublicIP string `json:"public_ip"`
Port int `json:"port"`
Realm string `json:"realm"`
AuthSecret string `json:"auth_secret"`
RelayMinPort uint16 `json:"relay_min_port"`
RelayMaxPort uint16 `json:"relay_max_port"`
TlsCertPath string `json:"tls_cert_path"`
TlsPrivateKeyPath string `json:"tls_private_key_path"`
tls tls.Certificate
isTlsEnabled bool
}
func NewGatewayRelay(configFilePath string) (*GatewayRelay, error) {
cfgFile, err := os.ReadFile(configFilePath)
if err != nil {
return nil, err
}
var cfg GatewayRelayConfig
if err := yaml.Unmarshal(cfgFile, &cfg); err != nil {
return nil, err
}
if cfg.PublicIP == "" {
return nil, fmt.Errorf("Missing public ip")
}
if cfg.AuthSecret == "" {
return nil, fmt.Errorf("Missing auth secret")
}
if cfg.Realm == "" {
cfg.Realm = "infisical.org"
}
if cfg.RelayMinPort == 0 {
cfg.RelayMinPort = 49152
}
if cfg.RelayMaxPort == 0 {
cfg.RelayMaxPort = 65535
}
if cfg.Port == 0 {
cfg.Port = 3478
} else if cfg.Port == 5349 {
if cfg.TlsCertPath == "" || cfg.TlsPrivateKeyPath == "" {
return nil, errMissingTlsCert
}
tlsCertFile, err := os.ReadFile(cfg.TlsCertPath)
if err != nil {
return nil, err
}
tlsPrivateKeyFile, err := os.ReadFile(cfg.TlsPrivateKeyPath)
if err != nil {
return nil, err
}
cert, err := tls.LoadX509KeyPair(string(tlsCertFile), string(tlsPrivateKeyFile))
if err != nil {
return nil, err
}
cfg.tls = cert
cfg.isTlsEnabled = true
}
return &GatewayRelay{
Config: &cfg,
}, nil
}
func (g *GatewayRelay) Run() error {
addr, err := net.ResolveTCPAddr("tcp", "0.0.0.0:"+strconv.Itoa(g.Config.Port))
if err != nil {
return fmt.Errorf("Failed to parse server address: %s", err)
}
// NewLongTermAuthHandler takes a pion.LeveledLogger. This allows you to intercept messages
// and process them yourself.
logger := logging.NewDefaultLeveledLoggerForScope("lt-creds", logging.LogLevelTrace, os.Stdout)
// Create `numThreads` UDP listeners to pass into pion/turn
// pion/turn itself doesn't allocate any UDP sockets, but lets the user pass them in
// this allows us to add logging, storage or modify inbound/outbound traffic
// UDP listeners share the same local address:port with setting SO_REUSEPORT and the kernel
// will load-balance received packets per the IP 5-tuple
listenerConfig := &net.ListenConfig{
Control: func(network, address string, conn syscall.RawConn) error { // nolint: revive
var operr error
if err = conn.Control(func(fd uintptr) {
operr = syscall.SetsockoptInt(int(fd), syscall.SOL_SOCKET, unix.SO_REUSEPORT, 1)
}); err != nil {
return err
}
return operr
},
}
publicIP := g.Config.PublicIP
relayAddressGenerator := &turn.RelayAddressGeneratorPortRange{
RelayAddress: net.ParseIP(publicIP), // Claim that we are listening on IP passed by user
Address: "0.0.0.0", // But actually be listening on every interface
MinPort: g.Config.RelayMinPort,
MaxPort: g.Config.RelayMaxPort,
}
threadNum := runtime.NumCPU()
listenerConfigs := make([]turn.ListenerConfig, threadNum)
for i := 0; i < threadNum; i++ {
conn, listErr := listenerConfig.Listen(context.Background(), addr.Network(), addr.String())
if listErr != nil {
return fmt.Errorf("Failed to allocate TCP listener at %s:%s %s", addr.Network(), addr.String(), listErr)
}
listenerConfigs[i] = turn.ListenerConfig{
RelayAddressGenerator: relayAddressGenerator,
}
if g.Config.isTlsEnabled {
listenerConfigs[i].Listener = tls.NewListener(conn, &tls.Config{
Certificates: []tls.Certificate{g.Config.tls},
})
} else {
listenerConfigs[i].Listener = conn
}
log.Printf("Server %d listening on %s\n", i, conn.Addr().String())
}
loggerF := logging.NewDefaultLoggerFactory()
loggerF.DefaultLogLevel = logging.LogLevelDebug
server, err := turn.NewServer(turn.ServerConfig{
Realm: g.Config.Realm,
AuthHandler: turn.LongTermTURNRESTAuthHandler(g.Config.AuthSecret, logger),
// PacketConnConfigs is a list of UDP Listeners and the configuration around them
ListenerConfigs: listenerConfigs,
LoggerFactory: loggerF,
})
if err != nil {
return fmt.Errorf("Failed to start server: %w", err)
}
// Block until user sends SIGINT or SIGTERM
sigs := make(chan os.Signal, 1)
signal.Notify(sigs, syscall.SIGINT, syscall.SIGTERM)
<-sigs
if err = server.Close(); err != nil {
return fmt.Errorf("Failed to close server: %w", err)
}
return nil
}