Implement remote DNS

This commit implements remote DNS. It introduces two new dependencies:
ttlcache and dns.

Remote DNS intercepts UDP DNS queries for A records on port 53. It
replies with an unused IP address from an address pool, 198.18.0.0/15 by
default. When obtaining a new address from the pool, tun2socks needs to
memorize which name the address belongs to, so that when a client
connects to the address, it can instruct the proxy to connect to the
FQDN. To implement this IP to name mapping, ttlcache is used.
To prevent using multiple addresses for the same name, ttlcache is also
used to implement a name to IP mapping. If an IP address is already
cached for a name, that address is returned instread.
When building a connection, the connection metadata is inspected and if
the destination address is associated with a DNS name, the proxy is
instructed to use this name instead of the IP address.

component/remotedns/handle.go

@@ -0,0 +1,88 @@
+package remotedns
+
+import (
+	"net"
+
+	"github.com/miekg/dns"
+	"gvisor.dev/gvisor/pkg/tcpip"
+	"gvisor.dev/gvisor/pkg/tcpip/adapters/gonet"
+	"gvisor.dev/gvisor/pkg/tcpip/stack"
+	"gvisor.dev/gvisor/pkg/waiter"
+
+	"github.com/xjasonlyu/tun2socks/v2/log"
+	M "github.com/xjasonlyu/tun2socks/v2/metadata"
+)
+
+func RewriteMetadata(metadata *M.Metadata) bool {
+	if !IsEnabled() {
+		return false
+	}
+	dstName, found := getCachedName(metadata.DstIP)
+	if !found {
+		return false
+	}
+	metadata.DstIP = nil
+	metadata.DstName = dstName
+	return true
+}
+
+func HandleDNSQuery(s *stack.Stack, id stack.TransportEndpointID, ptr *stack.PacketBuffer) bool {
+	if !IsEnabled() {
+		return false
+	}
+
+	msg := dns.Msg{}
+	err := msg.Unpack(ptr.Data().AsRange().ToSlice())
+
+	isCorrectEndpoint := id.LocalPort == 53 && (listenAddress.Equal(id.LocalAddress.AsSlice()) || listenAddress.IsUnspecified())
+
+	// Ignore UDP packets that are not matching the listen address and are not recursive queries
+	if !isCorrectEndpoint || err != nil || len(msg.Question) != 1 || msg.Question[0].Qtype != dns.TypeA &&
+		msg.Question[0].Qtype != dns.TypeAAAA || msg.Question[0].Qclass != dns.ClassINET || !msg.RecursionDesired ||
+		msg.Response {
+		return false
+	}
+
+	qname := msg.Question[0].Name
+	qtype := msg.Question[0].Qtype
+
+	log.Debugf("[DNS] query %s %s", dns.TypeToString[qtype], qname)
+
+	var ip net.IP
+	if qtype == dns.TypeA {
+		rr := dns.A{}
+		ip = findOrInsertNameAndReturnIP(4, qname)
+		if ip == nil {
+			log.Warnf("[DNS] IP space exhausted")
+			return true
+		}
+		rr.A = ip
+		rr.Hdr.Name = qname
+		rr.Hdr.Ttl = dnsTTL
+		rr.Hdr.Class = dns.ClassINET
+		rr.Hdr.Rrtype = qtype
+		msg.Answer = append(msg.Answer, &rr)
+	}
+
+	msg.Response = true
+	msg.RecursionDesired = false
+	msg.RecursionAvailable = true
+
+	var wq waiter.Queue
+
+	ep, err2 := s.NewEndpoint(ptr.TransportProtocolNumber, ptr.NetworkProtocolNumber, &wq)
+	if err2 != nil {
+		return true
+	}
+	defer ep.Close()
+
+	ep.Bind(tcpip.FullAddress{NIC: ptr.NICID, Addr: id.LocalAddress, Port: id.LocalPort})
+	conn := gonet.NewUDPConn(&wq, ep)
+	defer conn.Close()
+	packed, err := msg.Pack()
+	if err != nil {
+		return true
+	}
+	_, _ = conn.WriteTo(packed, &net.UDPAddr{IP: id.RemoteAddress.AsSlice(), Port: int(id.RemotePort)})
+	return true
+}

component/remotedns/iputil.go

@@ -0,0 +1,36 @@
+package remotedns
+
+import "net"
+
+func copyIP(ip net.IP) net.IP {
+	dup := make(net.IP, len(ip))
+	copy(dup, ip)
+	return dup
+}
+
+func incrementIP(ip net.IP) net.IP {
+	result := copyIP(ip)
+	for i := len(result) - 1; i >= 0; i-- {
+		result[i]++
+		if result[i] != 0 {
+			break
+		}
+	}
+	return result
+}
+
+func getBroadcastAddress(ipnet *net.IPNet) net.IP {
+	result := copyIP(ipnet.IP)
+	for i := 0; i < len(ipnet.IP); i++ {
+		result[i] |= ^ipnet.Mask[i]
+	}
+	return result
+}
+
+func getNetworkAddress(ipnet *net.IPNet) net.IP {
+	result := copyIP(ipnet.IP)
+	for i := 0; i < len(ipnet.IP); i++ {
+		result[i] &= ipnet.Mask[i]
+	}
+	return result
+}

component/remotedns/pool.go

@@ -0,0 +1,88 @@
+package remotedns
+
+import (
+	"net"
+	"sync"
+	"time"
+
+	"github.com/jellydator/ttlcache/v3"
+)
+
+var (
+	ipToName = ttlcache.New[string, string]()
+	nameToIP = ttlcache.New[string, net.IP]()
+	mutex    = sync.Mutex{}
+
+	ip4NextAddress      net.IP
+	ip4BroadcastAddress net.IP
+)
+
+func findOrInsertNameAndReturnIP(ipVersion int, name string) net.IP {
+	if ipVersion != 4 {
+		panic("Method not implemented for IPv6")
+	}
+	mutex.Lock()
+	defer mutex.Unlock()
+	var result net.IP = nil
+	var ipnet *net.IPNet
+	var nextAddress *net.IP
+	var broadcastAddress net.IP
+	if ipVersion == 4 {
+		ipnet = ip4net
+		nextAddress = &ip4NextAddress
+		broadcastAddress = ip4BroadcastAddress
+	}
+
+	entry := nameToIP.Get(name)
+	if entry != nil {
+		ip := entry.Value()
+		ipToName.Touch(ip.String())
+		return ip
+	}
+
+	// Beginning from the pointer to the next most likely free IP, loop through the IP address space
+	// until either a free IP is found or the space is exhausted
+	passedBroadcastAddress := false
+	for result == nil {
+		if nextAddress.Equal(broadcastAddress) {
+			*nextAddress = getNetworkAddress(ipnet)
+			*nextAddress = incrementIP(ipnet.IP)
+
+			// We have seen the broadcast address twice during looping
+			// This means that our IP address space is exhausted
+			if passedBroadcastAddress {
+				return nil
+			}
+			passedBroadcastAddress = true
+		}
+
+		// Skip the listen address if that is inside our pool range
+		if nextAddress.Equal(listenAddress) {
+			*nextAddress = incrementIP(*nextAddress)
+			continue
+		}
+
+		// Do not touch entries that exist in the cache already.
+		hasKey := ipToName.Has((*nextAddress).String())
+		if !hasKey {
+			_ = ipToName.Set((*nextAddress).String(), name, time.Duration(dnsTTL)*time.Second+cacheGraceTime)
+			_ = nameToIP.Set(name, *nextAddress, time.Duration(dnsTTL)*time.Second+cacheGraceTime)
+			result = *nextAddress
+		}
+
+		*nextAddress = incrementIP(*nextAddress)
+	}
+
+	return result
+}
+
+func getCachedName(address net.IP) (string, bool) {
+	mutex.Lock()
+	defer mutex.Unlock()
+	entry := ipToName.Get(address.String())
+	if entry == nil {
+		return "", false
+	}
+	nameToIP.Touch(entry.Value())
+	return entry.Value(), true
+}

component/remotedns/settings.go

@@ -0,0 +1,55 @@
+package remotedns
+
+import (
+	"errors"
+	"net"
+	"time"
+)
+
+// Timeouts are somewhat arbitrary. For example, netcat will resolve the DNS
+// names upon startup and then stick to the resolved IP address. A timeout of 1
+// second may therefore be too low in cases where the first UDP packet is not
+// sent immediately.
+// cacheGraceTime defines how long an entry should still be retained in the cache
+// after being resolved by DNS.
+const (
+	cacheGraceTime = 30 * time.Second
+)
+
+var (
+	enabled              = false
+	dnsTTL        uint32 = 0
+	ip4net        *net.IPNet
+	listenAddress net.IP
+)
+
+func IsEnabled() bool {
+	return enabled
+}
+
+func SetDNSTTL(timeout time.Duration) {
+	dnsTTL = uint32(timeout.Seconds())
+}
+
+func SetListenAddress(ip net.IP) {
+	listenAddress = ip
+}
+
+func SetNetwork(ipnet *net.IPNet) error {
+	leadingOnes, _ := ipnet.Mask.Size()
+	if len(ipnet.IP) == 4 {
+		if leadingOnes > 30 {
+			return errors.New("IPv4 remote DNS subnet too small")
+		}
+		ip4net = ipnet
+	} else {
+		return errors.New("unsupported protocol")
+	}
+	return nil
+}
+
+func Enable() {
+	ip4NextAddress = incrementIP(getNetworkAddress(ip4net))
+	ip4BroadcastAddress = getBroadcastAddress(ip4net)
+	enabled = true
+}

core/udp.go

@@ -7,31 +7,38 @@ import (
 	"gvisor.dev/gvisor/pkg/tcpip/transport/udp"
 	"gvisor.dev/gvisor/pkg/waiter"
 
+	"github.com/xjasonlyu/tun2socks/v2/component/remotedns"
 	"github.com/xjasonlyu/tun2socks/v2/core/adapter"
 	"github.com/xjasonlyu/tun2socks/v2/core/option"
 )
 
 func withUDPHandler(handle func(adapter.UDPConn)) option.Option {
 	return func(s *stack.Stack) error {
-		udpForwarder := udp.NewForwarder(s, func(r *udp.ForwarderRequest) {
-			var (
-				wq waiter.Queue
-				id = r.ID()
-			)
-			ep, err := r.CreateEndpoint(&wq)
-			if err != nil {
-				glog.Debugf("forward udp request: %s:%d->%s:%d: %s",
-					id.RemoteAddress, id.RemotePort, id.LocalAddress, id.LocalPort, err)
-				return
+		s.SetTransportProtocolHandler(udp.ProtocolNumber, func(id stack.TransportEndpointID, ptr *stack.PacketBuffer) bool {
+			if remotedns.HandleDNSQuery(s, id, ptr) {
+				return true
 			}
 
-			conn := &udpConn{
-				UDPConn: gonet.NewUDPConn(&wq, ep),
-				id:      id,
-			}
-			handle(conn)
+			udpForwarder := udp.NewForwarder(s, func(r *udp.ForwarderRequest) {
+				var (
+					wq waiter.Queue
+					id = r.ID()
+				)
+				ep, err := r.CreateEndpoint(&wq)
+				if err != nil {
+					glog.Debugf("forward udp request %s:%d->%s:%d: %s",
+						id.RemoteAddress, id.RemotePort, id.LocalAddress, id.LocalPort, err)
+					return
+				}
+
+				conn := &udpConn{
+					UDPConn: gonet.NewUDPConn(&wq, ep),
+					id:      id,
+				}
+				handle(conn)
+			})
+			return udpForwarder.HandlePacket(id, ptr)
 		})
-		s.SetTransportProtocolHandler(udp.ProtocolNumber, udpForwarder.HandlePacket)
 		return nil
 	}
 }