iptables: support SO_ORIGINAL_DST

Envoy (#170) uses this to get the original destination of redirected
packets.

13 files changed, 344 insertions, 13 deletions

diff --git a/pkg/abi/linux/netfilter.go b/pkg/abi/linux/netfilter.go
index a91f9f018..9c27f7bb2 100644
--- a/pkg/abi/linux/netfilter.go
+++ b/pkg/abi/linux/netfilter.go
@@ -59,7 +59,7 @@ var VerdictStrings = map[int32]string{
 	NF_RETURN:      "RETURN",
 }
 
-// Socket options. These correspond to values in
+// Socket options for SOL_SOCKET. These correspond to values in
 // include/uapi/linux/netfilter_ipv4/ip_tables.h.
 const (
 	IPT_BASE_CTL            = 64
@@ -74,6 +74,12 @@ const (
 	IPT_SO_GET_MAX             = IPT_SO_GET_REVISION_TARGET
 )
 
+// Socket option for SOL_IP. This corresponds to the value in
+// include/uapi/linux/netfilter_ipv4.h.
+const (
+	SO_ORIGINAL_DST = 80
+)
+
 // Name lengths. These correspond to values in
 // include/uapi/linux/netfilter/x_tables.h.
 const (
diff --git a/pkg/abi/linux/socket.go b/pkg/abi/linux/socket.go
index c24a8216e..d6946bb82 100644
--- a/pkg/abi/linux/socket.go
+++ b/pkg/abi/linux/socket.go
@@ -239,11 +239,13 @@ const SockAddrMax = 128
 type InetAddr [4]byte
 
 // SockAddrInet is struct sockaddr_in, from uapi/linux/in.h.
+//
+// +marshal
 type SockAddrInet struct {
 	Family uint16
 	Port   uint16
 	Addr   InetAddr
-	Zero   [8]uint8 // pad to sizeof(struct sockaddr).
+	_      [8]uint8 // pad to sizeof(struct sockaddr).
 }
 
 // InetMulticastRequest is struct ip_mreq, from uapi/linux/in.h.
diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go
index f86e6cd7a..31a168f7e 100644
--- a/pkg/sentry/socket/netstack/netstack.go
+++ b/pkg/sentry/socket/netstack/netstack.go
@@ -1490,6 +1490,10 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (marsha
 		vP := primitive.Int32(boolToInt32(v))
 		return &vP, nil
 
+	case linux.SO_ORIGINAL_DST:
+		// TODO(gvisor.dev/issue/170): ip6tables.
+		return nil, syserr.ErrInvalidArgument
+
 	default:
 		emitUnimplementedEventIPv6(t, name)
 	}
@@ -1600,6 +1604,19 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in
 		vP := primitive.Int32(boolToInt32(v))
 		return &vP, nil
 
+	case linux.SO_ORIGINAL_DST:
+		if outLen < int(binary.Size(linux.SockAddrInet{})) {
+			return nil, syserr.ErrInvalidArgument
+		}
+
+		var v tcpip.OriginalDestinationOption
+		if err := ep.GetSockOpt(&v); err != nil {
+			return nil, syserr.TranslateNetstackError(err)
+		}
+
+		a, _ := ConvertAddress(linux.AF_INET, tcpip.FullAddress(v))
+		return a.(*linux.SockAddrInet), nil
+
 	default:
 		emitUnimplementedEventIP(t, name)
 	}
diff --git a/pkg/sentry/strace/socket.go b/pkg/sentry/strace/socket.go
index c0512de89..b51c4c941 100644
--- a/pkg/sentry/strace/socket.go
+++ b/pkg/sentry/strace/socket.go
@@ -521,6 +521,7 @@ var sockOptNames = map[uint64]abi.ValueSet{
 		linux.IP_ROUTER_ALERT:           "IP_ROUTER_ALERT",
 		linux.IP_PKTOPTIONS:             "IP_PKTOPTIONS",
 		linux.IP_MTU:                    "IP_MTU",
+		linux.SO_ORIGINAL_DST:           "SO_ORIGINAL_DST",
 	},
 	linux.SOL_SOCKET: {
 		linux.SO_ERROR:        "SO_ERROR",
diff --git a/pkg/tcpip/stack/conntrack.go b/pkg/tcpip/stack/conntrack.go
index 559a1c4dd..470c265aa 100644
--- a/pkg/tcpip/stack/conntrack.go
+++ b/pkg/tcpip/stack/conntrack.go
@@ -240,7 +240,10 @@ func (ct *ConnTrack) connFor(pkt *PacketBuffer) (*conn, direction) {
 	if err != nil {
 		return nil, dirOriginal
 	}
+	return ct.connForTID(tid)
+}
 
+func (ct *ConnTrack) connForTID(tid tupleID) (*conn, direction) {
 	bucket := ct.bucket(tid)
 	now := time.Now()
 
@@ -604,3 +607,26 @@ func (ct *ConnTrack) reapTupleLocked(tuple *tuple, bucket int, now time.Time) bo
 
 	return true
 }
+
+func (ct *ConnTrack) originalDst(epID TransportEndpointID) (tcpip.Address, uint16, *tcpip.Error) {
+	// Lookup the connection. The reply's original destination
+	// describes the original address.
+	tid := tupleID{
+		srcAddr:    epID.LocalAddress,
+		srcPort:    epID.LocalPort,
+		dstAddr:    epID.RemoteAddress,
+		dstPort:    epID.RemotePort,
+		transProto: header.TCPProtocolNumber,
+		netProto:   header.IPv4ProtocolNumber,
+	}
+	conn, _ := ct.connForTID(tid)
+	if conn == nil {
+		// Not a tracked connection.
+		return "", 0, tcpip.ErrNotConnected
+	} else if conn.manip == manipNone {
+		// Unmanipulated connection.
+		return "", 0, tcpip.ErrInvalidOptionValue
+	}
+
+	return conn.original.dstAddr, conn.original.dstPort, nil
+}
diff --git a/pkg/tcpip/stack/iptables.go b/pkg/tcpip/stack/iptables.go
index cbbae4224..110ba073d 100644
--- a/pkg/tcpip/stack/iptables.go
+++ b/pkg/tcpip/stack/iptables.go
@@ -218,19 +218,16 @@ func (it *IPTables) Check(hook Hook, pkt *PacketBuffer, gso *GSO, r *Route, addr
 	// Many users never configure iptables. Spare them the cost of rule
 	// traversal if rules have never been set.
 	it.mu.RLock()
+	defer it.mu.RUnlock()
 	if !it.modified {
-		it.mu.RUnlock()
 		return true
 	}
-	it.mu.RUnlock()
 
 	// Packets are manipulated only if connection and matching
 	// NAT rule exists.
 	shouldTrack := it.connections.handlePacket(pkt, hook, gso, r)
 
 	// Go through each table containing the hook.
-	it.mu.RLock()
-	defer it.mu.RUnlock()
 	priorities := it.priorities[hook]
 	for _, tableID := range priorities {
 		// If handlePacket already NATed the packet, we don't need to
@@ -418,3 +415,9 @@ func (it *IPTables) checkRule(hook Hook, pkt *PacketBuffer, table Table, ruleIdx
 	// All the matchers matched, so run the target.
 	return rule.Target.Action(pkt, &it.connections, hook, gso, r, address)
 }
+
+// OriginalDst returns the original destination of redirected connections. It
+// returns an error if the connection doesn't exist or isn't redirected.
+func (it *IPTables) OriginalDst(epID TransportEndpointID) (tcpip.Address, uint16, *tcpip.Error) {
+	return it.connections.originalDst(epID)
+}
diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go
index a634b9b60..45f59b60f 100644
--- a/pkg/tcpip/tcpip.go
+++ b/pkg/tcpip/tcpip.go
@@ -954,6 +954,10 @@ type DefaultTTLOption uint8
 // classic BPF filter on a given endpoint.
 type SocketDetachFilterOption int
 
+// OriginalDestinationOption is used to get the original destination address
+// and port of a redirected packet.
+type OriginalDestinationOption FullAddress
+
 // IPPacketInfo is the message structure for IP_PKTINFO.
 //
 // +stateify savable
diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go
index 0f7487963..682687ebe 100644
--- a/pkg/tcpip/transport/tcp/endpoint.go
+++ b/pkg/tcpip/transport/tcp/endpoint.go
@@ -2017,6 +2017,17 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error {
 		*o = tcpip.TCPDeferAcceptOption(e.deferAccept)
 		e.UnlockUser()
 
+	case *tcpip.OriginalDestinationOption:
+		ipt := e.stack.IPTables()
+		addr, port, err := ipt.OriginalDst(e.ID)
+		if err != nil {
+			return err
+		}
+		*o = tcpip.OriginalDestinationOption{
+			Addr: addr,
+			Port: port,
+		}
+
 	default:
 		return tcpip.ErrUnknownProtocolOption
 	}
diff --git a/test/iptables/BUILD b/test/iptables/BUILD
index 40b63ebbe..66453772a 100644
--- a/test/iptables/BUILD
+++ b/test/iptables/BUILD
@@ -9,6 +9,7 @@ go_library(
         "filter_input.go",
         "filter_output.go",
         "iptables.go",
+        "iptables_unsafe.go",
         "iptables_util.go",
         "nat.go",
     ],
diff --git a/test/iptables/iptables_test.go b/test/iptables/iptables_test.go
index 550b6198a..fda5f694f 100644
--- a/test/iptables/iptables_test.go
+++ b/test/iptables/iptables_test.go
@@ -371,3 +371,11 @@ func TestFilterAddrs(t *testing.T) {
 		}
 	}
 }
+
+func TestNATPreOriginalDst(t *testing.T) {
+	singleTest(t, NATPreOriginalDst{})
+}
+
+func TestNATOutOriginalDst(t *testing.T) {
+	singleTest(t, NATOutOriginalDst{})
+}
diff --git a/test/iptables/iptables_unsafe.go b/test/iptables/iptables_unsafe.go
new file mode 100644
index 000000000..bd85a8fea
--- /dev/null
+++ b/test/iptables/iptables_unsafe.go
@@ -0,0 +1,63 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package iptables
+
+import (
+	"fmt"
+	"syscall"
+	"unsafe"
+)
+
+type originalDstError struct {
+	errno syscall.Errno
+}
+
+func (e originalDstError) Error() string {
+	return fmt.Sprintf("errno (%d) when calling getsockopt(SO_ORIGINAL_DST): %v", int(e.errno), e.errno.Error())
+}
+
+// SO_ORIGINAL_DST gets the original destination of a redirected packet via
+// getsockopt.
+const SO_ORIGINAL_DST = 80
+
+func originalDestination4(connfd int) (syscall.RawSockaddrInet4, error) {
+	var addr syscall.RawSockaddrInet4
+	var addrLen uint32 = syscall.SizeofSockaddrInet4
+	if errno := originalDestination(connfd, syscall.SOL_IP, unsafe.Pointer(&addr), &addrLen); errno != 0 {
+		return syscall.RawSockaddrInet4{}, originalDstError{errno}
+	}
+	return addr, nil
+}
+
+func originalDestination6(connfd int) (syscall.RawSockaddrInet6, error) {
+	var addr syscall.RawSockaddrInet6
+	var addrLen uint32 = syscall.SizeofSockaddrInet6
+	if errno := originalDestination(connfd, syscall.SOL_IPV6, unsafe.Pointer(&addr), &addrLen); errno != 0 {
+		return syscall.RawSockaddrInet6{}, originalDstError{errno}
+	}
+	return addr, nil
+}
+
+func originalDestination(connfd int, level uintptr, optval unsafe.Pointer, optlen *uint32) syscall.Errno {
+	_, _, errno := syscall.Syscall6(
+		syscall.SYS_GETSOCKOPT,
+		uintptr(connfd),
+		level,
+		SO_ORIGINAL_DST,
+		uintptr(optval),
+		uintptr(unsafe.Pointer(optlen)),
+		0)
+	return errno
+}
diff --git a/test/iptables/iptables_util.go b/test/iptables/iptables_util.go
index ca80a4b5f..5125fe47b 100644
--- a/test/iptables/iptables_util.go
+++ b/test/iptables/iptables_util.go
@@ -15,6 +15,8 @@
 package iptables
 
 import (
+	"encoding/binary"
+	"errors"
 	"fmt"
 	"net"
 	"os/exec"
@@ -218,17 +220,58 @@ func filterAddrs(addrs []string, ipv6 bool) []string {
 
 // getInterfaceName returns the name of the interface other than loopback.
 func getInterfaceName() (string, bool) {
-	var ifname string
+	iface, ok := getNonLoopbackInterface()
+	if !ok {
+		return "", false
+	}
+	return iface.Name, true
+}
+
+func getInterfaceAddrs(ipv6 bool) ([]net.IP, error) {
+	iface, ok := getNonLoopbackInterface()
+	if !ok {
+		return nil, errors.New("no non-loopback interface found")
+	}
+	addrs, err := iface.Addrs()
+	if err != nil {
+		return nil, err
+	}
+
+	// Get only IPv4 or IPv6 addresses.
+	ips := make([]net.IP, 0, len(addrs))
+	for _, addr := range addrs {
+		parts := strings.Split(addr.String(), "/")
+		var ip net.IP
+		// To16() returns IPv4 addresses as IPv4-mapped IPv6 addresses.
+		// So we check whether To4() returns nil to test whether the
+		// address is v4 or v6.
+		if v4 := net.ParseIP(parts[0]).To4(); ipv6 && v4 == nil {
+			ip = net.ParseIP(parts[0]).To16()
+		} else {
+			ip = v4
+		}
+		if ip != nil {
+			ips = append(ips, ip)
+		}
+	}
+	return ips, nil
+}
+
+func getNonLoopbackInterface() (net.Interface, bool) {
 	if interfaces, err := net.Interfaces(); err == nil {
 		for _, intf := range interfaces {
 			if intf.Name != "lo" {
-				ifname = intf.Name
-				break
+				return intf, true
 			}
 		}
 	}
+	return net.Interface{}, false
+}
 
-	return ifname, ifname != ""
+func htons(x uint16) uint16 {
+	buf := make([]byte, 2)
+	binary.BigEndian.PutUint16(buf, x)
+	return binary.LittleEndian.Uint16(buf)
 }
 
 func localIP(ipv6 bool) string {
diff --git a/test/iptables/nat.go b/test/iptables/nat.go
index ac0d91bb2..b7fea2527 100644
--- a/test/iptables/nat.go
+++ b/test/iptables/nat.go
@@ -18,12 +18,11 @@ import (
 	"errors"
 	"fmt"
 	"net"
+	"syscall"
 	"time"
 )
 
-const (
-	redirectPort = 42
-)
+const redirectPort = 42
 
 func init() {
 	RegisterTestCase(NATPreRedirectUDPPort{})
@@ -42,6 +41,8 @@ func init() {
 	RegisterTestCase(NATOutRedirectInvert{})
 	RegisterTestCase(NATRedirectRequiresProtocol{})
 	RegisterTestCase(NATLoopbackSkipsPrerouting{})
+	RegisterTestCase(NATPreOriginalDst{})
+	RegisterTestCase(NATOutOriginalDst{})
 }
 
 // NATPreRedirectUDPPort tests that packets are redirected to different port.
@@ -471,6 +472,151 @@ func (NATLoopbackSkipsPrerouting) LocalAction(ip net.IP, ipv6 bool) error {
 	return nil
 }
 
+// NATPreOriginalDst tests that SO_ORIGINAL_DST returns the pre-NAT destination
+// of PREROUTING NATted packets.
+type NATPreOriginalDst struct{}
+
+// Name implements TestCase.Name.
+func (NATPreOriginalDst) Name() string {
+	return "NATPreOriginalDst"
+}
+
+// ContainerAction implements TestCase.ContainerAction.
+func (NATPreOriginalDst) ContainerAction(ip net.IP, ipv6 bool) error {
+	// Redirect incoming TCP connections to acceptPort.
+	if err := natTable(ipv6, "-A", "PREROUTING",
+		"-p", "tcp",
+		"--destination-port", fmt.Sprintf("%d", dropPort),
+		"-j", "REDIRECT", "--to-port", fmt.Sprintf("%d", acceptPort)); err != nil {
+		return err
+	}
+
+	addrs, err := getInterfaceAddrs(ipv6)
+	if err != nil {
+		return err
+	}
+	return listenForRedirectedConn(ipv6, addrs)
+}
+
+// LocalAction implements TestCase.LocalAction.
+func (NATPreOriginalDst) LocalAction(ip net.IP, ipv6 bool) error {
+	return connectTCP(ip, dropPort, sendloopDuration)
+}
+
+// NATOutOriginalDst tests that SO_ORIGINAL_DST returns the pre-NAT destination
+// of OUTBOUND NATted packets.
+type NATOutOriginalDst struct{}
+
+// Name implements TestCase.Name.
+func (NATOutOriginalDst) Name() string {
+	return "NATOutOriginalDst"
+}
+
+// ContainerAction implements TestCase.ContainerAction.
+func (NATOutOriginalDst) ContainerAction(ip net.IP, ipv6 bool) error {
+	// Redirect incoming TCP connections to acceptPort.
+	if err := natTable(ipv6, "-A", "OUTPUT", "-p", "tcp", "-j", "REDIRECT", "--to-port", fmt.Sprintf("%d", acceptPort)); err != nil {
+		return err
+	}
+
+	connCh := make(chan error)
+	go func() {
+		connCh <- connectTCP(ip, dropPort, sendloopDuration)
+	}()
+
+	if err := listenForRedirectedConn(ipv6, []net.IP{ip}); err != nil {
+		return err
+	}
+	return <-connCh
+}
+
+// LocalAction implements TestCase.LocalAction.
+func (NATOutOriginalDst) LocalAction(ip net.IP, ipv6 bool) error {
+	// No-op.
+	return nil
+}
+
+func listenForRedirectedConn(ipv6 bool, originalDsts []net.IP) error {
+	// The net package doesn't give guarantee access to the connection's
+	// underlying FD, and thus we cannot call getsockopt. We have to use
+	// traditional syscalls for SO_ORIGINAL_DST.
+
+	// Create the listening socket, bind, listen, and accept.
+	family := syscall.AF_INET
+	if ipv6 {
+		family = syscall.AF_INET6
+	}
+	sockfd, err := syscall.Socket(family, syscall.SOCK_STREAM, 0)
+	if err != nil {
+		return err
+	}
+	defer syscall.Close(sockfd)
+
+	var bindAddr syscall.Sockaddr
+	if ipv6 {
+		bindAddr = &syscall.SockaddrInet6{
+			Port: acceptPort,
+			Addr: [16]byte{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}, // in6addr_any
+		}
+	} else {
+		bindAddr = &syscall.SockaddrInet4{
+			Port: acceptPort,
+			Addr: [4]byte{0, 0, 0, 0}, // INADDR_ANY
+		}
+	}
+	if err := syscall.Bind(sockfd, bindAddr); err != nil {
+		return err
+	}
+
+	if err := syscall.Listen(sockfd, 1); err != nil {
+		return err
+	}
+
+	connfd, _, err := syscall.Accept(sockfd)
+	if err != nil {
+		return err
+	}
+	defer syscall.Close(connfd)
+
+	// Verify that, despite listening on acceptPort, SO_ORIGINAL_DST
+	// indicates the packet was sent to originalDst:dropPort.
+	if ipv6 {
+		got, err := originalDestination6(connfd)
+		if err != nil {
+			return err
+		}
+		// The original destination could be any of our IPs.
+		for _, dst := range originalDsts {
+			want := syscall.RawSockaddrInet6{
+				Family: syscall.AF_INET6,
+				Port:   htons(dropPort),
+			}
+			copy(want.Addr[:], dst.To16())
+			if got == want {
+				return nil
+			}
+		}
+		return fmt.Errorf("SO_ORIGINAL_DST returned %+v, but wanted one of %+v (note: port numbers are in network byte order)", got, originalDsts)
+	} else {
+		got, err := originalDestination4(connfd)
+		if err != nil {
+			return err
+		}
+		// The original destination could be any of our IPs.
+		for _, dst := range originalDsts {
+			want := syscall.RawSockaddrInet4{
+				Family: syscall.AF_INET,
+				Port:   htons(dropPort),
+			}
+			copy(want.Addr[:], dst.To4())
+			if got == want {
+				return nil
+			}
+		}
+		return fmt.Errorf("SO_ORIGINAL_DST returned %+v, but wanted one of %+v (note: port numbers are in network byte order)", got, originalDsts)
+	}
+}
+
 // loopbackTests runs an iptables rule and ensures that packets sent to
 // dest:dropPort are received by localhost:acceptPort.
 func loopbackTest(ipv6 bool, dest net.IP, args ...string) error {