Internal change.

PiperOrigin-RevId: 294952610

pkg/abi/linux/socket.go

@@ -411,6 +411,15 @@ type ControlMessageCredentials struct {
 	GID uint32
 }
 
+// A ControlMessageIPPacketInfo is IP_PKTINFO socket control message.
+//
+// ControlMessageIPPacketInfo represents struct in_pktinfo from linux/in.h.
+type ControlMessageIPPacketInfo struct {
+	NIC             int32
+	LocalAddr       InetAddr
+	DestinationAddr InetAddr
+}
+
 // SizeOfControlMessageCredentials is the binary size of a
 // ControlMessageCredentials struct.
 var SizeOfControlMessageCredentials = int(binary.Size(ControlMessageCredentials{}))
@@ -431,6 +440,10 @@ const SizeOfControlMessageTOS = 1
 // SizeOfControlMessageTClass is the size of an IPV6_TCLASS control message.
 const SizeOfControlMessageTClass = 4
 
+// SizeOfControlMessageIPPacketInfo is the size of an IP_PKTINFO
+// control message.
+const SizeOfControlMessageIPPacketInfo = 12
+
 // SCM_MAX_FD is the maximum number of FDs accepted in a single sendmsg call.
 // From net/scm.h.
 const SCM_MAX_FD = 253

pkg/sentry/socket/control/BUILD

@@ -19,6 +19,7 @@ go_library(
         "//pkg/sentry/socket",
         "//pkg/sentry/socket/unix/transport",
         "//pkg/syserror",
+        "//pkg/tcpip",
         "//pkg/usermem",
     ],
 )

pkg/sentry/socket/control/control.go

@@ -26,6 +26,7 @@ import (
 	"gvisor.dev/gvisor/pkg/sentry/socket"
 	"gvisor.dev/gvisor/pkg/sentry/socket/unix/transport"
 	"gvisor.dev/gvisor/pkg/syserror"
+	"gvisor.dev/gvisor/pkg/tcpip"
 	"gvisor.dev/gvisor/pkg/usermem"
 )
 
@@ -338,6 +339,22 @@ func PackTClass(t *kernel.Task, tClass int32, buf []byte) []byte {
 	)
 }
 
+// PackIPPacketInfo packs an IP_PKTINFO socket control message.
+func PackIPPacketInfo(t *kernel.Task, packetInfo tcpip.IPPacketInfo, buf []byte) []byte {
+	var p linux.ControlMessageIPPacketInfo
+	p.NIC = int32(packetInfo.NIC)
+	copy(p.LocalAddr[:], []byte(packetInfo.LocalAddr))
+	copy(p.DestinationAddr[:], []byte(packetInfo.DestinationAddr))
+
+	return putCmsgStruct(
+		buf,
+		linux.SOL_IP,
+		linux.IP_PKTINFO,
+		t.Arch().Width(),
+		p,
+	)
+}
+
 // PackControlMessages packs control messages into the given buffer.
 //
 // We skip control messages specific to Unix domain sockets.
@@ -362,6 +379,10 @@ func PackControlMessages(t *kernel.Task, cmsgs socket.ControlMessages, buf []byt
 		buf = PackTClass(t, cmsgs.IP.TClass, buf)
 	}
 
+	if cmsgs.IP.HasIPPacketInfo {
+		buf = PackIPPacketInfo(t, cmsgs.IP.PacketInfo, buf)
+	}
+
 	return buf
 }
 
@@ -394,6 +415,16 @@ func CmsgsSpace(t *kernel.Task, cmsgs socket.ControlMessages) int {
 	return space
 }
 
+// NewIPPacketInfo returns the IPPacketInfo struct.
+func NewIPPacketInfo(packetInfo linux.ControlMessageIPPacketInfo) tcpip.IPPacketInfo {
+	var p tcpip.IPPacketInfo
+	p.NIC = tcpip.NICID(packetInfo.NIC)
+	copy([]byte(p.LocalAddr), packetInfo.LocalAddr[:])
+	copy([]byte(p.DestinationAddr), packetInfo.DestinationAddr[:])
+
+	return p
+}
+
 // Parse parses a raw socket control message into portable objects.
 func Parse(t *kernel.Task, socketOrEndpoint interface{}, buf []byte) (socket.ControlMessages, error) {
 	var (
@@ -468,6 +499,18 @@ func Parse(t *kernel.Task, socketOrEndpoint interface{}, buf []byte) (socket.Con
 				binary.Unmarshal(buf[i:i+linux.SizeOfControlMessageTOS], usermem.ByteOrder, &cmsgs.IP.TOS)
 				i += binary.AlignUp(length, width)
 
+			case linux.IP_PKTINFO:
+				if length < linux.SizeOfControlMessageIPPacketInfo {
+					return socket.ControlMessages{}, syserror.EINVAL
+				}
+
+				cmsgs.IP.HasIPPacketInfo = true
+				var packetInfo linux.ControlMessageIPPacketInfo
+				binary.Unmarshal(buf[i:i+linux.SizeOfControlMessageIPPacketInfo], usermem.ByteOrder, &packetInfo)
+
+				cmsgs.IP.PacketInfo = NewIPPacketInfo(packetInfo)
+				i += binary.AlignUp(length, width)
+
 			default:
 				return socket.ControlMessages{}, syserror.EINVAL
 			}

pkg/sentry/socket/hostinet/socket.go

@@ -289,7 +289,7 @@ func (s *socketOperations) GetSockOpt(t *kernel.Task, level int, name int, outPt
 	switch level {
 	case linux.SOL_IP:
 		switch name {
-		case linux.IP_TOS, linux.IP_RECVTOS:
+		case linux.IP_TOS, linux.IP_RECVTOS, linux.IP_PKTINFO:
 			optlen = sizeofInt32
 		}
 	case linux.SOL_IPV6:
@@ -336,6 +336,8 @@ func (s *socketOperations) SetSockOpt(t *kernel.Task, level int, name int, opt [
 		switch name {
 		case linux.IP_TOS, linux.IP_RECVTOS:
 			optlen = sizeofInt32
+		case linux.IP_PKTINFO:
+			optlen = linux.SizeOfControlMessageIPPacketInfo
 		}
 	case linux.SOL_IPV6:
 		switch name {
@@ -473,7 +475,14 @@ func (s *socketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags
 			case syscall.IP_TOS:
 				controlMessages.IP.HasTOS = true
 				binary.Unmarshal(unixCmsg.Data[:linux.SizeOfControlMessageTOS], usermem.ByteOrder, &controlMessages.IP.TOS)
+
+			case syscall.IP_PKTINFO:
+				controlMessages.IP.HasIPPacketInfo = true
+				var packetInfo linux.ControlMessageIPPacketInfo
+				binary.Unmarshal(unixCmsg.Data[:linux.SizeOfControlMessageIPPacketInfo], usermem.ByteOrder, &packetInfo)
+				controlMessages.IP.PacketInfo = control.NewIPPacketInfo(packetInfo)
 			}
+
 		case syscall.SOL_IPV6:
 			switch unixCmsg.Header.Type {
 			case syscall.IPV6_TCLASS:

pkg/sentry/socket/netstack/netstack.go

@@ -1414,6 +1414,21 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in
 		}
 		return o, nil
 
+	case linux.IP_PKTINFO:
+		if outLen < sizeOfInt32 {
+			return nil, syserr.ErrInvalidArgument
+		}
+
+		v, err := ep.GetSockOptBool(tcpip.ReceiveIPPacketInfoOption)
+		if err != nil {
+			return nil, syserr.TranslateNetstackError(err)
+		}
+		var o int32
+		if v {
+			o = 1
+		}
+		return o, nil
+
 	default:
 		emitUnimplementedEventIP(t, name)
 	}
@@ -1762,6 +1777,7 @@ func setSockOptIPv6(t *kernel.Task, ep commonEndpoint, name int, optVal []byte)
 		linux.IPV6_IPSEC_POLICY,
 		linux.IPV6_JOIN_ANYCAST,
 		linux.IPV6_LEAVE_ANYCAST,
+		// TODO(b/148887420): Add support for IPV6_PKTINFO.
 		linux.IPV6_PKTINFO,
 		linux.IPV6_ROUTER_ALERT,
 		linux.IPV6_XFRM_POLICY,
@@ -1949,6 +1965,16 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s
 		}
 		return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.ReceiveTOSOption, v != 0))
 
+	case linux.IP_PKTINFO:
+		if len(optVal) == 0 {
+			return nil
+		}
+		v, err := parseIntOrChar(optVal)
+		if err != nil {
+			return err
+		}
+		return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.ReceiveIPPacketInfoOption, v != 0))
+
 	case linux.IP_ADD_SOURCE_MEMBERSHIP,
 		linux.IP_BIND_ADDRESS_NO_PORT,
 		linux.IP_BLOCK_SOURCE,
@@ -1964,7 +1990,6 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s
 		linux.IP_NODEFRAG,
 		linux.IP_OPTIONS,
 		linux.IP_PASSSEC,
-		linux.IP_PKTINFO,
 		linux.IP_RECVERR,
 		linux.IP_RECVFRAGSIZE,
 		linux.IP_RECVOPTS,
@@ -2395,10 +2420,12 @@ func (s *SocketOperations) nonBlockingRead(ctx context.Context, dst usermem.IOSe
 func (s *SocketOperations) controlMessages() socket.ControlMessages {
 	return socket.ControlMessages{
 		IP: tcpip.ControlMessages{
-			HasTimestamp: s.readCM.HasTimestamp && s.sockOptTimestamp,
-			Timestamp:    s.readCM.Timestamp,
-			HasTOS:       s.readCM.HasTOS,
-			TOS:          s.readCM.TOS,
+			HasTimestamp:    s.readCM.HasTimestamp && s.sockOptTimestamp,
+			Timestamp:       s.readCM.Timestamp,
+			HasTOS:          s.readCM.HasTOS,
+			TOS:             s.readCM.TOS,
+			HasIPPacketInfo: s.readCM.HasIPPacketInfo,
+			PacketInfo:      s.readCM.PacketInfo,
 		},
 	}
 }

pkg/tcpip/tcpip.go

@@ -328,6 +328,12 @@ type ControlMessages struct {
 
 	// Tclass is the IPv6 traffic class of the associated packet.
 	TClass int32
+
+	// HasIPPacketInfo indicates whether PacketInfo is set.
+	HasIPPacketInfo bool
+
+	// PacketInfo holds interface and address data on an incoming packet.
+	PacketInfo IPPacketInfo
 }
 
 // Endpoint is the interface implemented by transport protocols (e.g., tcp, udp)
@@ -503,6 +509,11 @@ const (
 	// V6OnlyOption is used by {G,S}etSockOptBool to specify whether an IPv6
 	// socket is to be restricted to sending and receiving IPv6 packets only.
 	V6OnlyOption
+
+	// ReceiveIPPacketInfoOption is used by {G,S}etSockOptBool to specify
+	// if more inforamtion is provided with incoming packets such
+	// as interface index and address.
+	ReceiveIPPacketInfoOption
 )
 
 // SockOptInt represents socket options which values have the int type.
@@ -685,6 +696,20 @@ type IPv4TOSOption uint8
 // for all subsequent outgoing IPv6 packets from the endpoint.
 type IPv6TrafficClassOption uint8
 
+// IPPacketInfo is the message struture for IP_PKTINFO.
+//
+// +stateify savable
+type IPPacketInfo struct {
+	// NIC is the ID of the NIC to be used.
+	NIC NICID
+
+	// LocalAddr is the local address.
+	LocalAddr Address
+
+	// DestinationAddr is the destination address.
+	DestinationAddr Address
+}
+
 // Route is a row in the routing table. It specifies through which NIC (and
 // gateway) sets of packets should be routed. A row is considered viable if the
 // masked target address matches the destination address in the row.

pkg/tcpip/transport/udp/endpoint.go

@@ -29,6 +29,7 @@ import (
 type udpPacket struct {
 	udpPacketEntry
 	senderAddress tcpip.FullAddress
+	packetInfo    tcpip.IPPacketInfo
 	data          buffer.VectorisedView `state:".(buffer.VectorisedView)"`
 	timestamp     int64
 	tos           uint8
@@ -118,6 +119,9 @@ type endpoint struct {
 	// as ancillary data to ControlMessages on Read.
 	receiveTOS bool
 
+	// receiveIPPacketInfo determines if the packet info is returned by Read.
+	receiveIPPacketInfo bool
+
 	// shutdownFlags represent the current shutdown state of the endpoint.
 	shutdownFlags tcpip.ShutdownFlags
 
@@ -254,11 +258,17 @@ func (e *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMess
 	}
 	e.mu.RLock()
 	receiveTOS := e.receiveTOS
+	receiveIPPacketInfo := e.receiveIPPacketInfo
 	e.mu.RUnlock()
 	if receiveTOS {
 		cm.HasTOS = true
 		cm.TOS = p.tos
 	}
+
+	if receiveIPPacketInfo {
+		cm.HasIPPacketInfo = true
+		cm.PacketInfo = p.packetInfo
+	}
 	return p.data.ToView(), cm, nil
 }
 
@@ -495,6 +505,13 @@ func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error {
 		}
 
 		e.v6only = v
+		return nil
+
+	case tcpip.ReceiveIPPacketInfoOption:
+		e.mu.Lock()
+		e.receiveIPPacketInfo = v
+		e.mu.Unlock()
+		return nil
 	}
 
 	return nil
@@ -703,6 +720,12 @@ func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) {
 		e.mu.RUnlock()
 
 		return v, nil
+
+	case tcpip.ReceiveIPPacketInfoOption:
+		e.mu.RLock()
+		v := e.receiveIPPacketInfo
+		e.mu.RUnlock()
+		return v, nil
 	}
 
 	return false, tcpip.ErrUnknownProtocolOption
@@ -1247,6 +1270,9 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk
 	switch r.NetProto {
 	case header.IPv4ProtocolNumber:
 		packet.tos, _ = header.IPv4(pkt.NetworkHeader).TOS()
+		packet.packetInfo.LocalAddr = r.LocalAddress
+		packet.packetInfo.DestinationAddr = r.RemoteAddress
+		packet.packetInfo.NIC = r.NICID()
 	}
 
 	packet.timestamp = e.stack.NowNanoseconds()

test/syscalls/linux/socket_ip_udp_generic.cc

@@ -357,5 +357,49 @@ TEST_P(UDPSocketPairTest, SetReuseAddrReusePort) {
   EXPECT_EQ(get, kSockOptOn);
 }
 
+// Test getsockopt for a socket which is not set with IP_PKTINFO option.
+TEST_P(UDPSocketPairTest, IPPKTINFODefault) {
+  auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair());
+
+  int get = -1;
+  socklen_t get_len = sizeof(get);
+
+  ASSERT_THAT(
+      getsockopt(sockets->first_fd(), SOL_IP, IP_PKTINFO, &get, &get_len),
+      SyscallSucceedsWithValue(0));
+  EXPECT_EQ(get_len, sizeof(get));
+  EXPECT_EQ(get, kSockOptOff);
+}
+
+// Test setsockopt and getsockopt for a socket with IP_PKTINFO option.
+TEST_P(UDPSocketPairTest, SetAndGetIPPKTINFO) {
+  auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair());
+
+  int level = SOL_IP;
+  int type = IP_PKTINFO;
+
+  // Check getsockopt before IP_PKTINFO is set.
+  int get = -1;
+  socklen_t get_len = sizeof(get);
+
+  ASSERT_THAT(setsockopt(sockets->first_fd(), level, type, &kSockOptOn,
+                         sizeof(kSockOptOn)),
+              SyscallSucceedsWithValue(0));
+
+  ASSERT_THAT(getsockopt(sockets->first_fd(), level, type, &get, &get_len),
+              SyscallSucceedsWithValue(0));
+  EXPECT_EQ(get, kSockOptOn);
+  EXPECT_EQ(get_len, sizeof(get));
+
+  ASSERT_THAT(setsockopt(sockets->first_fd(), level, type, &kSockOptOff,
+                         sizeof(kSockOptOff)),
+              SyscallSucceedsWithValue(0));
+
+  ASSERT_THAT(getsockopt(sockets->first_fd(), level, type, &get, &get_len),
+              SyscallSucceedsWithValue(0));
+  EXPECT_EQ(get, kSockOptOff);
+  EXPECT_EQ(get_len, sizeof(get));
+}
+
 }  // namespace testing
 }  // namespace gvisor

test/syscalls/linux/socket_ipv4_udp_unbound.cc

@@ -15,6 +15,7 @@
 #include "test/syscalls/linux/socket_ipv4_udp_unbound.h"
 
 #include <arpa/inet.h>
+#include <net/if.h>
 #include <sys/ioctl.h>
 #include <sys/socket.h>
 #include <sys/un.h>
@@ -2128,5 +2129,88 @@ TEST_P(IPv4UDPUnboundSocketTest, ReuseAddrReusePortDistribution) {
               SyscallSucceedsWithValue(kMessageSize));
 }
 
+// Test that socket will receive packet info control message.
+TEST_P(IPv4UDPUnboundSocketTest, SetAndReceiveIPPKTINFO) {
+  // TODO(gvisor.dev/issue/1202): ioctl() is not supported by hostinet.
+  SKIP_IF((IsRunningWithHostinet()));
+
+  auto sender = ASSERT_NO_ERRNO_AND_VALUE(NewSocket());
+  auto receiver = ASSERT_NO_ERRNO_AND_VALUE(NewSocket());
+  auto sender_addr = V4Loopback();
+  int level = SOL_IP;
+  int type = IP_PKTINFO;
+
+  ASSERT_THAT(
+      bind(receiver->get(), reinterpret_cast<sockaddr*>(&sender_addr.addr),
+           sender_addr.addr_len),
+      SyscallSucceeds());
+  socklen_t sender_addr_len = sender_addr.addr_len;
+  ASSERT_THAT(getsockname(receiver->get(),
+                          reinterpret_cast<sockaddr*>(&sender_addr.addr),
+                          &sender_addr_len),
+              SyscallSucceeds());
+  EXPECT_EQ(sender_addr_len, sender_addr.addr_len);
+
+  auto receiver_addr = V4Loopback();
+  reinterpret_cast<sockaddr_in*>(&receiver_addr.addr)->sin_port =
+      reinterpret_cast<sockaddr_in*>(&sender_addr.addr)->sin_port;
+  ASSERT_THAT(
+      connect(sender->get(), reinterpret_cast<sockaddr*>(&receiver_addr.addr),
+              receiver_addr.addr_len),
+      SyscallSucceeds());
+
+  // Allow socket to receive control message.
+  ASSERT_THAT(
+      setsockopt(receiver->get(), level, type, &kSockOptOn, sizeof(kSockOptOn)),
+      SyscallSucceeds());
+
+  // Prepare message to send.
+  constexpr size_t kDataLength = 1024;
+  msghdr sent_msg = {};
+  iovec sent_iov = {};
+  char sent_data[kDataLength];
+  sent_iov.iov_base = sent_data;
+  sent_iov.iov_len = kDataLength;
+  sent_msg.msg_iov = &sent_iov;
+  sent_msg.msg_iovlen = 1;
+  sent_msg.msg_flags = 0;
+
+  ASSERT_THAT(RetryEINTR(sendmsg)(sender->get(), &sent_msg, 0),
+              SyscallSucceedsWithValue(kDataLength));
+
+  msghdr received_msg = {};
+  iovec received_iov = {};
+  char received_data[kDataLength];
+  char received_cmsg_buf[CMSG_SPACE(sizeof(in_pktinfo))] = {};
+  size_t cmsg_data_len = sizeof(in_pktinfo);
+  received_iov.iov_base = received_data;
+  received_iov.iov_len = kDataLength;
+  received_msg.msg_iov = &received_iov;
+  received_msg.msg_iovlen = 1;
+  received_msg.msg_controllen = CMSG_LEN(cmsg_data_len);
+  received_msg.msg_control = received_cmsg_buf;
+
+  ASSERT_THAT(RetryEINTR(recvmsg)(receiver->get(), &received_msg, 0),
+              SyscallSucceedsWithValue(kDataLength));
+
+  cmsghdr* cmsg = CMSG_FIRSTHDR(&received_msg);
+  ASSERT_NE(cmsg, nullptr);
+  EXPECT_EQ(cmsg->cmsg_len, CMSG_LEN(cmsg_data_len));
+  EXPECT_EQ(cmsg->cmsg_level, level);
+  EXPECT_EQ(cmsg->cmsg_type, type);
+
+  // Get loopback index.
+  ifreq ifr = {};
+  absl::SNPrintF(ifr.ifr_name, IFNAMSIZ, "lo");
+  ASSERT_THAT(ioctl(sender->get(), SIOCGIFINDEX, &ifr), SyscallSucceeds());
+  ASSERT_NE(ifr.ifr_ifindex, 0);
+
+  // Check the data
+  in_pktinfo received_pktinfo = {};
+  memcpy(&received_pktinfo, CMSG_DATA(cmsg), sizeof(in_pktinfo));
+  EXPECT_EQ(received_pktinfo.ipi_ifindex, ifr.ifr_ifindex);
+  EXPECT_EQ(received_pktinfo.ipi_spec_dst.s_addr, htonl(INADDR_LOOPBACK));
+  EXPECT_EQ(received_pktinfo.ipi_addr.s_addr, htonl(INADDR_LOOPBACK));
+}
 }  // namespace testing
 }  // namespace gvisor

test/syscalls/linux/udp_socket_test_cases.cc

@@ -1495,6 +1495,5 @@ TEST_P(UdpSocketTest, SendAndReceiveTOS) {
   memcpy(&received_tos, CMSG_DATA(cmsg), sizeof(received_tos));
   EXPECT_EQ(received_tos, sent_tos);
 }
-
 }  // namespace testing
 }  // namespace gvisor