Add support to return protocol in recvmsg for AF_PACKET.

Updates #173

PiperOrigin-RevId: 321690756

5 files changed, 58 insertions, 5 deletions

diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go
index 0b1be1bd2..49a04e613 100644
--- a/pkg/sentry/socket/netstack/netstack.go
+++ b/pkg/sentry/socket/netstack/netstack.go
@@ -297,8 +297,9 @@ type socketOpsCommon struct {
 	readView buffer.View
 	// readCM holds control message information for the last packet read
 	// from Endpoint.
-	readCM tcpip.ControlMessages
-	sender tcpip.FullAddress
+	readCM         tcpip.ControlMessages
+	sender         tcpip.FullAddress
+	linkPacketInfo tcpip.LinkPacketInfo
 
 	// sockOptTimestamp corresponds to SO_TIMESTAMP. When true, timestamps
 	// of returned messages can be returned via control messages. When
@@ -447,8 +448,21 @@ func (s *socketOpsCommon) fetchReadView() *syserr.Error {
 	}
 	s.readView = nil
 	s.sender = tcpip.FullAddress{}
+	s.linkPacketInfo = tcpip.LinkPacketInfo{}
 
-	v, cms, err := s.Endpoint.Read(&s.sender)
+	var v buffer.View
+	var cms tcpip.ControlMessages
+	var err *tcpip.Error
+
+	switch e := s.Endpoint.(type) {
+	// The ordering of these interfaces matters. The most specific
+	// interfaces must be specified before the more generic Endpoint
+	// interface.
+	case tcpip.PacketEndpoint:
+		v, cms, err = e.ReadPacket(&s.sender, &s.linkPacketInfo)
+	case tcpip.Endpoint:
+		v, cms, err = e.Read(&s.sender)
+	}
 	if err != nil {
 		atomic.StoreUint32(&s.readViewHasData, 0)
 		return syserr.TranslateNetstackError(err)
@@ -2509,6 +2523,10 @@ func (s *socketOpsCommon) nonBlockingRead(ctx context.Context, dst usermem.IOSeq
 	var addrLen uint32
 	if isPacket && senderRequested {
 		addr, addrLen = ConvertAddress(s.family, s.sender)
+		switch v := addr.(type) {
+		case *linux.SockAddrLink:
+			v.Protocol = htons(uint16(s.linkPacketInfo.Protocol))
+		}
 	}
 
 	if peek {
diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go
index 71bcee785..48ad56d4d 100644
--- a/pkg/tcpip/tcpip.go
+++ b/pkg/tcpip/tcpip.go
@@ -549,6 +549,25 @@ type Endpoint interface {
 	SetOwner(owner PacketOwner)
 }
 
+// LinkPacketInfo holds Link layer information for a received packet.
+//
+// +stateify savable
+type LinkPacketInfo struct {
+	// Protocol is the NetworkProtocolNumber for the packet.
+	Protocol NetworkProtocolNumber
+}
+
+// PacketEndpoint are additional methods that are only implemented by Packet
+// endpoints.
+type PacketEndpoint interface {
+	// ReadPacket reads a datagram/packet from the endpoint and optionally
+	// returns the sender and additional LinkPacketInfo.
+	//
+	// This method does not block if there is no data pending. It will also
+	// either return an error or data, never both.
+	ReadPacket(*FullAddress, *LinkPacketInfo) (buffer.View, ControlMessages, *Error)
+}
+
 // EndpointInfo is the interface implemented by each endpoint info struct.
 type EndpointInfo interface {
 	// IsEndpointInfo is an empty method to implement the tcpip.EndpointInfo
diff --git a/pkg/tcpip/transport/packet/endpoint.go b/pkg/tcpip/transport/packet/endpoint.go
index 92b487381..7b2083a09 100644
--- a/pkg/tcpip/transport/packet/endpoint.go
+++ b/pkg/tcpip/transport/packet/endpoint.go
@@ -45,6 +45,9 @@ type packet struct {
 	timestampNS int64
 	// senderAddr is the network address of the sender.
 	senderAddr tcpip.FullAddress
+	// packetInfo holds additional information like the protocol
+	// of the packet etc.
+	packetInfo tcpip.LinkPacketInfo
 }
 
 // endpoint is the packet socket implementation of tcpip.Endpoint. It is legal
@@ -151,8 +154,8 @@ func (ep *endpoint) Close() {
 // ModerateRecvBuf implements tcpip.Endpoint.ModerateRecvBuf.
 func (ep *endpoint) ModerateRecvBuf(copied int) {}
 
-// Read implements tcpip.Endpoint.Read.
-func (ep *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, *tcpip.Error) {
+// Read implements tcpip.PacketEndpoint.ReadPacket.
+func (ep *endpoint) ReadPacket(addr *tcpip.FullAddress, info *tcpip.LinkPacketInfo) (buffer.View, tcpip.ControlMessages, *tcpip.Error) {
 	ep.rcvMu.Lock()
 
 	// If there's no data to read, return that read would block or that the
@@ -177,9 +180,18 @@ func (ep *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMes
 		*addr = packet.senderAddr
 	}
 
+	if info != nil {
+		*info = packet.packetInfo
+	}
+
 	return packet.data.ToView(), tcpip.ControlMessages{HasTimestamp: true, Timestamp: packet.timestampNS}, nil
 }
 
+// Read implements tcpip.Endpoint.Read.
+func (ep *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, *tcpip.Error) {
+	return ep.ReadPacket(addr, nil)
+}
+
 func (ep *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) {
 	// TODO(b/129292371): Implement.
 	return 0, nil, tcpip.ErrInvalidOptionValue
@@ -428,12 +440,14 @@ func (ep *endpoint) HandlePacket(nicID tcpip.NICID, localAddr tcpip.LinkAddress,
 			NIC:  nicID,
 			Addr: tcpip.Address(hdr.SourceAddress()),
 		}
+		packet.packetInfo.Protocol = netProto
 	} else {
 		// Guess the would-be ethernet header.
 		packet.senderAddr = tcpip.FullAddress{
 			NIC:  nicID,
 			Addr: tcpip.Address(localAddr),
 		}
+		packet.packetInfo.Protocol = netProto
 	}
 
 	if ep.cooked {
diff --git a/test/syscalls/linux/packet_socket.cc b/test/syscalls/linux/packet_socket.cc
index ce63adb23..e94ddcb77 100644
--- a/test/syscalls/linux/packet_socket.cc
+++ b/test/syscalls/linux/packet_socket.cc
@@ -193,6 +193,7 @@ void ReceiveMessage(int sock, int ifindex) {
   EXPECT_EQ(src.sll_family, AF_PACKET);
   EXPECT_EQ(src.sll_ifindex, ifindex);
   EXPECT_EQ(src.sll_halen, ETH_ALEN);
+  EXPECT_EQ(ntohs(src.sll_protocol), ETH_P_IP);
   // This came from the loopback device, so the address is all 0s.
   for (int i = 0; i < src.sll_halen; i++) {
     EXPECT_EQ(src.sll_addr[i], 0);
diff --git a/test/syscalls/linux/packet_socket_raw.cc b/test/syscalls/linux/packet_socket_raw.cc
index e44062475..2fca9fe4d 100644
--- a/test/syscalls/linux/packet_socket_raw.cc
+++ b/test/syscalls/linux/packet_socket_raw.cc
@@ -200,6 +200,7 @@ TEST_P(RawPacketTest, Receive) {
   EXPECT_EQ(src.sll_family, AF_PACKET);
   EXPECT_EQ(src.sll_ifindex, GetLoopbackIndex());
   EXPECT_EQ(src.sll_halen, ETH_ALEN);
+  EXPECT_EQ(ntohs(src.sll_protocol), ETH_P_IP);
   // This came from the loopback device, so the address is all 0s.
   for (int i = 0; i < src.sll_halen; i++) {
     EXPECT_EQ(src.sll_addr[i], 0);