Move packet handling to NetworkEndpoint

The NIC should not hold network-layer state or logic - network packet
handling/forwarding should be performed at the network layer instead
of the NIC.

Fixes #4688

PiperOrigin-RevId: 342166985

1 files changed, 81 insertions, 21 deletions

diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go
index 0526190cc..38a0633bd 100644
--- a/pkg/tcpip/network/ipv6/ipv6.go
+++ b/pkg/tcpip/network/ipv6/ipv6.go
@@ -441,17 +441,13 @@ func (e *endpoint) handleFragments(r *stack.Route, gso *stack.GSO, networkMTU ui
 // WritePacket writes a packet to the given destination address and protocol.
 func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, pkt *stack.PacketBuffer) *tcpip.Error {
 	e.addIPHeader(r, pkt, params)
-	return e.writePacket(r, gso, pkt, params.Protocol)
-}
 
-func (e *endpoint) writePacket(r *stack.Route, gso *stack.GSO, pkt *stack.PacketBuffer, protocol tcpip.TransportProtocolNumber) *tcpip.Error {
 	// iptables filtering. All packets that reach here are locally
 	// generated.
 	nicName := e.protocol.stack.FindNICNameFromID(e.nic.ID())
-	ipt := e.protocol.stack.IPTables()
-	if ok := ipt.Check(stack.Output, pkt, gso, r, "", nicName); !ok {
+	if ok := e.protocol.stack.IPTables().Check(stack.Output, pkt, gso, r, "", nicName); !ok {
 		// iptables is telling us to drop the packet.
-		r.Stats().IP.IPTablesOutputDropped.Increment()
+		e.protocol.stack.Stats().IP.IPTablesOutputDropped.Increment()
 		return nil
 	}
 
@@ -467,24 +463,27 @@ func (e *endpoint) writePacket(r *stack.Route, gso *stack.GSO, pkt *stack.Packet
 		if ep, err := e.protocol.stack.FindNetworkEndpoint(ProtocolNumber, netHeader.DestinationAddress()); err == nil {
 			pkt := pkt.CloneToInbound()
 			if e.protocol.stack.ParsePacketBuffer(ProtocolNumber, pkt) == stack.ParsedOK {
-				route := r.ReverseRoute(netHeader.SourceAddress(), netHeader.DestinationAddress())
-				route.PopulatePacketInfo(pkt)
 				// Since we rewrote the packet but it is being routed back to us, we can
 				// safely assume the checksum is valid.
 				pkt.RXTransportChecksumValidated = true
-				ep.HandlePacket(pkt)
+				ep.(*endpoint).handlePacket(pkt)
 			}
 			return nil
 		}
 	}
 
+	return e.writePacket(r, gso, pkt, params.Protocol, false /* headerIncluded */)
+}
+
+func (e *endpoint) writePacket(r *stack.Route, gso *stack.GSO, pkt *stack.PacketBuffer, protocol tcpip.TransportProtocolNumber, headerIncluded bool) *tcpip.Error {
 	if r.Loop&stack.PacketLoop != 0 {
 		pkt := pkt.CloneToInbound()
 		if e.protocol.stack.ParsePacketBuffer(ProtocolNumber, pkt) == stack.ParsedOK {
-			loopedR := r.MakeLoopedRoute()
-			loopedR.PopulatePacketInfo(pkt)
-			loopedR.Release()
-			e.HandlePacket(pkt)
+			// If the packet was generated by the stack (not a raw/packet endpoint
+			// where a packet may be written with the header included), then we can
+			// safely assume the checksum is valid.
+			pkt.RXTransportChecksumValidated = !headerIncluded
+			e.handlePacket(pkt)
 		}
 	}
 	if r.Loop&stack.PacketOut == 0 {
@@ -558,8 +557,7 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe
 	// iptables filtering. All packets that reach here are locally
 	// generated.
 	nicName := e.protocol.stack.FindNICNameFromID(e.nic.ID())
-	ipt := e.protocol.stack.IPTables()
-	dropped, natPkts := ipt.CheckPackets(stack.Output, pkts, gso, r, nicName)
+	dropped, natPkts := e.protocol.stack.IPTables().CheckPackets(stack.Output, pkts, gso, r, nicName)
 	if len(dropped) == 0 && len(natPkts) == 0 {
 		// Fast path: If no packets are to be dropped then we can just invoke the
 		// faster WritePackets API directly.
@@ -584,9 +582,10 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe
 			if ep, err := e.protocol.stack.FindNetworkEndpoint(ProtocolNumber, netHeader.DestinationAddress()); err == nil {
 				pkt := pkt.CloneToInbound()
 				if e.protocol.stack.ParsePacketBuffer(ProtocolNumber, pkt) == stack.ParsedOK {
-					route := r.ReverseRoute(netHeader.SourceAddress(), netHeader.DestinationAddress())
-					route.PopulatePacketInfo(pkt)
-					ep.HandlePacket(pkt)
+					// Since we rewrote the packet but it is being routed back to us, we
+					// can safely assume the checksum is valid.
+					pkt.RXTransportChecksumValidated = true
+					ep.(*endpoint).handlePacket(pkt)
 				}
 				n++
 				continue
@@ -640,16 +639,66 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt *stack.PacketBu
 		return tcpip.ErrMalformedHeader
 	}
 
-	return e.writePacket(r, nil /* gso */, pkt, proto)
+	return e.writePacket(r, nil /* gso */, pkt, proto, true /* headerIncluded */)
+}
+
+// forwardPacket attempts to forward a packet to its final destination.
+func (e *endpoint) forwardPacket(pkt *stack.PacketBuffer) *tcpip.Error {
+	h := header.IPv6(pkt.NetworkHeader().View())
+	dstAddr := h.DestinationAddress()
+
+	// Check if the destination is owned by the stack.
+	networkEndpoint, err := e.protocol.stack.FindNetworkEndpoint(ProtocolNumber, dstAddr)
+	if err == nil {
+		networkEndpoint.(*endpoint).handlePacket(pkt)
+		return nil
+	}
+	if err != tcpip.ErrBadAddress {
+		return err
+	}
+
+	r, err := e.protocol.stack.FindRoute(0, "", dstAddr, ProtocolNumber, false /* multicastLoop */)
+	if err != nil {
+		return err
+	}
+	defer r.Release()
+
+	// TODO(b/143425874) Decrease the TTL field in forwarded packets.
+	return r.WriteHeaderIncludedPacket(stack.NewPacketBuffer(stack.PacketBufferOptions{
+		ReserveHeaderBytes: int(r.MaxHeaderLength()),
+		// We need to do a deep copy of the IP packet because
+		// WriteHeaderIncludedPacket takes ownership of the packet buffer, but we do
+		// not own it.
+		Data: stack.PayloadSince(pkt.NetworkHeader()).ToVectorisedView(),
+	}))
 }
 
 // HandlePacket is called by the link layer when new ipv6 packets arrive for
 // this endpoint.
 func (e *endpoint) HandlePacket(pkt *stack.PacketBuffer) {
+	stats := e.protocol.stack.Stats()
+	stats.IP.PacketsReceived.Increment()
+
 	if !e.isEnabled() {
+		stats.IP.DisabledPacketsReceived.Increment()
 		return
 	}
 
+	// Loopback traffic skips the prerouting chain.
+	if !e.nic.IsLoopback() {
+		if ok := e.protocol.stack.IPTables().Check(stack.Prerouting, pkt, nil, nil, e.MainAddress().Address, ""); !ok {
+			// iptables is telling us to drop the packet.
+			stats.IP.IPTablesPreroutingDropped.Increment()
+			return
+		}
+	}
+
+	e.handlePacket(pkt)
+}
+
+// handlePacket is like HandlePacket except it does not perform the prerouting
+// iptables hook.
+func (e *endpoint) handlePacket(pkt *stack.PacketBuffer) {
 	pkt.NICID = e.nic.ID()
 	stats := e.protocol.stack.Stats()
 
@@ -669,6 +718,18 @@ func (e *endpoint) HandlePacket(pkt *stack.PacketBuffer) {
 		return
 	}
 
+	addressEndpoint := e.AcquireAssignedAddress(dstAddr, e.nic.Promiscuous(), stack.CanBePrimaryEndpoint)
+	if addressEndpoint == nil {
+		if !e.protocol.Forwarding() {
+			stats.IP.InvalidDestinationAddressesReceived.Increment()
+			return
+		}
+
+		_ = e.forwardPacket(pkt)
+		return
+	}
+	addressEndpoint.DecRef()
+
 	// vv consists of:
 	// - Any IPv6 header bytes after the first 40 (i.e. extensions).
 	// - The transport header, if present.
@@ -681,8 +742,7 @@ func (e *endpoint) HandlePacket(pkt *stack.PacketBuffer) {
 
 	// iptables filtering. All packets that reach here are intended for
 	// this machine and need not be forwarded.
-	ipt := e.protocol.stack.IPTables()
-	if ok := ipt.Check(stack.Input, pkt, nil, nil, "", ""); !ok {
+	if ok := e.protocol.stack.IPTables().Check(stack.Input, pkt, nil, nil, "", ""); !ok {
 		// iptables is telling us to drop the packet.
 		stats.IP.IPTablesInputDropped.Increment()
 		return