netstack: parse incoming packet headers up-front

Netstack has traditionally parsed headers on-demand as a packet moves up the
stack. This is conceptually simple and convenient, but incompatible with
iptables, where headers can be inspected and mangled before even a routing
decision is made.

This changes header parsing to happen early in the incoming packet path, as soon
as the NIC gets the packet from a link endpoint. Even if an invalid packet is
found (e.g. a TCP header of insufficient length), the packet is passed up the
stack for proper stats bookkeeping.

PiperOrigin-RevId: 315179302

1 files changed, 40 insertions, 34 deletions

diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go
index 959f7e007..7e9f16c90 100644
--- a/pkg/tcpip/network/ipv4/ipv4.go
+++ b/pkg/tcpip/network/ipv4/ipv4.go
@@ -21,6 +21,7 @@
 package ipv4
 
 import (
+	"fmt"
 	"sync/atomic"
 
 	"gvisor.dev/gvisor/pkg/tcpip"
@@ -268,14 +269,14 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw
 		ep, err := e.stack.FindNetworkEndpoint(header.IPv4ProtocolNumber, netHeader.DestinationAddress())
 		if err == nil {
 			route := r.ReverseRoute(netHeader.SourceAddress(), netHeader.DestinationAddress())
-			handleLoopback(&route, pkt, ep)
+			ep.HandlePacket(&route, pkt)
 			return nil
 		}
 	}
 
 	if r.Loop&stack.PacketLoop != 0 {
 		loopedR := r.MakeLoopedRoute()
-		handleLoopback(&loopedR, pkt, e)
+		e.HandlePacket(&loopedR, pkt)
 		loopedR.Release()
 	}
 	if r.Loop&stack.PacketOut == 0 {
@@ -291,17 +292,6 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw
 	return nil
 }
 
-func handleLoopback(route *stack.Route, pkt *stack.PacketBuffer, ep stack.NetworkEndpoint) {
-	// The inbound path expects the network header to still be in
-	// the PacketBuffer's Data field.
-	views := make([]buffer.View, 1, 1+len(pkt.Data.Views()))
-	views[0] = pkt.Header.View()
-	views = append(views, pkt.Data.Views()...)
-	ep.HandlePacket(route, &stack.PacketBuffer{
-		Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views),
-	})
-}
-
 // WritePackets implements stack.NetworkEndpoint.WritePackets.
 func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.PacketBufferList, params stack.NetworkHeaderParams) (int, *tcpip.Error) {
 	if r.Loop&stack.PacketLoop != 0 {
@@ -339,12 +329,11 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe
 		}
 		if _, ok := natPkts[pkt]; ok {
 			netHeader := header.IPv4(pkt.NetworkHeader)
-			ep, err := e.stack.FindNetworkEndpoint(header.IPv4ProtocolNumber, netHeader.DestinationAddress())
-			if err == nil {
+			if ep, err := e.stack.FindNetworkEndpoint(header.IPv4ProtocolNumber, netHeader.DestinationAddress()); err == nil {
 				src := netHeader.SourceAddress()
 				dst := netHeader.DestinationAddress()
 				route := r.ReverseRoute(src, dst)
-				handleLoopback(&route, pkt, ep)
+				ep.HandlePacket(&route, pkt)
 				n++
 				continue
 			}
@@ -418,22 +407,11 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt *stack.PacketBu
 // HandlePacket is called by the link layer when new ipv4 packets arrive for
 // this endpoint.
 func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) {
-	headerView, ok := pkt.Data.PullUp(header.IPv4MinimumSize)
-	if !ok {
+	h := header.IPv4(pkt.NetworkHeader)
+	if !h.IsValid(pkt.Data.Size() + len(pkt.NetworkHeader) + len(pkt.TransportHeader)) {
 		r.Stats().IP.MalformedPacketsReceived.Increment()
 		return
 	}
-	h := header.IPv4(headerView)
-	if !h.IsValid(pkt.Data.Size()) {
-		r.Stats().IP.MalformedPacketsReceived.Increment()
-		return
-	}
-	pkt.NetworkHeader = headerView[:h.HeaderLength()]
-
-	hlen := int(h.HeaderLength())
-	tlen := int(h.TotalLength())
-	pkt.Data.TrimFront(hlen)
-	pkt.Data.CapLength(tlen - hlen)
 
 	// iptables filtering. All packets that reach here are intended for
 	// this machine and will not be forwarded.
@@ -443,9 +421,8 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) {
 		return
 	}
 
-	more := (h.Flags() & header.IPv4FlagMoreFragments) != 0
-	if more || h.FragmentOffset() != 0 {
-		if pkt.Data.Size() == 0 {
+	if h.More() || h.FragmentOffset() != 0 {
+		if pkt.Data.Size()+len(pkt.TransportHeader) == 0 {
 			// Drop the packet as it's marked as a fragment but has
 			// no payload.
 			r.Stats().IP.MalformedPacketsReceived.Increment()
@@ -464,7 +441,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) {
 		}
 		var ready bool
 		var err error
-		pkt.Data, ready, err = e.fragmentation.Process(hash.IPv4FragmentHash(h), h.FragmentOffset(), last, more, pkt.Data)
+		pkt.Data, ready, err = e.fragmentation.Process(hash.IPv4FragmentHash(h), h.FragmentOffset(), last, h.More(), pkt.Data)
 		if err != nil {
 			r.Stats().IP.MalformedPacketsReceived.Increment()
 			r.Stats().IP.MalformedFragmentsReceived.Increment()
@@ -476,7 +453,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) {
 	}
 	p := h.TransportProtocol()
 	if p == header.ICMPv4ProtocolNumber {
-		headerView.CapLength(hlen)
+		pkt.NetworkHeader.CapLength(int(h.HeaderLength()))
 		e.handleICMP(r, pkt)
 		return
 	}
@@ -556,6 +533,35 @@ func (*protocol) Close() {}
 // Wait implements stack.TransportProtocol.Wait.
 func (*protocol) Wait() {}
 
+// Parse implements stack.TransportProtocol.Parse.
+func (*protocol) Parse(pkt *stack.PacketBuffer) (proto tcpip.TransportProtocolNumber, hasTransportHdr bool, ok bool) {
+	hdr, ok := pkt.Data.PullUp(header.IPv4MinimumSize)
+	if !ok {
+		return 0, false, false
+	}
+	ipHdr := header.IPv4(hdr)
+
+	// If there are options, pull those into hdr as well.
+	if headerLen := int(ipHdr.HeaderLength()); headerLen > header.IPv4MinimumSize && headerLen <= pkt.Data.Size() {
+		hdr, ok = pkt.Data.PullUp(headerLen)
+		if !ok {
+			panic(fmt.Sprintf("There are only %d bytes in pkt.Data, but there should be at least %d", pkt.Data.Size(), headerLen))
+		}
+		ipHdr = header.IPv4(hdr)
+	}
+
+	// If this is a fragment, don't bother parsing the transport header.
+	parseTransportHeader := true
+	if ipHdr.More() || ipHdr.FragmentOffset() != 0 {
+		parseTransportHeader = false
+	}
+
+	pkt.NetworkHeader = hdr
+	pkt.Data.TrimFront(len(hdr))
+	pkt.Data.CapLength(int(ipHdr.TotalLength()) - len(hdr))
+	return ipHdr.TransportProtocol(), parseTransportHeader, true
+}
+
 // calculateMTU calculates the network-layer payload MTU based on the link-layer
 // payload mtu.
 func calculateMTU(mtu uint32) uint32 {