UDP and TCP raw socket support.

PiperOrigin-RevId: 249511348
Change-Id: I34539092cc85032d9473ff4dd308fc29dc9bfd6b

9 files changed, 48 insertions, 17 deletions

diff --git a/pkg/sentry/socket/epsocket/provider.go b/pkg/sentry/socket/epsocket/provider.go
index 5a89a63fb..fb1815c2d 100644
--- a/pkg/sentry/socket/epsocket/provider.go
+++ b/pkg/sentry/socket/epsocket/provider.go
@@ -70,6 +70,10 @@ func getTransportProtocol(ctx context.Context, stype transport.SockType, protoco
 		switch protocol {
 		case syscall.IPPROTO_ICMP:
 			return header.ICMPv4ProtocolNumber, nil
+		case syscall.IPPROTO_UDP:
+			return header.UDPProtocolNumber, nil
+		case syscall.IPPROTO_TCP:
+			return header.TCPProtocolNumber, nil
 		}
 	}
 	return 0, syserr.ErrInvalidArgument
diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go
index a4117d98e..50d35de88 100644
--- a/pkg/tcpip/stack/nic.go
+++ b/pkg/tcpip/stack/nic.go
@@ -593,6 +593,14 @@ func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolN
 	}
 
 	transProto := state.proto
+
+	// Raw socket packets are delivered based solely on the transport
+	// protocol number. We do not inspect the payload to ensure it's
+	// validly formed.
+	if !n.demux.deliverRawPacket(r, protocol, netHeader, vv) {
+		n.stack.demux.deliverRawPacket(r, protocol, netHeader, vv)
+	}
+
 	if len(vv.First()) < transProto.MinimumPacketSize() {
 		n.stack.stats.MalformedRcvdPackets.Increment()
 		return
diff --git a/pkg/tcpip/stack/transport_demuxer.go b/pkg/tcpip/stack/transport_demuxer.go
index 807c3ba5e..605bfadeb 100644
--- a/pkg/tcpip/stack/transport_demuxer.go
+++ b/pkg/tcpip/stack/transport_demuxer.go
@@ -286,20 +286,10 @@ func (d *transportDemuxer) deliverPacket(r *Route, protocol tcpip.TransportProto
 		destEps = append(destEps, ep)
 	}
 
-	// As in net/ipv4/ip_input.c:ip_local_deliver, attempt to deliver via
-	// raw endpoint first. If there are multipe raw endpoints, they all
-	// receive the packet.
-	foundRaw := false
-	for _, rawEP := range eps.rawEndpoints {
-		// Each endpoint gets its own copy of the packet for the sake
-		// of save/restore.
-		rawEP.HandlePacket(r, buffer.NewViewFromBytes(netHeader), vv.ToView().ToVectorisedView())
-		foundRaw = true
-	}
 	eps.mu.RUnlock()
 
 	// Fail if we didn't find at least one matching transport endpoint.
-	if len(destEps) == 0 && !foundRaw {
+	if len(destEps) == 0 {
 		// UDP packet could not be delivered to an unknown destination port.
 		if protocol == header.UDPProtocolNumber {
 			r.Stats().UDP.UnknownPortErrors.Increment()
@@ -315,6 +305,30 @@ func (d *transportDemuxer) deliverPacket(r *Route, protocol tcpip.TransportProto
 	return true
 }
 
+// deliverRawPacket attempts to deliver the given packet and returns whether it
+// was delivered successfully.
+func (d *transportDemuxer) deliverRawPacket(r *Route, protocol tcpip.TransportProtocolNumber, netHeader buffer.View, vv buffer.VectorisedView) bool {
+	eps, ok := d.protocol[protocolIDs{r.NetProto, protocol}]
+	if !ok {
+		return false
+	}
+
+	// As in net/ipv4/ip_input.c:ip_local_deliver, attempt to deliver via
+	// raw endpoint first. If there are multiple raw endpoints, they all
+	// receive the packet.
+	foundRaw := false
+	eps.mu.RLock()
+	for _, rawEP := range eps.rawEndpoints {
+		// Each endpoint gets its own copy of the packet for the sake
+		// of save/restore.
+		rawEP.HandlePacket(r, buffer.NewViewFromBytes(netHeader), vv.ToView().ToVectorisedView())
+		foundRaw = true
+	}
+	eps.mu.RUnlock()
+
+	return foundRaw
+}
+
 // deliverControlPacket attempts to deliver the given control packet. Returns
 // true if it found an endpoint, false otherwise.
 func (d *transportDemuxer) deliverControlPacket(net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, vv buffer.VectorisedView, id TransportEndpointID) bool {
diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go
index 1a16a3607..e7b383ad5 100644
--- a/pkg/tcpip/transport/raw/endpoint.go
+++ b/pkg/tcpip/transport/raw/endpoint.go
@@ -280,7 +280,7 @@ func (ep *endpoint) finishWrite(payload tcpip.Payload, route *stack.Route) (uint
 	switch ep.netProto {
 	case header.IPv4ProtocolNumber:
 		hdr := buffer.NewPrependable(len(payloadBytes) + int(route.MaxHeaderLength()))
-		if err := route.WritePacket(nil /* gso */, hdr, buffer.View(payloadBytes).ToVectorisedView(), header.ICMPv4ProtocolNumber, route.DefaultTTL()); err != nil {
+		if err := route.WritePacket(nil /* gso */, hdr, buffer.View(payloadBytes).ToVectorisedView(), ep.transProto, route.DefaultTTL()); err != nil {
 			return 0, nil, err
 		}
 
diff --git a/pkg/tcpip/transport/tcp/BUILD b/pkg/tcpip/transport/tcp/BUILD
index d44d63e95..e31b03f7d 100644
--- a/pkg/tcpip/transport/tcp/BUILD
+++ b/pkg/tcpip/transport/tcp/BUILD
@@ -49,6 +49,7 @@ go_library(
         "//pkg/tcpip/header",
         "//pkg/tcpip/seqnum",
         "//pkg/tcpip/stack",
+        "//pkg/tcpip/transport/raw",
         "//pkg/tmutex",
         "//pkg/waiter",
         "@com_github_google_btree//:go_default_library",
diff --git a/pkg/tcpip/transport/tcp/protocol.go b/pkg/tcpip/transport/tcp/protocol.go
index b86473891..d31a1edcb 100644
--- a/pkg/tcpip/transport/tcp/protocol.go
+++ b/pkg/tcpip/transport/tcp/protocol.go
@@ -29,6 +29,7 @@ import (
 	"gvisor.googlesource.com/gvisor/pkg/tcpip/header"
 	"gvisor.googlesource.com/gvisor/pkg/tcpip/seqnum"
 	"gvisor.googlesource.com/gvisor/pkg/tcpip/stack"
+	"gvisor.googlesource.com/gvisor/pkg/tcpip/transport/raw"
 	"gvisor.googlesource.com/gvisor/pkg/waiter"
 )
 
@@ -104,7 +105,7 @@ func (*protocol) NewEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolN
 // NewRawEndpoint creates a new raw TCP endpoint. Raw TCP sockets are currently
 // unsupported. It implements stack.TransportProtocol.NewRawEndpoint.
 func (p *protocol) NewRawEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) {
-	return nil, tcpip.ErrUnknownProtocol
+	return raw.NewEndpoint(stack, netProto, header.TCPProtocolNumber, waiterQueue)
 }
 
 // MinimumPacketSize returns the minimum valid tcp packet size.
diff --git a/pkg/tcpip/transport/udp/BUILD b/pkg/tcpip/transport/udp/BUILD
index 361132a25..b9520d6e0 100644
--- a/pkg/tcpip/transport/udp/BUILD
+++ b/pkg/tcpip/transport/udp/BUILD
@@ -28,12 +28,12 @@ go_library(
     imports = ["gvisor.googlesource.com/gvisor/pkg/tcpip/buffer"],
     visibility = ["//visibility:public"],
     deps = [
-        "//pkg/log",
         "//pkg/sleep",
         "//pkg/tcpip",
         "//pkg/tcpip/buffer",
         "//pkg/tcpip/header",
         "//pkg/tcpip/stack",
+        "//pkg/tcpip/transport/raw",
         "//pkg/waiter",
     ],
 )
diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go
index 0ed0902b0..d9ca097c9 100644
--- a/pkg/tcpip/transport/udp/endpoint.go
+++ b/pkg/tcpip/transport/udp/endpoint.go
@@ -51,6 +51,8 @@ const (
 // have concurrent goroutines make calls into the endpoint, they are properly
 // synchronized.
 //
+// It implements tcpip.Endpoint.
+//
 // +stateify savable
 type endpoint struct {
 	// The following fields are initialized at creation time and do not
diff --git a/pkg/tcpip/transport/udp/protocol.go b/pkg/tcpip/transport/udp/protocol.go
index 8b47cce17..3d31dfbf1 100644
--- a/pkg/tcpip/transport/udp/protocol.go
+++ b/pkg/tcpip/transport/udp/protocol.go
@@ -25,6 +25,7 @@ import (
 	"gvisor.googlesource.com/gvisor/pkg/tcpip/buffer"
 	"gvisor.googlesource.com/gvisor/pkg/tcpip/header"
 	"gvisor.googlesource.com/gvisor/pkg/tcpip/stack"
+	"gvisor.googlesource.com/gvisor/pkg/tcpip/transport/raw"
 	"gvisor.googlesource.com/gvisor/pkg/waiter"
 )
 
@@ -48,10 +49,10 @@ func (*protocol) NewEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolN
 	return newEndpoint(stack, netProto, waiterQueue), nil
 }
 
-// NewRawEndpoint creates a new raw UDP endpoint. Raw UDP sockets are currently
-// unsupported. It implements stack.TransportProtocol.NewRawEndpoint.
+// NewRawEndpoint creates a new raw UDP endpoint. It implements
+// stack.TransportProtocol.NewRawEndpoint.
 func (p *protocol) NewRawEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) {
-	return nil, tcpip.ErrUnknownProtocol
+	return raw.NewEndpoint(stack, netProto, header.UDPProtocolNumber, waiterQueue)
 }
 
 // MinimumPacketSize returns the minimum valid udp packet size.