2 files changed, 182 insertions, 2 deletions
diff --git a/pkg/tcpip/network/ipv4/icmp.go b/pkg/tcpip/network/ipv4/icmp.go
index b5659a36b..5fe73315f 100644
--- a/pkg/tcpip/network/ipv4/icmp.go
+++ b/pkg/tcpip/network/ipv4/icmp.go
@@ -15,6 +15,7 @@
 package ipv4
 
 import (
+	"gvisor.dev/gvisor/pkg/tcpip"
 	"gvisor.dev/gvisor/pkg/tcpip/buffer"
 	"gvisor.dev/gvisor/pkg/tcpip/header"
 	"gvisor.dev/gvisor/pkg/tcpip/stack"
@@ -105,7 +106,7 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer) {
 		// source address MUST be one of its own IP addresses (but not a broadcast
 		// or multicast address).
 		localAddr := r.LocalAddress
-		if r.IsInboundBroadcast() || header.IsV4MulticastAddress(r.LocalAddress) {
+		if r.IsInboundBroadcast() || header.IsV4MulticastAddress(localAddr) {
 			localAddr = ""
 		}
 
@@ -131,7 +132,10 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer) {
 			ReserveHeaderBytes: int(r.MaxHeaderLength()),
 			Data:               dataVV,
 		})
-
+		// TODO(gvisor.dev/issue/3810): When adding protocol numbers into the header
+		// information we will have to change this code to handle the ICMP header
+		// no longer being in the data buffer.
+		replyPkt.TransportProtocolNumber = header.ICMPv4ProtocolNumber
 		// Send out the reply packet.
 		sent := stats.ICMP.V4PacketsSent
 		if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{
@@ -193,3 +197,175 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer) {
 		received.Invalid.Increment()
 	}
 }
+
+// ======= ICMP Error packet generation =========
+
+// ReturnError implements stack.TransportProtocol.ReturnError.
+func (p *protocol) ReturnError(r *stack.Route, reason tcpip.ICMPReason, pkt *stack.PacketBuffer) *tcpip.Error {
+	switch reason.(type) {
+	case *tcpip.ICMPReasonPortUnreachable:
+		return returnError(r, &icmpReasonPortUnreachable{}, pkt)
+	default:
+		return tcpip.ErrNotSupported
+	}
+}
+
+// icmpReason is a marker interface for IPv4 specific ICMP errors.
+type icmpReason interface {
+	isICMPReason()
+}
+
+// icmpReasonPortUnreachable is an error where the transport protocol has no
+// listener and no alternative means to inform the sender.
+type icmpReasonPortUnreachable struct{}
+
+func (*icmpReasonPortUnreachable) isICMPReason() {}
+
+// returnError takes an error descriptor and generates the appropriate ICMP
+// error packet for IPv4 and sends it back to the remote device that sent
+// the problematic packet. It incorporates as much of that packet as
+// possible as well as any error metadata as is available. returnError
+// expects pkt to hold a valid IPv4 packet as per the wire format.
+func returnError(r *stack.Route, reason icmpReason, pkt *stack.PacketBuffer) *tcpip.Error {
+	sent := r.Stats().ICMP.V4PacketsSent
+	if !r.Stack().AllowICMPMessage() {
+		sent.RateLimited.Increment()
+		return nil
+	}
+
+	// We check we are responding only when we are allowed to.
+	// See RFC 1812 section 4.3.2.7 (shown below).
+	//
+	// =========
+	// 4.3.2.7 When Not to Send ICMP Errors
+	//
+	//  An ICMP error message MUST NOT be sent as the result of receiving:
+	//
+	//  o An ICMP error message, or
+	//
+	//  o A packet which fails the IP header validation tests described in
+	//    Section [5.2.2] (except where that section specifically permits
+	//    the sending of an ICMP error message), or
+	//
+	//  o A packet destined to an IP broadcast or IP multicast address, or
+	//
+	//  o A packet sent as a Link Layer broadcast or multicast, or
+	//
+	//  o Any fragment of a datagram other then the first fragment (i.e., a
+	// packet for which the fragment offset in the IP header is nonzero).
+	//
+	// TODO(gvisor.dev/issues/4058): Make sure we don't send ICMP errors in
+	// response to a non-initial fragment, but it currently can not happen.
+
+	if r.IsInboundBroadcast() || header.IsV4MulticastAddress(r.LocalAddress) || r.RemoteAddress == header.IPv4Any {
+		return nil
+	}
+
+	networkHeader := pkt.NetworkHeader().View()
+	transportHeader := pkt.TransportHeader().View()
+
+	// Don't respond to icmp error packets.
+	if header.IPv4(networkHeader).Protocol() == uint8(header.ICMPv4ProtocolNumber) {
+		// TODO(gvisor.dev/issue/3810):
+		// Unfortunately the current stack pretty much always has ICMPv4 headers
+		// in the Data section of the packet but there is no guarantee that is the
+		// case. If this is the case grab the header to make it like all other
+		// packet types. When this is cleaned up the Consume should be removed.
+		if transportHeader.IsEmpty() {
+			var ok bool
+			transportHeader, ok = pkt.TransportHeader().Consume(header.ICMPv4MinimumSize)
+			if !ok {
+				return nil
+			}
+		} else if transportHeader.Size() < header.ICMPv4MinimumSize {
+			return nil
+		}
+		// We need to decide to explicitly name the packets we can respond to or
+		// the ones we can not respond to. The decision is somewhat arbitrary and
+		// if problems arise this could be reversed. It was judged less of a breach
+		// of protocol to not respond to unknown non-error packets than to respond
+		// to unknown error packets so we take the first approach.
+		switch header.ICMPv4(transportHeader).Type() {
+		case
+			header.ICMPv4EchoReply,
+			header.ICMPv4Echo,
+			header.ICMPv4Timestamp,
+			header.ICMPv4TimestampReply,
+			header.ICMPv4InfoRequest,
+			header.ICMPv4InfoReply:
+		default:
+			// Assume any type we don't know about may be an error type.
+			return nil
+		}
+	} else if transportHeader.IsEmpty() {
+		return nil
+	}
+
+	// Now work out how much of the triggering packet we should return.
+	// As per RFC 1812 Section 4.3.2.3
+	//
+	//   ICMP datagram SHOULD contain as much of the original
+	//   datagram as possible without the length of the ICMP
+	//   datagram exceeding 576 bytes.
+	//
+	// NOTE: The above RFC referenced is different from the original
+	// recommendation in RFC 1122 and RFC 792 where it mentioned that at
+	// least 8 bytes of the payload must be included. Today linux and other
+	// systems implement the RFC 1812 definition and not the original
+	// requirement. We treat 8 bytes as the minimum but will try send more.
+	mtu := int(r.MTU())
+	if mtu > header.IPv4MinimumProcessableDatagramSize {
+		mtu = header.IPv4MinimumProcessableDatagramSize
+	}
+	headerLen := int(r.MaxHeaderLength()) + header.ICMPv4MinimumSize
+	available := int(mtu) - headerLen
+
+	if available < header.IPv4MinimumSize+header.ICMPv4MinimumErrorPayloadSize {
+		return nil
+	}
+
+	payloadLen := networkHeader.Size() + transportHeader.Size() + pkt.Data.Size()
+	if payloadLen > available {
+		payloadLen = available
+	}
+
+	// The buffers used by pkt may be used elsewhere in the system.
+	// For example, an AF_RAW or AF_PACKET socket may use what the transport
+	// protocol considers an unreachable destination. Thus we deep copy pkt to
+	// prevent multiple ownership and SR errors. The new copy is a vectorized
+	// view with the entire incoming IP packet reassembled and truncated as
+	// required. This is now the payload of the new ICMP packet and no longer
+	// considered a packet in its own right.
+	newHeader := append(buffer.View(nil), networkHeader...)
+	newHeader = append(newHeader, transportHeader...)
+	payload := newHeader.ToVectorisedView()
+	payload.AppendView(pkt.Data.ToView())
+	payload.CapLength(payloadLen)
+
+	icmpPkt := stack.NewPacketBuffer(stack.PacketBufferOptions{
+		ReserveHeaderBytes: headerLen,
+		Data:               payload,
+	})
+	icmpPkt.TransportProtocolNumber = header.ICMPv4ProtocolNumber
+
+	icmpHdr := header.ICMPv4(icmpPkt.TransportHeader().Push(header.ICMPv4MinimumSize))
+	icmpHdr.SetType(header.ICMPv4DstUnreachable)
+	icmpHdr.SetCode(header.ICMPv4PortUnreachable)
+	counter := sent.DstUnreachable
+	icmpHdr.SetChecksum(header.ICMPv4Checksum(icmpHdr, icmpPkt.Data))
+
+	if err := r.WritePacket(
+		nil, /* gso */
+		stack.NetworkHeaderParams{
+			Protocol: header.ICMPv4ProtocolNumber,
+			TTL:      r.DefaultTTL(),
+			TOS:      stack.DefaultTOS,
+		},
+		icmpPkt,
+	); err != nil {
+		sent.Dropped.Increment()
+		return err
+	}
+	counter.Increment()
+	return nil
+}
diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go
index b14b356d6..135444222 100644
--- a/pkg/tcpip/network/ipv4/ipv4.go
+++ b/pkg/tcpip/network/ipv4/ipv4.go
@@ -455,6 +455,10 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) {
 	}
 	p := h.TransportProtocol()
 	if p == header.ICMPv4ProtocolNumber {
+		// TODO(gvisor.dev/issues/3810): when we sort out ICMP and transport
+		// headers, the setting of the transport number here should be
+		// unnecessary and removed.
+		pkt.TransportProtocolNumber = p
 		e.handleICMP(r, pkt)
 		return
 	}