Extract ICMP error sender from UDP

Store transport protocol number on packet buffers for use in ICMP error
generation.

Updates #2211.

PiperOrigin-RevId: 333252762

pkg/tcpip/header/icmpv4.go

@@ -31,6 +31,27 @@ const (
 	// ICMPv4MinimumSize is the minimum size of a valid ICMP packet.
 	ICMPv4MinimumSize = 8
 
+	// ICMPv4MinimumErrorPayloadSize Is the smallest number of bytes of an
+	// errant packet's transport layer that an ICMP error type packet should
+	// attempt to send as per RFC 792 (see each type) and RFC 1122
+	// section 3.2.2 which states:
+	//      Every ICMP error message includes the Internet header and at
+	//      least the first 8 data octets of the datagram that triggered
+	//      the error; more than 8 octets MAY be sent; this header and data
+	//      MUST be unchanged from the received datagram.
+	//
+	// RFC 792 shows:
+	//   0                   1                   2                   3
+	//  0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+	// +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	// |     Type      |     Code      |          Checksum             |
+	// +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	// |                             unused                            |
+	// +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	// |      Internet Header + 64 bits of Original Data Datagram      |
+	// +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	ICMPv4MinimumErrorPayloadSize = 8
+
 	// ICMPv4ProtocolNumber is the ICMP transport protocol number.
 	ICMPv4ProtocolNumber tcpip.TransportProtocolNumber = 1
 
@@ -39,15 +60,19 @@ const (
 	icmpv4ChecksumOffset = 2
 
 	// icmpv4MTUOffset is the offset of the MTU field
-	// in a ICMPv4FragmentationNeeded message.
+	// in an ICMPv4FragmentationNeeded message.
 	icmpv4MTUOffset = 6
 
 	// icmpv4IdentOffset is the offset of the ident field
-	// in a ICMPv4EchoRequest/Reply message.
+	// in an ICMPv4EchoRequest/Reply message.
 	icmpv4IdentOffset = 4
 
+	// icmpv4PointerOffset is the offset of the pointer field
+	// in an ICMPv4ParamProblem message.
+	icmpv4PointerOffset = 4
+
 	// icmpv4SequenceOffset is the offset of the sequence field
-	// in a ICMPv4EchoRequest/Reply message.
+	// in an ICMPv4EchoRequest/Reply message.
 	icmpv4SequenceOffset = 6
 )
 
@@ -72,15 +97,23 @@ const (
 	ICMPv4InfoReply      ICMPv4Type = 16
 )
 
+// ICMP codes for ICMPv4 Time Exceeded messages as defined in RFC 792.
+const (
+	ICMPv4TTLExceeded ICMPv4Code = 0
+)
+
 // ICMP codes for ICMPv4 Destination Unreachable messages as defined in RFC 792.
 const (
-	ICMPv4TTLExceeded         ICMPv4Code = 0
+	ICMPv4NetUnreachable      ICMPv4Code = 0
 	ICMPv4HostUnreachable     ICMPv4Code = 1
 	ICMPv4ProtoUnreachable    ICMPv4Code = 2
 	ICMPv4PortUnreachable     ICMPv4Code = 3
 	ICMPv4FragmentationNeeded ICMPv4Code = 4
 )
 
+// ICMPv4UnusedCode is a code to use in ICMP messages where no code is needed.
+const ICMPv4UnusedCode ICMPv4Code = 0
+
 // Type is the ICMP type field.
 func (b ICMPv4) Type() ICMPv4Type { return ICMPv4Type(b[0]) }
 
@@ -93,6 +126,15 @@ func (b ICMPv4) Code() ICMPv4Code { return ICMPv4Code(b[1]) }
 // SetCode sets the ICMP code field.
 func (b ICMPv4) SetCode(c ICMPv4Code) { b[1] = byte(c) }
 
+// SetPointer sets the pointer field in a Parameter error packet.
+// This is the first byte of the type specific data field.
+func (b ICMPv4) SetPointer(c byte) { b[icmpv4PointerOffset] = c }
+
+// SetTypeSpecific sets the full 32 bit type specific data field.
+func (b ICMPv4) SetTypeSpecific(val uint32) {
+	binary.BigEndian.PutUint32(b[icmpv4PointerOffset:], val)
+}
+
 // Checksum is the ICMP checksum field.
 func (b ICMPv4) Checksum() uint16 {
 	return binary.BigEndian.Uint16(b[icmpv4ChecksumOffset:])

pkg/tcpip/header/icmpv6.go

@@ -54,9 +54,17 @@ const (
 	// address.
 	ICMPv6NeighborAdvertSize = ICMPv6HeaderSize + NDPNAMinimumSize + NDPLinkLayerAddressSize
 
-	// ICMPv6EchoMinimumSize is the minimum size of a valid ICMP echo packet.
+	// ICMPv6EchoMinimumSize is the minimum size of a valid echo packet.
 	ICMPv6EchoMinimumSize = 8
 
+	// ICMPv6ErrorHeaderSize is the size of an ICMP error packet header,
+	// as per RFC 4443, Apendix A, item 4 and the errata.
+	//   ... all ICMP error messages shall have exactly
+	//   32 bits of type-specific data, so that receivers can reliably find
+	//   the embedded invoking packet even when they don't recognize the
+	//   ICMP message Type.
+	ICMPv6ErrorHeaderSize = 8
+
 	// ICMPv6DstUnreachableMinimumSize is the minimum size of a valid ICMP
 	// destination unreachable packet.
 	ICMPv6DstUnreachableMinimumSize = ICMPv6MinimumSize
@@ -69,6 +77,10 @@ const (
 	// in an ICMPv6 message.
 	icmpv6ChecksumOffset = 2
 
+	// icmpv6PointerOffset is the offset of the pointer
+	// in an ICMPv6 Parameter problem message.
+	icmpv6PointerOffset = 4
+
 	// icmpv6MTUOffset is the offset of the MTU field in an ICMPv6
 	// PacketTooBig message.
 	icmpv6MTUOffset = 4
@@ -89,9 +101,10 @@ const (
 	NDPHopLimit = 255
 )
 
-// ICMPv6Type is the ICMP type field described in RFC 4443 and friends.
+// ICMPv6Type is the ICMP type field described in RFC 4443.
 type ICMPv6Type byte
 
+// Values for use in the Type field of ICMPv6 packet from RFC 4433.
 const (
 	ICMPv6DstUnreachable ICMPv6Type = 1
 	ICMPv6PacketTooBig   ICMPv6Type = 2
@@ -109,7 +122,18 @@ const (
 	ICMPv6RedirectMsg     ICMPv6Type = 137
 )
 
-// ICMPv6Code is the ICMP code field described in RFC 4443.
+// IsErrorType returns true if the receiver is an ICMP error type.
+func (typ ICMPv6Type) IsErrorType() bool {
+	// Per RFC 4443 section 2.1:
+	//   ICMPv6 messages are grouped into two classes: error messages and
+	//   informational messages.  Error messages are identified as such by a
+	//   zero in the high-order bit of their message Type field values.  Thus,
+	//   error messages have message types from 0 to 127; informational
+	//   messages have message types from 128 to 255.
+	return typ&0x80 == 0
+}
+
+// ICMPv6Code is the ICMP Code field described in RFC 4443.
 type ICMPv6Code byte
 
 // ICMP codes used with Destination Unreachable (Type 1). As per RFC 4443
@@ -153,6 +177,11 @@ func (b ICMPv6) Code() ICMPv6Code { return ICMPv6Code(b[1]) }
 // SetCode sets the ICMP code field.
 func (b ICMPv6) SetCode(c ICMPv6Code) { b[1] = byte(c) }
 
+// SetTypeSpecific sets the full 32 bit type specific data field.
+func (b ICMPv6) SetTypeSpecific(val uint32) {
+	binary.BigEndian.PutUint32(b[icmpv6PointerOffset:], val)
+}
+
 // Checksum is the ICMP checksum field.
 func (b ICMPv6) Checksum() uint16 {
 	return binary.BigEndian.Uint16(b[icmpv6ChecksumOffset:])

pkg/tcpip/header/ipv4.go

@@ -80,7 +80,8 @@ type IPv4Fields struct {
 type IPv4 []byte
 
 const (
-	// IPv4MinimumSize is the minimum size of a valid IPv4 packet.
+	// IPv4MinimumSize is the minimum size of a valid IPv4 packet;
+	// i.e. a packet header with no options.
 	IPv4MinimumSize = 20
 
 	// IPv4MaximumHeaderSize is the maximum size of an IPv4 header. Given
@@ -327,7 +328,7 @@ func IsV4MulticastAddress(addr tcpip.Address) bool {
 }
 
 // IsV4LoopbackAddress determines if the provided address is an IPv4 loopback
-// address (belongs to 127.0.0.1/8 subnet).
+// address (belongs to 127.0.0.0/8 subnet). See RFC 1122 section 3.2.1.3.
 func IsV4LoopbackAddress(addr tcpip.Address) bool {
 	if len(addr) != IPv4AddressSize {
 		return false

pkg/tcpip/network/arp/arp.go

@@ -238,6 +238,12 @@ func (*protocol) Parse(pkt *stack.PacketBuffer) (proto tcpip.TransportProtocolNu
 	return 0, false, parse.ARP(pkt)
 }
 
+// ReturnError implements stack.TransportProtocol.ReturnError.
+func (*protocol) ReturnError(*stack.Route, tcpip.ICMPReason, *stack.PacketBuffer) *tcpip.Error {
+	// In ARP, there is no such response so do nothing.
+	return nil
+}
+
 // NewProtocol returns an ARP network protocol.
 func NewProtocol() stack.NetworkProtocol {
 	return &protocol{}

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
+}

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
 	}

pkg/tcpip/network/ipv6/icmp.go

@@ -318,6 +318,7 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme
 			ReserveHeaderBytes: int(r.MaxHeaderLength()) + header.ICMPv6NeighborAdvertMinimumSize + int(optsSerializer.Length()),
 		})
 		packet := header.ICMPv6(pkt.TransportHeader().Push(header.ICMPv6NeighborAdvertSize))
+		pkt.TransportProtocolNumber = header.ICMPv6ProtocolNumber
 		packet.SetType(header.ICMPv6NeighborAdvert)
 		na := header.NDPNeighborAdvert(packet.NDPPayload())
 		na.SetSolicitedFlag(solicited)
@@ -438,6 +439,7 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme
 			Data:               pkt.Data,
 		})
 		packet := header.ICMPv6(replyPkt.TransportHeader().Push(header.ICMPv6EchoMinimumSize))
+		pkt.TransportProtocolNumber = header.ICMPv6ProtocolNumber
 		copy(packet, icmpHdr)
 		packet.SetType(header.ICMPv6EchoReply)
 		packet.SetChecksum(header.ICMPv6Checksum(packet, r.LocalAddress, r.RemoteAddress, pkt.Data))
@@ -637,6 +639,7 @@ func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, remoteLinkAdd
 		ReserveHeaderBytes: int(linkEP.MaxHeaderLength()) + header.IPv6MinimumSize + header.ICMPv6NeighborAdvertSize,
 	})
 	icmpHdr := header.ICMPv6(pkt.TransportHeader().Push(header.ICMPv6NeighborAdvertSize))
+	pkt.TransportProtocolNumber = header.ICMPv6ProtocolNumber
 	icmpHdr.SetType(header.ICMPv6NeighborSolicit)
 	copy(icmpHdr[icmpV6OptOffset-len(addr):], addr)
 	icmpHdr[icmpV6OptOffset] = ndpOptSrcLinkAddr
@@ -665,3 +668,123 @@ func (*protocol) ResolveStaticAddress(addr tcpip.Address) (tcpip.LinkAddress, bo
 	}
 	return tcpip.LinkAddress([]byte(nil)), false
 }
+
+// ======= 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 IPv6 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 IPv6 and sends it.
+func returnError(r *stack.Route, reason icmpReason, pkt *stack.PacketBuffer) *tcpip.Error {
+	stats := r.Stats().ICMP
+	sent := stats.V6PacketsSent
+	if !r.Stack().AllowICMPMessage() {
+		sent.RateLimited.Increment()
+		return nil
+	}
+
+	// Only send ICMP error if the address is not a multicast v6
+	// address and the source is not the unspecified address.
+	//
+	// TODO(b/164522993) There are exceptions to this rule.
+	// See: point e.3) RFC 4443 section-2.4
+	//
+	//	 (e) An ICMPv6 error message MUST NOT be originated as a result of
+	//       receiving the following:
+	//
+	//       (e.1) An ICMPv6 error message.
+	//
+	//       (e.2) An ICMPv6 redirect message [IPv6-DISC].
+	//
+	//       (e.3) A packet destined to an IPv6 multicast address.  (There are
+	//             two exceptions to this rule: (1) the Packet Too Big Message
+	//             (Section 3.2) to allow Path MTU discovery to work for IPv6
+	//             multicast, and (2) the Parameter Problem Message, Code 2
+	//             (Section 3.4) reporting an unrecognized IPv6 option (see
+	//             Section 4.2 of [IPv6]) that has the Option Type highest-
+	//             order two bits set to 10).
+	//
+	if header.IsV6MulticastAddress(r.LocalAddress) || r.RemoteAddress == header.IPv6Any {
+		return nil
+	}
+
+	network, transport := pkt.NetworkHeader().View(), pkt.TransportHeader().View()
+
+	if pkt.TransportProtocolNumber == header.ICMPv6ProtocolNumber {
+		// TODO(gvisor.dev/issues/3810): Sort this out when ICMP headers are stored.
+		// Unfortunately at this time ICMP Packets do not have a transport
+		// header separated out. It is in the Data part so we need to
+		// separate it out now. We will just pretend it is a minimal length
+		// ICMP packet as we don't really care if any later bits of a
+		// larger ICMP packet are in the header view or in the Data view.
+		transport, ok := pkt.TransportHeader().Consume(header.ICMPv6MinimumSize)
+		if !ok {
+			return nil
+		}
+		typ := header.ICMPv6(transport).Type()
+		if typ.IsErrorType() || typ == header.ICMPv6RedirectMsg {
+			return nil
+		}
+	}
+
+	// As per RFC 4443 section 2.4
+	//
+	//    (c) Every ICMPv6 error message (type < 128) MUST include
+	//    as much of the IPv6 offending (invoking) packet (the
+	//    packet that caused the error) as possible without making
+	//    the error message packet exceed the minimum IPv6 MTU
+	//    [IPv6].
+	mtu := int(r.MTU())
+	if mtu > header.IPv6MinimumMTU {
+		mtu = header.IPv6MinimumMTU
+	}
+	headerLen := int(r.MaxHeaderLength()) + header.ICMPv6ErrorHeaderSize
+	available := int(mtu) - headerLen
+	if available < header.IPv6MinimumSize {
+		return nil
+	}
+	payloadLen := network.Size() + transport.Size() + pkt.Data.Size()
+	if payloadLen > available {
+		payloadLen = available
+	}
+	payload := buffer.NewVectorisedView(pkt.Size(), pkt.Views())
+	payload.CapLength(payloadLen)
+
+	newPkt := stack.NewPacketBuffer(stack.PacketBufferOptions{
+		ReserveHeaderBytes: headerLen,
+		Data:               payload,
+	})
+	newPkt.TransportProtocolNumber = header.ICMPv6ProtocolNumber
+
+	icmpHdr := header.ICMPv6(newPkt.TransportHeader().Push(header.ICMPv6DstUnreachableMinimumSize))
+	icmpHdr.SetCode(header.ICMPv6PortUnreachable)
+	icmpHdr.SetType(header.ICMPv6DstUnreachable)
+	icmpHdr.SetChecksum(header.ICMPv6Checksum(icmpHdr, r.LocalAddress, r.RemoteAddress, newPkt.Data))
+	counter := sent.DstUnreachable
+	err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, newPkt)
+	if err != nil {
+		sent.Dropped.Increment()
+		return err
+	}
+	counter.Increment()
+	return nil
+}

pkg/tcpip/network/ipv6/ipv6.go

@@ -348,7 +348,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) {
 					it, done, err := it.Next()
 					if err != nil {
 						r.Stats().IP.MalformedPacketsReceived.Increment()
-						r.Stats().IP.MalformedPacketsReceived.Increment()
+						r.Stats().IP.MalformedFragmentsReceived.Increment()
 						return
 					}
 					if done {
@@ -476,6 +476,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) {
 			pkt.Data = extHdr.Buf
 
 			if p := tcpip.TransportProtocolNumber(extHdr.Identifier); p == header.ICMPv6ProtocolNumber {
+				pkt.TransportProtocolNumber = p
 				e.handleICMP(r, pkt, hasFragmentHeader)
 			} else {
 				r.Stats().IP.PacketsDelivered.Increment()

pkg/tcpip/stack/forwarder_test.go

@@ -145,6 +145,10 @@ func (*fwdTestNetworkProtocol) Parse(pkt *PacketBuffer) (tcpip.TransportProtocol
 	return tcpip.TransportProtocolNumber(netHeader[protocolNumberOffset]), true, true
 }
 
+func (*fwdTestNetworkProtocol) ReturnError(*Route, tcpip.ICMPReason, *PacketBuffer) *tcpip.Error {
+	return nil
+}
+
 func (f *fwdTestNetworkProtocol) NewEndpoint(nicID tcpip.NICID, _ LinkAddressCache, _ NUDHandler, dispatcher TransportDispatcher, ep LinkEndpoint, _ *Stack) NetworkEndpoint {
 	return &fwdTestNetworkEndpoint{
 		nicID:      nicID,

pkg/tcpip/stack/nic.go

@@ -1242,9 +1242,9 @@ func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcp
 		local = n.linkEP.LinkAddress()
 	}
 
-	// Are any packet sockets listening for this network protocol?
+	// Are any packet type sockets listening for this network protocol?
 	packetEPs := n.mu.packetEPs[protocol]
-	// Add any other packet sockets that maybe listening for all protocols.
+	// Add any other packet type sockets that may be listening for all protocols.
 	packetEPs = append(packetEPs, n.mu.packetEPs[header.EthernetProtocolAll]...)
 	n.mu.RUnlock()
 	for _, ep := range packetEPs {
@@ -1265,6 +1265,7 @@ func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcp
 		return
 	}
 	if hasTransportHdr {
+		pkt.TransportProtocolNumber = transProtoNum
 		// Parse the transport header if present.
 		if state, ok := n.stack.transportProtocols[transProtoNum]; ok {
 			state.proto.Parse(pkt)
@@ -1453,10 +1454,28 @@ func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolN
 		}
 	}
 
-	// We could not find an appropriate destination for this packet, so
-	// deliver it to the global handler.
-	if !transProto.HandleUnknownDestinationPacket(r, id, pkt) {
+	// We could not find an appropriate destination for this packet so
+	// give the protocol specific error handler a chance to handle it.
+	// If it doesn't handle it then we should do so.
+	switch transProto.HandleUnknownDestinationPacket(r, id, pkt) {
+	case UnknownDestinationPacketMalformed:
 		n.stack.stats.MalformedRcvdPackets.Increment()
+	case UnknownDestinationPacketUnhandled:
+		// As per RFC: 1122 Section 3.2.2.1 A host SHOULD generate Destination
+		//   Unreachable messages with code:
+		//     3 (Port Unreachable), when the designated transport protocol
+		//     (e.g., UDP) is unable to demultiplex the datagram but has no
+		//     protocol mechanism to inform the sender.
+		np, ok := n.stack.networkProtocols[r.NetProto]
+		if !ok {
+			// For this to happen stack.makeRoute() must have been called with the
+			// incorrect protocol number. Since we have successfully completed
+			// network layer processing this should be impossible.
+			panic(fmt.Sprintf("expected stack to have a NetworkProtocol for proto = %d", r.NetProto))
+		}
+
+		_ = np.ReturnError(r, &tcpip.ICMPReasonPortUnreachable{}, pkt)
+	case UnknownDestinationPacketHandled:
 	}
 }
 

pkg/tcpip/stack/nic_test.go

@@ -221,6 +221,11 @@ func (*testIPv6Protocol) Parse(*PacketBuffer) (tcpip.TransportProtocolNumber, bo
 	return 0, false, false
 }
 
+// ReturnError implements NetworkProtocol.ReturnError.
+func (*testIPv6Protocol) ReturnError(*Route, tcpip.ICMPReason, *PacketBuffer) *tcpip.Error {
+	return nil
+}
+
 var _ LinkAddressResolver = (*testIPv6Protocol)(nil)
 
 // LinkAddressProtocol implements LinkAddressResolver.

pkg/tcpip/stack/packet_buffer.go

@@ -80,11 +80,17 @@ type PacketBuffer struct {
 	// data are held in the same underlying buffer storage.
 	header buffer.Prependable
 
-	// NetworkProtocolNumber is only valid when NetworkHeader is set.
+	// NetworkProtocolNumber is only valid when NetworkHeader().View().IsEmpty()
+	// returns false.
 	// TODO(gvisor.dev/issue/3574): Remove the separately passed protocol
 	// numbers in registration APIs that take a PacketBuffer.
 	NetworkProtocolNumber tcpip.NetworkProtocolNumber
 
+	// TransportProtocol is only valid if it is non zero.
+	// TODO(gvisor.dev/issue/3810): This and the network protocol number should
+	// be moved into the headerinfo. This should resolve the validity issue.
+	TransportProtocolNumber tcpip.TransportProtocolNumber
+
 	// Hash is the transport layer hash of this packet. A value of zero
 	// indicates no valid hash has been set.
 	Hash uint32
@@ -234,16 +240,17 @@ func (pk *PacketBuffer) consume(typ headerType, size int) (v buffer.View, consum
 // underlying packet payload.
 func (pk *PacketBuffer) Clone() *PacketBuffer {
 	newPk := &PacketBuffer{
-		PacketBufferEntry:     pk.PacketBufferEntry,
-		Data:                  pk.Data.Clone(nil),
-		headers:               pk.headers,
-		header:                pk.header,
-		Hash:                  pk.Hash,
-		Owner:                 pk.Owner,
-		EgressRoute:           pk.EgressRoute,
-		GSOOptions:            pk.GSOOptions,
-		NetworkProtocolNumber: pk.NetworkProtocolNumber,
-		NatDone:               pk.NatDone,
+		PacketBufferEntry:       pk.PacketBufferEntry,
+		Data:                    pk.Data.Clone(nil),
+		headers:                 pk.headers,
+		header:                  pk.header,
+		Hash:                    pk.Hash,
+		Owner:                   pk.Owner,
+		EgressRoute:             pk.EgressRoute,
+		GSOOptions:              pk.GSOOptions,
+		NetworkProtocolNumber:   pk.NetworkProtocolNumber,
+		NatDone:                 pk.NatDone,
+		TransportProtocolNumber: pk.TransportProtocolNumber,
 	}
 	return newPk
 }

pkg/tcpip/stack/registration.go

@@ -125,6 +125,26 @@ type PacketEndpoint interface {
 	HandlePacket(nicID tcpip.NICID, addr tcpip.LinkAddress, netProto tcpip.NetworkProtocolNumber, pkt *PacketBuffer)
 }
 
+// UnknownDestinationPacketDisposition enumerates the possible return vaues from
+// HandleUnknownDestinationPacket().
+type UnknownDestinationPacketDisposition int
+
+const (
+	// UnknownDestinationPacketMalformed denotes that the packet was malformed
+	// and no further processing should be attempted other than updating
+	// statistics.
+	UnknownDestinationPacketMalformed UnknownDestinationPacketDisposition = iota
+
+	// UnknownDestinationPacketUnhandled tells the caller that the packet was
+	// well formed but that the issue was not handled and the stack should take
+	// the default action.
+	UnknownDestinationPacketUnhandled
+
+	// UnknownDestinationPacketHandled tells the caller that it should do
+	// no further processing.
+	UnknownDestinationPacketHandled
+)
+
 // TransportProtocol is the interface that needs to be implemented by transport
 // protocols (e.g., tcp, udp) that want to be part of the networking stack.
 type TransportProtocol interface {
@@ -147,14 +167,12 @@ type TransportProtocol interface {
 	ParsePorts(v buffer.View) (src, dst uint16, err *tcpip.Error)
 
 	// HandleUnknownDestinationPacket handles packets targeted at this
-	// protocol but that don't match any existing endpoint. For example,
-	// it is targeted at a port that have no listeners.
+	// protocol that don't match any existing endpoint. For example,
+	// it is targeted at a port that has no listeners.
 	//
-	// The return value indicates whether the packet was well-formed (for
-	// stats purposes only).
-	//
-	// HandleUnknownDestinationPacket takes ownership of pkt.
-	HandleUnknownDestinationPacket(r *Route, id TransportEndpointID, pkt *PacketBuffer) bool
+	// HandleUnknownDestinationPacket takes ownership of pkt if it handles
+	// the issue.
+	HandleUnknownDestinationPacket(r *Route, id TransportEndpointID, pkt *PacketBuffer) UnknownDestinationPacketDisposition
 
 	// SetOption allows enabling/disabling protocol specific features.
 	// SetOption returns an error if the option is not supported or the
@@ -324,6 +342,19 @@ type NetworkProtocol interface {
 	//   does not encapsulate anything).
 	// - Whether pkt.Data was large enough to parse and set pkt.NetworkHeader.
 	Parse(pkt *PacketBuffer) (proto tcpip.TransportProtocolNumber, hasTransportHdr bool, ok bool)
+
+	// ReturnError attempts to send a suitable error message to the sender
+	// of a received packet.
+	// - pkt holds the problematic packet.
+	// - reason indicates what the reason for wanting a message is.
+	// - route is the routing information for the received packet
+	// ReturnError returns an error if the send failed and nil on success.
+	// Note that deciding to deliberately send no message is a success.
+	//
+	// TODO(gvisor.dev/issues/3871): This method should be removed or simplified
+	// after all (or all but one) of the ICMP error dispatch occurs through the
+	// protocol specific modules. May become SendPortNotFound(r, pkt).
+	ReturnError(r *Route, reason tcpip.ICMPReason, pkt *PacketBuffer) *tcpip.Error
 }
 
 // NetworkDispatcher contains the methods used by the network stack to deliver

pkg/tcpip/stack/stack_test.go

@@ -216,13 +216,18 @@ func (f *fakeNetworkProtocol) Option(option tcpip.GettableNetworkProtocolOption)
 	}
 }
 
-// Close implements TransportProtocol.Close.
+// ReturnError implements NetworkProtocol.ReturnError
+func (*fakeNetworkProtocol) ReturnError(*stack.Route, tcpip.ICMPReason, *stack.PacketBuffer) *tcpip.Error {
+	return nil
+}
+
+// Close implements NetworkProtocol.Close.
 func (*fakeNetworkProtocol) Close() {}
 
-// Wait implements TransportProtocol.Wait.
+// Wait implements NetworkProtocol.Wait.
 func (*fakeNetworkProtocol) Wait() {}
 
-// Parse implements TransportProtocol.Parse.
+// Parse implements NetworkProtocol.Parse.
 func (*fakeNetworkProtocol) Parse(pkt *stack.PacketBuffer) (tcpip.TransportProtocolNumber, bool, bool) {
 	hdr, ok := pkt.NetworkHeader().Consume(fakeNetHeaderLen)
 	if !ok {

pkg/tcpip/stack/transport_test.go

@@ -287,8 +287,8 @@ func (*fakeTransportProtocol) ParsePorts(buffer.View) (src, dst uint16, err *tcp
 	return 0, 0, nil
 }
 
-func (*fakeTransportProtocol) HandleUnknownDestinationPacket(*stack.Route, stack.TransportEndpointID, *stack.PacketBuffer) bool {
-	return true
+func (*fakeTransportProtocol) HandleUnknownDestinationPacket(*stack.Route, stack.TransportEndpointID, *stack.PacketBuffer) stack.UnknownDestinationPacketDisposition {
+	return stack.UnknownDestinationPacketHandled
 }
 
 func (f *fakeTransportProtocol) SetOption(option tcpip.SettableTransportProtocolOption) *tcpip.Error {

pkg/tcpip/tcpip.go

@@ -1987,3 +1987,14 @@ func DeleteDanglingEndpoint(e Endpoint) {
 // AsyncLoading is the global barrier for asynchronous endpoint loading
 // activities.
 var AsyncLoading sync.WaitGroup
+
+// ICMPReason is a marker interface for network protocol agnostic ICMP errors.
+type ICMPReason interface {
+	isICMP()
+}
+
+// ICMPReasonPortUnreachable is an error where the transport protocol has no
+// listener and no alternative means to inform the sender.
+type ICMPReasonPortUnreachable struct{}
+
+func (*ICMPReasonPortUnreachable) isICMP() {}

pkg/tcpip/transport/icmp/endpoint.go

@@ -446,6 +446,7 @@ func send4(r *stack.Route, ident uint16, data buffer.View, ttl uint8, owner tcpi
 	pkt.Owner = owner
 
 	icmpv4 := header.ICMPv4(pkt.TransportHeader().Push(header.ICMPv4MinimumSize))
+	pkt.TransportProtocolNumber = header.ICMPv4ProtocolNumber
 	copy(icmpv4, data)
 	// Set the ident to the user-specified port. Sequence number should
 	// already be set by the user.
@@ -478,6 +479,7 @@ func send6(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Err
 	})
 
 	icmpv6 := header.ICMPv6(pkt.TransportHeader().Push(header.ICMPv6MinimumSize))
+	pkt.TransportProtocolNumber = header.ICMPv6ProtocolNumber
 	copy(icmpv6, data)
 	// Set the ident. Sequence number is provided by the user.
 	icmpv6.SetIdent(ident)

pkg/tcpip/transport/icmp/protocol.go

@@ -104,8 +104,8 @@ func (p *protocol) ParsePorts(v buffer.View) (src, dst uint16, err *tcpip.Error)
 
 // HandleUnknownDestinationPacket handles packets targeted at this protocol but
 // that don't match any existing endpoint.
-func (*protocol) HandleUnknownDestinationPacket(*stack.Route, stack.TransportEndpointID, *stack.PacketBuffer) bool {
-	return true
+func (*protocol) HandleUnknownDestinationPacket(*stack.Route, stack.TransportEndpointID, *stack.PacketBuffer) stack.UnknownDestinationPacketDisposition {
+	return stack.UnknownDestinationPacketHandled
 }
 
 // SetOption implements stack.TransportProtocol.SetOption.

pkg/tcpip/transport/tcp/connect.go

@@ -747,6 +747,7 @@ func (e *endpoint) sendTCP(r *stack.Route, tf tcpFields, data buffer.VectorisedV
 func buildTCPHdr(r *stack.Route, tf tcpFields, pkt *stack.PacketBuffer, gso *stack.GSO) {
 	optLen := len(tf.opts)
 	tcp := header.TCP(pkt.TransportHeader().Push(header.TCPMinimumSize + optLen))
+	pkt.TransportProtocolNumber = header.TCPProtocolNumber
 	tcp.Encode(&header.TCPFields{
 		SrcPort:    tf.id.LocalPort,
 		DstPort:    tf.id.RemotePort,

pkg/tcpip/transport/tcp/protocol.go

@@ -201,21 +201,20 @@ func (p *protocol) QueuePacket(r *stack.Route, ep stack.TransportEndpoint, id st
 // a reset is sent in response to any incoming segment except another reset. In
 // particular, SYNs addressed to a non-existent connection are rejected by this
 // means."
-func (*protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.TransportEndpointID, pkt *stack.PacketBuffer) bool {
+
+func (*protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.TransportEndpointID, pkt *stack.PacketBuffer) stack.UnknownDestinationPacketDisposition {
 	s := newSegment(r, id, pkt)
 	defer s.decRef()
 
 	if !s.parse() || !s.csumValid {
-		return false
+		return stack.UnknownDestinationPacketMalformed
 	}
 
-	// There's nothing to do if this is already a reset packet.
-	if s.flagIsSet(header.TCPFlagRst) {
-		return true
+	if !s.flagIsSet(header.TCPFlagRst) {
+		replyWithReset(s, stack.DefaultTOS, s.route.DefaultTTL())
 	}
 
-	replyWithReset(s, stack.DefaultTOS, s.route.DefaultTTL())
-	return true
+	return stack.UnknownDestinationPacketHandled
 }
 
 // replyWithReset replies to the given segment with a reset segment.