Merge release-20201109.0-118-gbc81fcced (automated)

9 files changed, 392 insertions, 64 deletions

diff --git a/pkg/tcpip/header/icmpv6.go b/pkg/tcpip/header/icmpv6.go
index 4303fc5d5..2eef64b4d 100644
--- a/pkg/tcpip/header/icmpv6.go
+++ b/pkg/tcpip/header/icmpv6.go
@@ -115,6 +115,12 @@ const (
 	ICMPv6NeighborSolicit ICMPv6Type = 135
 	ICMPv6NeighborAdvert  ICMPv6Type = 136
 	ICMPv6RedirectMsg     ICMPv6Type = 137
+
+	// Multicast Listener Discovery (MLD) messages, see RFC 2710.
+
+	ICMPv6MulticastListenerQuery  ICMPv6Type = 130
+	ICMPv6MulticastListenerReport ICMPv6Type = 131
+	ICMPv6MulticastListenerDone   ICMPv6Type = 132
 )
 
 // IsErrorType returns true if the receiver is an ICMP error type.
@@ -245,10 +251,9 @@ func (b ICMPv6) SetSequence(sequence uint16) {
 	binary.BigEndian.PutUint16(b[icmpv6SequenceOffset:], sequence)
 }
 
-// NDPPayload returns the NDP payload buffer. That is, it returns the ICMPv6
-// packet's message body as defined by RFC 4443 section 2.1; the portion of the
-// ICMPv6 buffer after the first ICMPv6HeaderSize bytes.
-func (b ICMPv6) NDPPayload() []byte {
+// MessageBody returns the message body as defined by RFC 4443 section 2.1; the
+// portion of the ICMPv6 buffer after the first ICMPv6HeaderSize bytes.
+func (b ICMPv6) MessageBody() []byte {
 	return b[ICMPv6HeaderSize:]
 }
 
diff --git a/pkg/tcpip/header/mld.go b/pkg/tcpip/header/mld.go
index f70623092..ffe03c76a 100644
--- a/pkg/tcpip/header/mld.go
+++ b/pkg/tcpip/header/mld.go
@@ -23,6 +23,13 @@ import (
 )
 
 const (
+	// MLDMinimumSize is the minimum size for an MLD message.
+	MLDMinimumSize = 20
+
+	// MLDHopLimit is the Hop Limit for all IPv6 packets with an MLD message, as
+	// per RFC 2710 section 3.
+	MLDHopLimit = 1
+
 	// mldMaximumResponseDelayOffset is the offset to the Maximum Response Delay
 	// field within MLD.
 	mldMaximumResponseDelayOffset = 0
diff --git a/pkg/tcpip/network/ipv4/igmp.go b/pkg/tcpip/network/ipv4/igmp.go
index c9bf117de..37f1822ca 100644
--- a/pkg/tcpip/network/ipv4/igmp.go
+++ b/pkg/tcpip/network/ipv4/igmp.go
@@ -51,6 +51,16 @@ const (
 	UnsolicitedReportIntervalMax = 10 * time.Second
 )
 
+// IGMPOptions holds options for IGMP.
+type IGMPOptions struct {
+	// Enabled indicates whether IGMP will be performed.
+	//
+	// When enabled, IGMP may transmit IGMP report and leave messages when
+	// joining and leaving multicast groups respectively, and handle incoming
+	// IGMP packets.
+	Enabled bool
+}
+
 var _ ip.MulticastGroupProtocol = (*igmpState)(nil)
 
 // igmpState is the per-interface IGMP state.
@@ -58,7 +68,8 @@ var _ ip.MulticastGroupProtocol = (*igmpState)(nil)
 // igmpState.init() MUST be called after creating an IGMP state.
 type igmpState struct {
 	// The IPv4 endpoint this igmpState is for.
-	ep *endpoint
+	ep   *endpoint
+	opts IGMPOptions
 
 	// igmpV1Present is for maintaining compatibility with IGMPv1 Routers, from
 	// RFC 2236 Section 4 Page 6: "The IGMPv1 router expects Version 1
@@ -108,10 +119,11 @@ func (igmp *igmpState) SendLeave(groupAddress tcpip.Address) *tcpip.Error {
 
 // init sets up an igmpState struct, and is required to be called before using
 // a new igmpState.
-func (igmp *igmpState) init(ep *endpoint) {
+func (igmp *igmpState) init(ep *endpoint, opts IGMPOptions) {
 	igmp.mu.Lock()
 	defer igmp.mu.Unlock()
 	igmp.ep = ep
+	igmp.opts = opts
 	igmp.mu.genericMulticastProtocol.Init(ep.protocol.stack.Rand(), ep.protocol.stack.Clock(), igmp, UnsolicitedReportIntervalMax)
 	igmp.igmpV1Present = igmpV1PresentDefault
 	igmp.mu.igmpV1Job = igmp.ep.protocol.stack.NewJob(&igmp.mu, func() {
@@ -189,6 +201,10 @@ func (igmp *igmpState) setV1Present(v bool) {
 }
 
 func (igmp *igmpState) handleMembershipQuery(groupAddress tcpip.Address, maxRespTime time.Duration) {
+	if !igmp.opts.Enabled {
+		return
+	}
+
 	igmp.mu.Lock()
 	defer igmp.mu.Unlock()
 
@@ -206,6 +222,10 @@ func (igmp *igmpState) handleMembershipQuery(groupAddress tcpip.Address, maxResp
 }
 
 func (igmp *igmpState) handleMembershipReport(groupAddress tcpip.Address) {
+	if !igmp.opts.Enabled {
+		return
+	}
+
 	igmp.mu.Lock()
 	defer igmp.mu.Unlock()
 	igmp.mu.genericMulticastProtocol.HandleReport(groupAddress)
@@ -226,11 +246,8 @@ func (igmp *igmpState) writePacket(destAddress tcpip.Address, groupAddress tcpip
 
 	// TODO(gvisor.dev/issue/4888): We should not use the unspecified address,
 	// rather we should select an appropriate local address.
-	r := stack.Route{
-		LocalAddress:  header.IPv4Any,
-		RemoteAddress: destAddress,
-	}
-	igmp.ep.addIPHeader(&r, pkt, stack.NetworkHeaderParams{
+	localAddr := header.IPv4Any
+	igmp.ep.addIPHeader(localAddr, destAddress, pkt, stack.NetworkHeaderParams{
 		Protocol: header.IGMPProtocolNumber,
 		TTL:      header.IGMPTTL,
 		TOS:      stack.DefaultTOS,
@@ -239,7 +256,7 @@ func (igmp *igmpState) writePacket(destAddress tcpip.Address, groupAddress tcpip
 	// TODO(b/162198658): set the ROUTER_ALERT option when sending Host
 	// Membership Reports.
 	sent := igmp.ep.protocol.stack.Stats().IGMP.PacketsSent
-	if err := igmp.ep.nic.WritePacketToRemote(header.EthernetAddressFromMulticastIPv4Address(destAddress), nil /* gso */, header.IPv4ProtocolNumber, pkt); err != nil {
+	if err := igmp.ep.nic.WritePacketToRemote(header.EthernetAddressFromMulticastIPv4Address(destAddress), nil /* gso */, ProtocolNumber, pkt); err != nil {
 		sent.Dropped.Increment()
 		return err
 	}
@@ -263,6 +280,26 @@ func (igmp *igmpState) writePacket(destAddress tcpip.Address, groupAddress tcpip
 // If the group already exists in the membership map, returns
 // tcpip.ErrDuplicateAddress.
 func (igmp *igmpState) joinGroup(groupAddress tcpip.Address) *tcpip.Error {
+	if !igmp.opts.Enabled {
+		return nil
+	}
+
+	// As per RFC 2236 section 6 page 10,
+	//
+	//   The all-systems group (address 224.0.0.1) is handled as a special
+	//   case. The host starts in Idle Member state for that group on every
+	//   interface, never transitions to another state, and never sends a
+	//   report for that group.
+	//
+	// This is equivalent to not performing IGMP for the all-systems multicast
+	// address. Simply not performing IGMP when the group is added will prevent
+	// any work from being done on the all-systems multicast group when leaving
+	// the group or when query or report messages are received for it since the
+	// MGP state will not know about it.
+	if groupAddress == header.IPv4AllSystems {
+		return nil
+	}
+
 	igmp.mu.Lock()
 	defer igmp.mu.Unlock()
 
@@ -280,6 +317,10 @@ func (igmp *igmpState) joinGroup(groupAddress tcpip.Address) *tcpip.Error {
 // If the group does not exist in the membership map, this function will
 // silently return.
 func (igmp *igmpState) leaveGroup(groupAddress tcpip.Address) {
+	if !igmp.opts.Enabled {
+		return
+	}
+
 	igmp.mu.Lock()
 	defer igmp.mu.Unlock()
 	igmp.mu.genericMulticastProtocol.LeaveGroup(groupAddress)
diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go
index 7c759be9a..be9c8e2f9 100644
--- a/pkg/tcpip/network/ipv4/ipv4.go
+++ b/pkg/tcpip/network/ipv4/ipv4.go
@@ -95,7 +95,7 @@ func (p *protocol) NewEndpoint(nic stack.NetworkInterface, _ stack.LinkAddressCa
 		protocol:   p,
 	}
 	e.mu.addressableEndpointState.Init(e)
-	e.igmp.init(e)
+	e.igmp.init(e, p.options.IGMP)
 	return e
 }
 
@@ -126,7 +126,7 @@ func (e *endpoint) Enable() *tcpip.Error {
 	// As per RFC 1122 section 3.3.7, all hosts should join the all-hosts
 	// multicast group. Note, the IANA calls the all-hosts multicast group the
 	// all-systems multicast group.
-	_, err = e.mu.addressableEndpointState.JoinGroup(header.IPv4AllSystems)
+	_, err = e.joinGroupLocked(header.IPv4AllSystems)
 	return err
 }
 
@@ -164,7 +164,7 @@ func (e *endpoint) disableLocked() {
 	}
 
 	// The endpoint may have already left the multicast group.
-	if _, err := e.mu.addressableEndpointState.LeaveGroup(header.IPv4AllSystems); err != nil && err != tcpip.ErrBadLocalAddress {
+	if _, err := e.leaveGroupLocked(header.IPv4AllSystems); err != nil && err != tcpip.ErrBadLocalAddress {
 		panic(fmt.Sprintf("unexpected error when leaving group = %s: %s", header.IPv4AllSystems, err))
 	}
 
@@ -200,7 +200,7 @@ func (e *endpoint) NetworkProtocolNumber() tcpip.NetworkProtocolNumber {
 	return e.protocol.Number()
 }
 
-func (e *endpoint) addIPHeader(r *stack.Route, pkt *stack.PacketBuffer, params stack.NetworkHeaderParams) {
+func (e *endpoint) addIPHeader(srcAddr, dstAddr tcpip.Address, pkt *stack.PacketBuffer, params stack.NetworkHeaderParams) {
 	hdrLen := header.IPv4MinimumSize
 	var opts header.IPv4Options
 	if params.Options != nil {
@@ -221,15 +221,15 @@ func (e *endpoint) addIPHeader(r *stack.Route, pkt *stack.PacketBuffer, params s
 	// RFC 6864 section 4.3 mandates uniqueness of ID values for non-atomic
 	// datagrams. Since the DF bit is never being set here, all datagrams
 	// are non-atomic and need an ID.
-	id := atomic.AddUint32(&e.protocol.ids[hashRoute(r, params.Protocol, e.protocol.hashIV)%buckets], 1)
+	id := atomic.AddUint32(&e.protocol.ids[hashRoute(srcAddr, dstAddr, params.Protocol, e.protocol.hashIV)%buckets], 1)
 	ip.Encode(&header.IPv4Fields{
 		TotalLength: length,
 		ID:          uint16(id),
 		TTL:         params.TTL,
 		TOS:         params.TOS,
 		Protocol:    uint8(params.Protocol),
-		SrcAddr:     r.LocalAddress,
-		DstAddr:     r.RemoteAddress,
+		SrcAddr:     srcAddr,
+		DstAddr:     dstAddr,
 		Options:     opts,
 	})
 	ip.SetChecksum(^ip.CalculateChecksum())
@@ -261,7 +261,7 @@ 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)
+	e.addIPHeader(r.LocalAddress, r.RemoteAddress, pkt, params)
 
 	// iptables filtering. All packets that reach here are locally
 	// generated.
@@ -349,7 +349,7 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe
 	}
 
 	for pkt := pkts.Front(); pkt != nil; pkt = pkt.Next() {
-		e.addIPHeader(r, pkt, params)
+		e.addIPHeader(r.LocalAddress, r.RemoteAddress, pkt, params)
 		networkMTU, err := calculateNetworkMTU(e.nic.MTU(), uint32(pkt.NetworkHeader().View().Size()))
 		if err != nil {
 			r.Stats().IP.OutgoingPacketErrors.IncrementBy(uint64(pkts.Len()))
@@ -463,7 +463,7 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt *stack.PacketBu
 		// non-atomic datagrams, so assign an ID to all such datagrams
 		// according to the definition given in RFC 6864 section 4.
 		if ip.Flags()&header.IPv4FlagDontFragment == 0 || ip.Flags()&header.IPv4FlagMoreFragments != 0 || ip.FragmentOffset() > 0 {
-			ip.SetID(uint16(atomic.AddUint32(&e.protocol.ids[hashRoute(r, 0 /* protocol */, e.protocol.hashIV)%buckets], 1)))
+			ip.SetID(uint16(atomic.AddUint32(&e.protocol.ids[hashRoute(r.LocalAddress, r.RemoteAddress, 0 /* protocol */, e.protocol.hashIV)%buckets], 1)))
 		}
 	}
 
@@ -706,10 +706,7 @@ func (e *endpoint) handlePacket(pkt *stack.PacketBuffer) {
 		return
 	}
 	if p == header.IGMPProtocolNumber {
-		if e.protocol.options.IGMPEnabled {
-			e.igmp.handleIGMP(pkt)
-		}
-		// Nothing further to do with an IGMP packet, even if IGMP is not enabled.
+		e.igmp.handleIGMP(pkt)
 		return
 	}
 	if opts := h.Options(); len(opts) != 0 {
@@ -837,32 +834,55 @@ func (e *endpoint) PermanentAddresses() []tcpip.AddressWithPrefix {
 
 // JoinGroup implements stack.GroupAddressableEndpoint.
 func (e *endpoint) JoinGroup(addr tcpip.Address) (bool, *tcpip.Error) {
+	e.mu.Lock()
+	defer e.mu.Unlock()
+	return e.joinGroupLocked(addr)
+}
+
+// joinGroupLocked is like JoinGroup, but with locking requirements.
+//
+// Precondition: e.mu must be locked.
+func (e *endpoint) joinGroupLocked(addr tcpip.Address) (bool, *tcpip.Error) {
 	if !header.IsV4MulticastAddress(addr) {
 		return false, tcpip.ErrBadAddress
 	}
+	// TODO(gvisor.dev/issue/4916): Keep track of join count and IGMP state in a
+	// single type.
+	joined, err := e.mu.addressableEndpointState.JoinGroup(addr)
+	if err != nil || !joined {
+		return joined, err
+	}
 
-	e.mu.Lock()
-	defer e.mu.Unlock()
-
-	joinedGroup, err := e.mu.addressableEndpointState.JoinGroup(addr)
-	if err == nil && joinedGroup && e.protocol.options.IGMPEnabled {
-		_ = e.igmp.joinGroup(addr)
+	// joinGroup only returns an error if we try to join a group twice, but we
+	// checked above to make sure that the group was newly joined.
+	if err := e.igmp.joinGroup(addr); err != nil {
+		panic(fmt.Sprintf("e.igmp.joinGroup(%s): %s", addr, err))
 	}
 
-	return joinedGroup, err
+	return true, nil
 }
 
 // LeaveGroup implements stack.GroupAddressableEndpoint.
 func (e *endpoint) LeaveGroup(addr tcpip.Address) (bool, *tcpip.Error) {
 	e.mu.Lock()
 	defer e.mu.Unlock()
+	return e.leaveGroupLocked(addr)
+}
+
+// leaveGroupLocked is like LeaveGroup, but with locking requirements.
+//
+// Precondition: e.mu must be locked.
+func (e *endpoint) leaveGroupLocked(addr tcpip.Address) (bool, *tcpip.Error) {
+	left, err := e.mu.addressableEndpointState.LeaveGroup(addr)
+	if err != nil {
+		return left, err
+	}
 
-	leftGroup, err := e.mu.addressableEndpointState.LeaveGroup(addr)
-	if err == nil && leftGroup && e.protocol.options.IGMPEnabled {
+	if left {
 		e.igmp.leaveGroup(addr)
 	}
 
-	return leftGroup, err
+	return left, nil
 }
 
 // IsInGroup implements stack.GroupAddressableEndpoint.
@@ -1021,20 +1041,19 @@ func addressToUint32(addr tcpip.Address) uint32 {
 	return uint32(addr[0]) | uint32(addr[1])<<8 | uint32(addr[2])<<16 | uint32(addr[3])<<24
 }
 
-// hashRoute calculates a hash value for the given route. It uses the source &
-// destination address, the transport protocol number and a 32-bit number to
-// generate the hash.
-func hashRoute(r *stack.Route, protocol tcpip.TransportProtocolNumber, hashIV uint32) uint32 {
-	a := addressToUint32(r.LocalAddress)
-	b := addressToUint32(r.RemoteAddress)
+// hashRoute calculates a hash value for the given source/destination pair using
+// the addresses, transport protocol number and a 32-bit number to generate the
+// hash.
+func hashRoute(srcAddr, dstAddr tcpip.Address, protocol tcpip.TransportProtocolNumber, hashIV uint32) uint32 {
+	a := addressToUint32(srcAddr)
+	b := addressToUint32(dstAddr)
 	return hash.Hash3Words(a, b, uint32(protocol), hashIV)
 }
 
 // Options holds options to configure a new protocol.
 type Options struct {
-	// IGMPEnabled indicates whether incoming IGMP packets will be handled and if
-	// this endpoint will transmit IGMP packets on IGMP related events.
-	IGMPEnabled bool
+	// IGMP holds options for IGMP.
+	IGMP IGMPOptions
 }
 
 // NewProtocolWithOptions returns an IPv4 network protocol.
diff --git a/pkg/tcpip/network/ipv6/icmp.go b/pkg/tcpip/network/ipv6/icmp.go
index 386d98a29..0fb98c578 100644
--- a/pkg/tcpip/network/ipv6/icmp.go
+++ b/pkg/tcpip/network/ipv6/icmp.go
@@ -163,7 +163,7 @@ func (e *endpoint) handleICMP(pkt *stack.PacketBuffer, hasFragmentHeader bool) {
 	}
 
 	// TODO(b/112892170): Meaningfully handle all ICMP types.
-	switch h.Type() {
+	switch icmpType := h.Type(); icmpType {
 	case header.ICMPv6PacketTooBig:
 		received.PacketTooBig.Increment()
 		hdr, ok := pkt.Data.PullUp(header.ICMPv6PacketTooBigMinimumSize)
@@ -358,7 +358,7 @@ func (e *endpoint) handleICMP(pkt *stack.PacketBuffer, hasFragmentHeader bool) {
 		pkt.TransportProtocolNumber = header.ICMPv6ProtocolNumber
 		packet := header.ICMPv6(pkt.TransportHeader().Push(neighborAdvertSize))
 		packet.SetType(header.ICMPv6NeighborAdvert)
-		na := header.NDPNeighborAdvert(packet.NDPPayload())
+		na := header.NDPNeighborAdvert(packet.MessageBody())
 
 		// As per RFC 4861 section 7.2.4:
 		//
@@ -644,8 +644,31 @@ func (e *endpoint) handleICMP(pkt *stack.PacketBuffer, hasFragmentHeader bool) {
 			return
 		}
 
+	case header.ICMPv6MulticastListenerQuery, header.ICMPv6MulticastListenerReport, header.ICMPv6MulticastListenerDone:
+		var handler func(header.MLD)
+		switch icmpType {
+		case header.ICMPv6MulticastListenerQuery:
+			received.MulticastListenerQuery.Increment()
+			handler = e.mld.handleMulticastListenerQuery
+		case header.ICMPv6MulticastListenerReport:
+			received.MulticastListenerReport.Increment()
+			handler = e.mld.handleMulticastListenerReport
+		case header.ICMPv6MulticastListenerDone:
+			received.MulticastListenerDone.Increment()
+		default:
+			panic(fmt.Sprintf("unrecognized MLD message = %d", icmpType))
+		}
+		if pkt.Data.Size()-header.ICMPv6HeaderSize < header.MLDMinimumSize {
+			received.Invalid.Increment()
+			return
+		}
+
+		if handler != nil {
+			handler(header.MLD(payload.ToView()))
+		}
+
 	default:
-		received.Invalid.Increment()
+		received.Unrecognized.Increment()
 	}
 }
 
@@ -681,7 +704,7 @@ func (p *protocol) LinkAddressRequest(targetAddr, localAddr tcpip.Address, remot
 	pkt.TransportProtocolNumber = header.ICMPv6ProtocolNumber
 	packet := header.ICMPv6(pkt.TransportHeader().Push(neighborSolicitSize))
 	packet.SetType(header.ICMPv6NeighborSolicit)
-	ns := header.NDPNeighborSolicit(packet.NDPPayload())
+	ns := header.NDPNeighborSolicit(packet.MessageBody())
 	ns.SetTargetAddress(targetAddr)
 	ns.Options().Serialize(optsSerializer)
 	packet.SetChecksum(header.ICMPv6Checksum(packet, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{}))
diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go
index 181c50cc7..ac67d4ac5 100644
--- a/pkg/tcpip/network/ipv6/ipv6.go
+++ b/pkg/tcpip/network/ipv6/ipv6.go
@@ -86,6 +86,8 @@ type endpoint struct {
 		addressableEndpointState stack.AddressableEndpointState
 		ndp                      ndpState
 	}
+
+	mld mldState
 }
 
 // NICNameFromID is a function that returns a stable name for the specified NIC,
@@ -243,7 +245,7 @@ func (e *endpoint) Enable() *tcpip.Error {
 	// (NDP NS) messages may be sent to the All-Nodes multicast group if the
 	// source address of the NDP NS is the unspecified address, as per RFC 4861
 	// section 7.2.4.
-	if _, err := e.mu.addressableEndpointState.JoinGroup(header.IPv6AllNodesMulticastAddress); err != nil {
+	if _, err := e.joinGroupLocked(header.IPv6AllNodesMulticastAddress); err != nil {
 		return err
 	}
 
@@ -333,7 +335,7 @@ func (e *endpoint) disableLocked() {
 	e.stopDADForPermanentAddressesLocked()
 
 	// The endpoint may have already left the multicast group.
-	if _, err := e.mu.addressableEndpointState.LeaveGroup(header.IPv6AllNodesMulticastAddress); err != nil && err != tcpip.ErrBadLocalAddress {
+	if _, err := e.leaveGroupLocked(header.IPv6AllNodesMulticastAddress); err != nil && err != tcpip.ErrBadLocalAddress {
 		panic(fmt.Sprintf("unexpected error when leaving group = %s: %s", header.IPv6AllNodesMulticastAddress, err))
 	}
 }
@@ -378,7 +380,7 @@ func (e *endpoint) MaxHeaderLength() uint16 {
 	return e.nic.MaxHeaderLength() + header.IPv6MinimumSize
 }
 
-func (e *endpoint) addIPHeader(r *stack.Route, pkt *stack.PacketBuffer, params stack.NetworkHeaderParams) {
+func (e *endpoint) addIPHeader(srcAddr, dstAddr tcpip.Address, pkt *stack.PacketBuffer, params stack.NetworkHeaderParams) {
 	length := uint16(pkt.Size())
 	ip := header.IPv6(pkt.NetworkHeader().Push(header.IPv6MinimumSize))
 	ip.Encode(&header.IPv6Fields{
@@ -386,8 +388,8 @@ func (e *endpoint) addIPHeader(r *stack.Route, pkt *stack.PacketBuffer, params s
 		NextHeader:    uint8(params.Protocol),
 		HopLimit:      params.TTL,
 		TrafficClass:  params.TOS,
-		SrcAddr:       r.LocalAddress,
-		DstAddr:       r.RemoteAddress,
+		SrcAddr:       srcAddr,
+		DstAddr:       dstAddr,
 	})
 	pkt.NetworkProtocolNumber = ProtocolNumber
 }
@@ -442,7 +444,7 @@ 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)
+	e.addIPHeader(r.LocalAddress, r.RemoteAddress, pkt, params)
 
 	// iptables filtering. All packets that reach here are locally
 	// generated.
@@ -531,7 +533,7 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe
 
 	linkMTU := e.nic.MTU()
 	for pb := pkts.Front(); pb != nil; pb = pb.Next() {
-		e.addIPHeader(r, pb, params)
+		e.addIPHeader(r.LocalAddress, r.RemoteAddress, pb, params)
 
 		networkMTU, err := calculateNetworkMTU(linkMTU, uint32(pb.NetworkHeader().View().Size()))
 		if err != nil {
@@ -1164,7 +1166,7 @@ func (e *endpoint) addAndAcquirePermanentAddressLocked(addr tcpip.AddressWithPre
 	}
 
 	snmc := header.SolicitedNodeAddr(addr.Address)
-	if _, err := e.mu.addressableEndpointState.JoinGroup(snmc); err != nil {
+	if _, err := e.joinGroupLocked(snmc); err != nil {
 		return nil, err
 	}
 
@@ -1221,7 +1223,7 @@ func (e *endpoint) removePermanentEndpointLocked(addressEndpoint stack.AddressEn
 	}
 
 	snmc := header.SolicitedNodeAddr(addr.Address)
-	if _, err := e.mu.addressableEndpointState.LeaveGroup(snmc); err != nil && err != tcpip.ErrBadLocalAddress {
+	if _, err := e.leaveGroupLocked(snmc); err != nil && err != tcpip.ErrBadLocalAddress {
 		return err
 	}
 
@@ -1387,20 +1389,56 @@ func (e *endpoint) PermanentAddresses() []tcpip.AddressWithPrefix {
 
 // JoinGroup implements stack.GroupAddressableEndpoint.
 func (e *endpoint) JoinGroup(addr tcpip.Address) (bool, *tcpip.Error) {
+	e.mu.Lock()
+	defer e.mu.Unlock()
+	return e.joinGroupLocked(addr)
+}
+
+// joinGroupLocked is like JoinGroup, but with locking requirements.
+//
+// Precondition: e.mu must be locked.
+func (e *endpoint) joinGroupLocked(addr tcpip.Address) (bool, *tcpip.Error) {
 	if !header.IsV6MulticastAddress(addr) {
 		return false, tcpip.ErrBadAddress
 	}
 
-	e.mu.Lock()
-	defer e.mu.Unlock()
-	return e.mu.addressableEndpointState.JoinGroup(addr)
+	// TODO(gvisor.dev/issue/4916): Keep track of join count and MLD state in a
+	// single type.
+	joined, err := e.mu.addressableEndpointState.JoinGroup(addr)
+	if err != nil || !joined {
+		return joined, err
+	}
+
+	// joinGroup only returns an error if we try to join a group twice, but we
+	// checked above to make sure that the group was newly joined.
+	if err := e.mld.joinGroup(addr); err != nil {
+		panic(fmt.Sprintf("e.mld.joinGroup(%s): %s", addr, err))
+	}
+
+	return true, nil
 }
 
 // LeaveGroup implements stack.GroupAddressableEndpoint.
 func (e *endpoint) LeaveGroup(addr tcpip.Address) (bool, *tcpip.Error) {
 	e.mu.Lock()
 	defer e.mu.Unlock()
-	return e.mu.addressableEndpointState.LeaveGroup(addr)
+	return e.leaveGroupLocked(addr)
+}
+
+// leaveGroupLocked is like LeaveGroup, but with locking requirements.
+//
+// Precondition: e.mu must be locked.
+func (e *endpoint) leaveGroupLocked(addr tcpip.Address) (bool, *tcpip.Error) {
+	left, err := e.mu.addressableEndpointState.LeaveGroup(addr)
+	if err != nil {
+		return left, err
+	}
+
+	if left {
+		e.mld.leaveGroup(addr)
+	}
+
+	return left, nil
 }
 
 // IsInGroup implements stack.GroupAddressableEndpoint.
@@ -1482,6 +1520,7 @@ func (p *protocol) NewEndpoint(nic stack.NetworkInterface, linkAddrCache stack.L
 		slaacPrefixes:  make(map[tcpip.Subnet]slaacPrefixState),
 	}
 	e.mu.ndp.initializeTempAddrState()
+	e.mld.init(e, p.options.MLD)
 
 	p.mu.Lock()
 	defer p.mu.Unlock()
@@ -1638,6 +1677,9 @@ type Options struct {
 	// seed that is too small would reduce randomness and increase predictability,
 	// defeating the purpose of temporary SLAAC addresses.
 	TempIIDSeed []byte
+
+	// MLD holds options for MLD.
+	MLD MLDOptions
 }
 
 // NewProtocolWithOptions returns an IPv6 network protocol.
diff --git a/pkg/tcpip/network/ipv6/mld.go b/pkg/tcpip/network/ipv6/mld.go
new file mode 100644
index 000000000..b16a1afb0
--- /dev/null
+++ b/pkg/tcpip/network/ipv6/mld.go
@@ -0,0 +1,175 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package ipv6
+
+import (
+	"fmt"
+	"time"
+
+	"gvisor.dev/gvisor/pkg/tcpip"
+	"gvisor.dev/gvisor/pkg/tcpip/buffer"
+	"gvisor.dev/gvisor/pkg/tcpip/header"
+	"gvisor.dev/gvisor/pkg/tcpip/network/ip"
+	"gvisor.dev/gvisor/pkg/tcpip/stack"
+)
+
+const (
+	// UnsolicitedReportIntervalMax is the maximum delay between sending
+	// unsolicited MLD reports.
+	//
+	// Obtained from RFC 2710 Section 7.10.
+	UnsolicitedReportIntervalMax = 10 * time.Second
+)
+
+// MLDOptions holds options for MLD.
+type MLDOptions struct {
+	// Enabled indicates whether MLD will be performed.
+	//
+	// When enabled, MLD may transmit MLD report and done messages when
+	// joining and leaving multicast groups respectively, and handle incoming
+	// MLD packets.
+	Enabled bool
+}
+
+var _ ip.MulticastGroupProtocol = (*mldState)(nil)
+
+// mldState is the per-interface MLD state.
+//
+// mldState.init MUST be called to initialize the MLD state.
+type mldState struct {
+	// The IPv6 endpoint this mldState is for.
+	ep   *endpoint
+	opts MLDOptions
+
+	genericMulticastProtocol ip.GenericMulticastProtocolState
+}
+
+// SendReport implements ip.MulticastGroupProtocol.
+func (mld *mldState) SendReport(groupAddress tcpip.Address) *tcpip.Error {
+	return mld.writePacket(groupAddress, groupAddress, header.ICMPv6MulticastListenerReport)
+}
+
+// SendLeave implements ip.MulticastGroupProtocol.
+func (mld *mldState) SendLeave(groupAddress tcpip.Address) *tcpip.Error {
+	return mld.writePacket(header.IPv6AllRoutersMulticastAddress, groupAddress, header.ICMPv6MulticastListenerDone)
+}
+
+// init sets up an mldState struct, and is required to be called before using
+// a new mldState.
+func (mld *mldState) init(ep *endpoint, opts MLDOptions) {
+	mld.ep = ep
+	mld.opts = opts
+	mld.genericMulticastProtocol.Init(ep.protocol.stack.Rand(), ep.protocol.stack.Clock(), mld, UnsolicitedReportIntervalMax)
+}
+
+func (mld *mldState) handleMulticastListenerQuery(mldHdr header.MLD) {
+	if !mld.opts.Enabled {
+		return
+	}
+
+	mld.genericMulticastProtocol.HandleQuery(mldHdr.MulticastAddress(), mldHdr.MaximumResponseDelay())
+}
+
+func (mld *mldState) handleMulticastListenerReport(mldHdr header.MLD) {
+	if !mld.opts.Enabled {
+		return
+	}
+
+	mld.genericMulticastProtocol.HandleReport(mldHdr.MulticastAddress())
+}
+
+// joinGroup handles joining a new group and sending and scheduling the required
+// messages.
+//
+// If the group is already joined, returns tcpip.ErrDuplicateAddress.
+func (mld *mldState) joinGroup(groupAddress tcpip.Address) *tcpip.Error {
+	if !mld.opts.Enabled {
+		return nil
+	}
+
+	// As per RFC 2710 section 5 page 10,
+	//
+	//   The link-scope all-nodes address (FF02::1) is handled as a special
+	//   case. The node starts in Idle Listener state for that address on
+	//   every interface, never transitions to another state, and never sends
+	//   a Report or Done for that address.
+	//
+	// This is equivalent to not performing MLD for the all-nodes multicast
+	// address. Simply not performing MLD when the group is added will prevent
+	// any work from being done on the all-nodes multicast group when leaving the
+	// group or when query or report messages are received for it since the MGP
+	// state will not know about it.
+	if groupAddress == header.IPv6AllNodesMulticastAddress {
+		return nil
+	}
+
+	// JoinGroup returns false if we have already joined the group.
+	if !mld.genericMulticastProtocol.JoinGroup(groupAddress) {
+		return tcpip.ErrDuplicateAddress
+	}
+	return nil
+}
+
+// leaveGroup handles removing the group from the membership map, cancels any
+// delay timers associated with that group, and sends the Done message, if
+// required.
+//
+// If the group is not joined, this function will do nothing.
+func (mld *mldState) leaveGroup(groupAddress tcpip.Address) {
+	if !mld.opts.Enabled {
+		return
+	}
+
+	mld.genericMulticastProtocol.LeaveGroup(groupAddress)
+}
+
+func (mld *mldState) writePacket(destAddress, groupAddress tcpip.Address, mldType header.ICMPv6Type) *tcpip.Error {
+	sentStats := mld.ep.protocol.stack.Stats().ICMP.V6PacketsSent
+	var mldStat *tcpip.StatCounter
+	switch mldType {
+	case header.ICMPv6MulticastListenerReport:
+		mldStat = sentStats.MulticastListenerReport
+	case header.ICMPv6MulticastListenerDone:
+		mldStat = sentStats.MulticastListenerDone
+	default:
+		panic(fmt.Sprintf("unrecognized mld type = %d", mldType))
+	}
+
+	icmp := header.ICMPv6(buffer.NewView(header.ICMPv6HeaderSize + header.MLDMinimumSize))
+	icmp.SetType(mldType)
+	header.MLD(icmp.MessageBody()).SetMulticastAddress(groupAddress)
+	// TODO(gvisor.dev/issue/4888): We should not use the unspecified address,
+	// rather we should select an appropriate local address.
+	localAddress := header.IPv6Any
+	icmp.SetChecksum(header.ICMPv6Checksum(icmp, localAddress, destAddress, buffer.VectorisedView{}))
+
+	pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{
+		ReserveHeaderBytes: int(mld.ep.MaxHeaderLength()),
+		Data:               buffer.View(icmp).ToVectorisedView(),
+	})
+
+	mld.ep.addIPHeader(localAddress, destAddress, pkt, stack.NetworkHeaderParams{
+		Protocol: header.ICMPv6ProtocolNumber,
+		TTL:      header.MLDHopLimit,
+	})
+	// TODO(b/162198658): set the ROUTER_ALERT option when sending Host
+	// Membership Reports.
+	if err := mld.ep.nic.WritePacketToRemote(header.EthernetAddressFromMulticastIPv6Address(destAddress), nil /* gso */, ProtocolNumber, pkt); err != nil {
+		sentStats.Dropped.Increment()
+		return err
+	}
+	mldStat.Increment()
+	return nil
+}
diff --git a/pkg/tcpip/network/ipv6/ndp.go b/pkg/tcpip/network/ipv6/ndp.go
index c138358af..b83e9d931 100644
--- a/pkg/tcpip/network/ipv6/ndp.go
+++ b/pkg/tcpip/network/ipv6/ndp.go
@@ -773,7 +773,7 @@ func (ndp *ndpState) sendDADPacket(addr tcpip.Address, addressEndpoint stack.Add
 
 	icmpData := header.ICMPv6(buffer.NewView(header.ICMPv6NeighborSolicitMinimumSize))
 	icmpData.SetType(header.ICMPv6NeighborSolicit)
-	ns := header.NDPNeighborSolicit(icmpData.NDPPayload())
+	ns := header.NDPNeighborSolicit(icmpData.MessageBody())
 	ns.SetTargetAddress(addr)
 	icmpData.SetChecksum(header.ICMPv6Checksum(icmpData, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{}))
 
@@ -1944,7 +1944,7 @@ func (ndp *ndpState) startSolicitingRouters() {
 		payloadSize := header.ICMPv6HeaderSize + header.NDPRSMinimumSize + int(optsSerializer.Length())
 		icmpData := header.ICMPv6(buffer.NewView(payloadSize))
 		icmpData.SetType(header.ICMPv6RouterSolicit)
-		rs := header.NDPRouterSolicit(icmpData.NDPPayload())
+		rs := header.NDPRouterSolicit(icmpData.MessageBody())
 		rs.Options().Serialize(optsSerializer)
 		icmpData.SetChecksum(header.ICMPv6Checksum(icmpData, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{}))
 
diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go
index 40f6e8aa9..6ed00e74f 100644
--- a/pkg/tcpip/tcpip.go
+++ b/pkg/tcpip/tcpip.go
@@ -1373,6 +1373,18 @@ type ICMPv6PacketStats struct {
 	// RedirectMsg is the total number of ICMPv6 redirect message packets
 	// counted.
 	RedirectMsg *StatCounter
+
+	// MulticastListenerQuery is the total number of Multicast Listener Query
+	// messages counted.
+	MulticastListenerQuery *StatCounter
+
+	// MulticastListenerReport is the total number of Multicast Listener Report
+	// messages counted.
+	MulticastListenerReport *StatCounter
+
+	// MulticastListenerDone is the total number of Multicast Listener Done
+	// messages counted.
+	MulticastListenerDone *StatCounter
 }
 
 // ICMPv4SentPacketStats collects outbound ICMPv4-specific stats.
@@ -1414,6 +1426,10 @@ type ICMPv6SentPacketStats struct {
 type ICMPv6ReceivedPacketStats struct {
 	ICMPv6PacketStats
 
+	// Unrecognized is the total number of ICMPv6 packets received that the
+	// transport layer does not know how to parse.
+	Unrecognized *StatCounter
+
 	// Invalid is the total number of ICMPv6 packets received that the
 	// transport layer could not parse.
 	Invalid *StatCounter