Merge branch 'master' into ip-forwarding

- Merges aleksej-paschenko's with HEAD
- Adds vfs2 support for ip_forward

1 files changed, 380 insertions, 60 deletions

diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go
index 5898f8f9e..0eafe9790 100644
--- a/pkg/tcpip/network/ipv6/ipv6.go
+++ b/pkg/tcpip/network/ipv6/ipv6.go
@@ -21,11 +21,13 @@
 package ipv6
 
 import (
+	"fmt"
 	"sync/atomic"
 
 	"gvisor.dev/gvisor/pkg/tcpip"
 	"gvisor.dev/gvisor/pkg/tcpip/buffer"
 	"gvisor.dev/gvisor/pkg/tcpip/header"
+	"gvisor.dev/gvisor/pkg/tcpip/network/fragmentation"
 	"gvisor.dev/gvisor/pkg/tcpip/stack"
 )
 
@@ -43,13 +45,12 @@ const (
 )
 
 type endpoint struct {
-	nicid         tcpip.NICID
-	id            stack.NetworkEndpointID
-	prefixLen     int
+	nicID         tcpip.NICID
 	linkEP        stack.LinkEndpoint
 	linkAddrCache stack.LinkAddressCache
 	dispatcher    stack.TransportDispatcher
 	protocol      *protocol
+	stack         *stack.Stack
 }
 
 // DefaultTTL is the default hop limit for this endpoint.
@@ -65,17 +66,7 @@ func (e *endpoint) MTU() uint32 {
 
 // NICID returns the ID of the NIC this endpoint belongs to.
 func (e *endpoint) NICID() tcpip.NICID {
-	return e.nicid
-}
-
-// ID returns the ipv6 endpoint ID.
-func (e *endpoint) ID() *stack.NetworkEndpointID {
-	return &e.id
-}
-
-// PrefixLen returns the ipv6 endpoint subnet prefix length in bits.
-func (e *endpoint) PrefixLen() int {
-	return e.prefixLen
+	return e.nicID
 }
 
 // Capabilities implements stack.NetworkEndpoint.Capabilities.
@@ -97,9 +88,9 @@ func (e *endpoint) GSOMaxSize() uint32 {
 	return 0
 }
 
-func (e *endpoint) addIPHeader(r *stack.Route, hdr *buffer.Prependable, payloadSize int, params stack.NetworkHeaderParams) {
-	length := uint16(hdr.UsedLength() + payloadSize)
-	ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize))
+func (e *endpoint) addIPHeader(r *stack.Route, pkt *stack.PacketBuffer, params stack.NetworkHeaderParams) {
+	length := uint16(pkt.Size())
+	ip := header.IPv6(pkt.NetworkHeader().Push(header.IPv6MinimumSize))
 	ip.Encode(&header.IPv6Fields{
 		PayloadLength: length,
 		NextHeader:    uint8(params.Protocol),
@@ -108,86 +99,336 @@ func (e *endpoint) addIPHeader(r *stack.Route, hdr *buffer.Prependable, payloadS
 		SrcAddr:       r.LocalAddress,
 		DstAddr:       r.RemoteAddress,
 	})
+	pkt.NetworkProtocolNumber = header.IPv6ProtocolNumber
 }
 
 // WritePacket writes a packet to the given destination address and protocol.
-func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) *tcpip.Error {
-	e.addIPHeader(r, &hdr, payload.Size(), params)
-
-	if loop&stack.PacketLoop != 0 {
-		views := make([]buffer.View, 1, 1+len(payload.Views()))
-		views[0] = hdr.View()
-		views = append(views, payload.Views()...)
-		vv := buffer.NewVectorisedView(len(views[0])+payload.Size(), views)
+func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, pkt *stack.PacketBuffer) *tcpip.Error {
+	e.addIPHeader(r, pkt, params)
+
+	if r.Loop&stack.PacketLoop != 0 {
 		loopedR := r.MakeLoopedRoute()
-		e.HandlePacket(&loopedR, vv)
+
+		e.HandlePacket(&loopedR, stack.NewPacketBuffer(stack.PacketBufferOptions{
+			// The inbound path expects an unparsed packet.
+			Data: buffer.NewVectorisedView(pkt.Size(), pkt.Views()),
+		}))
+
 		loopedR.Release()
 	}
-	if loop&stack.PacketOut == 0 {
+	if r.Loop&stack.PacketOut == 0 {
 		return nil
 	}
 
 	r.Stats().IP.PacketsSent.Increment()
-	return e.linkEP.WritePacket(r, gso, hdr, payload, ProtocolNumber)
+	return e.linkEP.WritePacket(r, gso, ProtocolNumber, pkt)
 }
 
 // WritePackets implements stack.LinkEndpoint.WritePackets.
-func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) (int, *tcpip.Error) {
-	if loop&stack.PacketLoop != 0 {
+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 {
 		panic("not implemented")
 	}
-	if loop&stack.PacketOut == 0 {
-		return len(hdrs), nil
+	if r.Loop&stack.PacketOut == 0 {
+		return pkts.Len(), nil
 	}
 
-	for i := range hdrs {
-		hdr := &hdrs[i].Hdr
-		size := hdrs[i].Size
-		e.addIPHeader(r, hdr, size, params)
+	for pb := pkts.Front(); pb != nil; pb = pb.Next() {
+		e.addIPHeader(r, pb, params)
 	}
 
-	n, err := e.linkEP.WritePackets(r, gso, hdrs, payload, ProtocolNumber)
+	n, err := e.linkEP.WritePackets(r, gso, pkts, ProtocolNumber)
 	r.Stats().IP.PacketsSent.IncrementBy(uint64(n))
 	return n, err
 }
 
 // WriteHeaderIncludedPacker implements stack.NetworkEndpoint. It is not yet
 // supported by IPv6.
-func (*endpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.VectorisedView, loop stack.PacketLooping) *tcpip.Error {
-	// TODO(b/119580726): Support IPv6 header-included packets.
+func (*endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt *stack.PacketBuffer) *tcpip.Error {
+	// TODO(b/146666412): Support IPv6 header-included packets.
 	return tcpip.ErrNotSupported
 }
 
 // HandlePacket is called by the link layer when new ipv6 packets arrive for
 // this endpoint.
-func (e *endpoint) HandlePacket(r *stack.Route, vv buffer.VectorisedView) {
-	headerView := vv.First()
-	h := header.IPv6(headerView)
-	if !h.IsValid(vv.Size()) {
+func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) {
+	h := header.IPv6(pkt.NetworkHeader().View())
+	if !h.IsValid(pkt.Data.Size() + pkt.NetworkHeader().View().Size() + pkt.TransportHeader().View().Size()) {
+		r.Stats().IP.MalformedPacketsReceived.Increment()
 		return
 	}
 
-	vv.TrimFront(header.IPv6MinimumSize)
-	vv.CapLength(int(h.PayloadLength()))
-
-	p := h.TransportProtocol()
-	if p == header.ICMPv6ProtocolNumber {
-		e.handleICMP(r, headerView, vv)
-		return
+	// vv consists of:
+	// - Any IPv6 header bytes after the first 40 (i.e. extensions).
+	// - The transport header, if present.
+	// - Any other payload data.
+	vv := pkt.NetworkHeader().View()[header.IPv6MinimumSize:].ToVectorisedView()
+	vv.AppendView(pkt.TransportHeader().View())
+	vv.Append(pkt.Data)
+	it := header.MakeIPv6PayloadIterator(header.IPv6ExtensionHeaderIdentifier(h.NextHeader()), vv)
+	hasFragmentHeader := false
+
+	for firstHeader := true; ; firstHeader = false {
+		extHdr, done, err := it.Next()
+		if err != nil {
+			r.Stats().IP.MalformedPacketsReceived.Increment()
+			return
+		}
+		if done {
+			break
+		}
+
+		switch extHdr := extHdr.(type) {
+		case header.IPv6HopByHopOptionsExtHdr:
+			// As per RFC 8200 section 4.1, the Hop By Hop extension header is
+			// restricted to appear immediately after an IPv6 fixed header.
+			//
+			// TODO(b/152019344): Send an ICMPv6 Parameter Problem, Code 1
+			// (unrecognized next header) error in response to an extension header's
+			// Next Header field with the Hop By Hop extension header identifier.
+			if !firstHeader {
+				return
+			}
+
+			optsIt := extHdr.Iter()
+
+			for {
+				opt, done, err := optsIt.Next()
+				if err != nil {
+					r.Stats().IP.MalformedPacketsReceived.Increment()
+					return
+				}
+				if done {
+					break
+				}
+
+				// We currently do not support any IPv6 Hop By Hop extension header
+				// options.
+				switch opt.UnknownAction() {
+				case header.IPv6OptionUnknownActionSkip:
+				case header.IPv6OptionUnknownActionDiscard:
+					return
+				case header.IPv6OptionUnknownActionDiscardSendICMP:
+					// TODO(b/152019344): Send an ICMPv6 Parameter Problem Code 2 for
+					// unrecognized IPv6 extension header options.
+					return
+				case header.IPv6OptionUnknownActionDiscardSendICMPNoMulticastDest:
+					// TODO(b/152019344): Send an ICMPv6 Parameter Problem Code 2 for
+					// unrecognized IPv6 extension header options.
+					return
+				default:
+					panic(fmt.Sprintf("unrecognized action for an unrecognized Hop By Hop extension header option = %d", opt))
+				}
+			}
+
+		case header.IPv6RoutingExtHdr:
+			// As per RFC 8200 section 4.4, if a node encounters a routing header with
+			// an unrecognized routing type value, with a non-zero Segments Left
+			// value, the node must discard the packet and send an ICMP Parameter
+			// Problem, Code 0. If the Segments Left is 0, the node must ignore the
+			// Routing extension header and process the next header in the packet.
+			//
+			// Note, the stack does not yet handle any type of routing extension
+			// header, so we just make sure Segments Left is zero before processing
+			// the next extension header.
+			//
+			// TODO(b/152019344): Send an ICMPv6 Parameter Problem Code 0 for
+			// unrecognized routing types with a non-zero Segments Left value.
+			if extHdr.SegmentsLeft() != 0 {
+				return
+			}
+
+		case header.IPv6FragmentExtHdr:
+			hasFragmentHeader = true
+
+			if extHdr.IsAtomic() {
+				// This fragment extension header indicates that this packet is an
+				// atomic fragment. An atomic fragment is a fragment that contains
+				// all the data required to reassemble a full packet. As per RFC 6946,
+				// atomic fragments must not interfere with "normal" fragmented traffic
+				// so we skip processing the fragment instead of feeding it through the
+				// reassembly process below.
+				continue
+			}
+
+			// Don't consume the iterator if we have the first fragment because we
+			// will use it to validate that the first fragment holds the upper layer
+			// header.
+			rawPayload := it.AsRawHeader(extHdr.FragmentOffset() != 0 /* consume */)
+
+			if extHdr.FragmentOffset() == 0 {
+				// Check that the iterator ends with a raw payload as the first fragment
+				// should include all headers up to and including any upper layer
+				// headers, as per RFC 8200 section 4.5; only upper layer data
+				// (non-headers) should follow the fragment extension header.
+				var lastHdr header.IPv6PayloadHeader
+
+				for {
+					it, done, err := it.Next()
+					if err != nil {
+						r.Stats().IP.MalformedPacketsReceived.Increment()
+						r.Stats().IP.MalformedPacketsReceived.Increment()
+						return
+					}
+					if done {
+						break
+					}
+
+					lastHdr = it
+				}
+
+				// If the last header is a raw header, then the last portion of the IPv6
+				// payload is not a known IPv6 extension header. Note, this does not
+				// mean that the last portion is an upper layer header or not an
+				// extension header because:
+				//  1) we do not yet support all extension headers
+				//  2) we do not validate the upper layer header before reassembling.
+				//
+				// This check makes sure that a known IPv6 extension header is not
+				// present after the Fragment extension header in a non-initial
+				// fragment.
+				//
+				// TODO(#2196): Support IPv6 Authentication and Encapsulated
+				// Security Payload extension headers.
+				// TODO(#2333): Validate that the upper layer header is valid.
+				switch lastHdr.(type) {
+				case header.IPv6RawPayloadHeader:
+				default:
+					r.Stats().IP.MalformedPacketsReceived.Increment()
+					r.Stats().IP.MalformedFragmentsReceived.Increment()
+					return
+				}
+			}
+
+			fragmentPayloadLen := rawPayload.Buf.Size()
+			if fragmentPayloadLen == 0 {
+				// Drop the packet as it's marked as a fragment but has no payload.
+				r.Stats().IP.MalformedPacketsReceived.Increment()
+				r.Stats().IP.MalformedFragmentsReceived.Increment()
+				return
+			}
+
+			// The packet is a fragment, let's try to reassemble it.
+			start := extHdr.FragmentOffset() * header.IPv6FragmentExtHdrFragmentOffsetBytesPerUnit
+			last := start + uint16(fragmentPayloadLen) - 1
+
+			// Drop the packet if the fragmentOffset is incorrect. i.e the
+			// combination of fragmentOffset and pkt.Data.size() causes a
+			// wrap around resulting in last being less than the offset.
+			if last < start {
+				r.Stats().IP.MalformedPacketsReceived.Increment()
+				r.Stats().IP.MalformedFragmentsReceived.Increment()
+				return
+			}
+
+			var ready bool
+			// Note that pkt doesn't have its transport header set after reassembly,
+			// and won't until DeliverNetworkPacket sets it.
+			pkt.Data, ready, err = e.protocol.fragmentation.Process(
+				// IPv6 ignores the Protocol field since the ID only needs to be unique
+				// across source-destination pairs, as per RFC 8200 section 4.5.
+				fragmentation.FragmentID{
+					Source:      h.SourceAddress(),
+					Destination: h.DestinationAddress(),
+					ID:          extHdr.ID(),
+				},
+				start,
+				last,
+				extHdr.More(),
+				rawPayload.Buf,
+			)
+			if err != nil {
+				r.Stats().IP.MalformedPacketsReceived.Increment()
+				r.Stats().IP.MalformedFragmentsReceived.Increment()
+				return
+			}
+
+			if ready {
+				// We create a new iterator with the reassembled packet because we could
+				// have more extension headers in the reassembled payload, as per RFC
+				// 8200 section 4.5.
+				it = header.MakeIPv6PayloadIterator(rawPayload.Identifier, pkt.Data)
+			}
+
+		case header.IPv6DestinationOptionsExtHdr:
+			optsIt := extHdr.Iter()
+
+			for {
+				opt, done, err := optsIt.Next()
+				if err != nil {
+					r.Stats().IP.MalformedPacketsReceived.Increment()
+					return
+				}
+				if done {
+					break
+				}
+
+				// We currently do not support any IPv6 Destination extension header
+				// options.
+				switch opt.UnknownAction() {
+				case header.IPv6OptionUnknownActionSkip:
+				case header.IPv6OptionUnknownActionDiscard:
+					return
+				case header.IPv6OptionUnknownActionDiscardSendICMP:
+					// TODO(b/152019344): Send an ICMPv6 Parameter Problem Code 2 for
+					// unrecognized IPv6 extension header options.
+					return
+				case header.IPv6OptionUnknownActionDiscardSendICMPNoMulticastDest:
+					// TODO(b/152019344): Send an ICMPv6 Parameter Problem Code 2 for
+					// unrecognized IPv6 extension header options.
+					return
+				default:
+					panic(fmt.Sprintf("unrecognized action for an unrecognized Destination extension header option = %d", opt))
+				}
+			}
+
+		case header.IPv6RawPayloadHeader:
+			// If the last header in the payload isn't a known IPv6 extension header,
+			// handle it as if it is transport layer data.
+
+			// For unfragmented packets, extHdr still contains the transport header.
+			// Get rid of it.
+			//
+			// For reassembled fragments, pkt.TransportHeader is unset, so this is a
+			// no-op and pkt.Data begins with the transport header.
+			extHdr.Buf.TrimFront(pkt.TransportHeader().View().Size())
+			pkt.Data = extHdr.Buf
+
+			if p := tcpip.TransportProtocolNumber(extHdr.Identifier); p == header.ICMPv6ProtocolNumber {
+				e.handleICMP(r, pkt, hasFragmentHeader)
+			} else {
+				r.Stats().IP.PacketsDelivered.Increment()
+				// TODO(b/152019344): Send an ICMPv6 Parameter Problem, Code 1 error
+				// in response to unrecognized next header values.
+				e.dispatcher.DeliverTransportPacket(r, p, pkt)
+			}
+
+		default:
+			// If we receive a packet for an extension header we do not yet handle,
+			// drop the packet for now.
+			//
+			// TODO(b/152019344): Send an ICMPv6 Parameter Problem, Code 1 error
+			// in response to unrecognized next header values.
+			r.Stats().UnknownProtocolRcvdPackets.Increment()
+			return
+		}
 	}
-
-	r.Stats().IP.PacketsDelivered.Increment()
-	e.dispatcher.DeliverTransportPacket(r, p, headerView, vv)
 }
 
 // Close cleans up resources associated with the endpoint.
 func (*endpoint) Close() {}
 
+// NetworkProtocolNumber implements stack.NetworkEndpoint.NetworkProtocolNumber.
+func (e *endpoint) NetworkProtocolNumber() tcpip.NetworkProtocolNumber {
+	return e.protocol.Number()
+}
+
 type protocol struct {
 	// defaultTTL is the current default TTL for the protocol. Only the
 	// uint8 portion of it is meaningful and it must be accessed
 	// atomically.
-	defaultTTL uint32
+	defaultTTL    uint32
+	fragmentation *fragmentation.Fragmentation
 }
 
 // Number returns the ipv6 protocol number.
@@ -212,16 +453,15 @@ func (*protocol) ParseAddresses(v buffer.View) (src, dst tcpip.Address) {
 }
 
 // NewEndpoint creates a new ipv6 endpoint.
-func (p *protocol) NewEndpoint(nicid tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache stack.LinkAddressCache, dispatcher stack.TransportDispatcher, linkEP stack.LinkEndpoint) (stack.NetworkEndpoint, *tcpip.Error) {
+func (p *protocol) NewEndpoint(nicID tcpip.NICID, linkAddrCache stack.LinkAddressCache, dispatcher stack.TransportDispatcher, linkEP stack.LinkEndpoint, st *stack.Stack) stack.NetworkEndpoint {
 	return &endpoint{
-		nicid:         nicid,
-		id:            stack.NetworkEndpointID{LocalAddress: addrWithPrefix.Address},
-		prefixLen:     addrWithPrefix.PrefixLen,
+		nicID:         nicID,
 		linkEP:        linkEP,
 		linkAddrCache: linkAddrCache,
 		dispatcher:    dispatcher,
 		protocol:      p,
-	}, nil
+		stack:         st,
+	}
 }
 
 // SetOption implements NetworkProtocol.SetOption.
@@ -256,6 +496,83 @@ func (p *protocol) DefaultTTL() uint8 {
 	return uint8(atomic.LoadUint32(&p.defaultTTL))
 }
 
+// Close implements stack.TransportProtocol.Close.
+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.IPv6MinimumSize)
+	if !ok {
+		return 0, false, false
+	}
+	ipHdr := header.IPv6(hdr)
+
+	// dataClone consists of:
+	// - Any IPv6 header bytes after the first 40 (i.e. extensions).
+	// - The transport header, if present.
+	// - Any other payload data.
+	views := [8]buffer.View{}
+	dataClone := pkt.Data.Clone(views[:])
+	dataClone.TrimFront(header.IPv6MinimumSize)
+	it := header.MakeIPv6PayloadIterator(header.IPv6ExtensionHeaderIdentifier(ipHdr.NextHeader()), dataClone)
+
+	// Iterate over the IPv6 extensions to find their length.
+	//
+	// Parsing occurs again in HandlePacket because we don't track the
+	// extensions in PacketBuffer. Unfortunately, that means HandlePacket
+	// has to do the parsing work again.
+	var nextHdr tcpip.TransportProtocolNumber
+	foundNext := true
+	extensionsSize := 0
+traverseExtensions:
+	for extHdr, done, err := it.Next(); ; extHdr, done, err = it.Next() {
+		if err != nil {
+			break
+		}
+		// If we exhaust the extension list, the entire packet is the IPv6 header
+		// and (possibly) extensions.
+		if done {
+			extensionsSize = dataClone.Size()
+			foundNext = false
+			break
+		}
+
+		switch extHdr := extHdr.(type) {
+		case header.IPv6FragmentExtHdr:
+			// If this is an atomic fragment, we don't have to treat it specially.
+			if !extHdr.More() && extHdr.FragmentOffset() == 0 {
+				continue
+			}
+			// This is a non-atomic fragment and has to be re-assembled before we can
+			// examine the payload for a transport header.
+			foundNext = false
+
+		case header.IPv6RawPayloadHeader:
+			// We've found the payload after any extensions.
+			extensionsSize = dataClone.Size() - extHdr.Buf.Size()
+			nextHdr = tcpip.TransportProtocolNumber(extHdr.Identifier)
+			break traverseExtensions
+
+		default:
+			// Any other extension is a no-op, keep looping until we find the payload.
+		}
+	}
+
+	// Put the IPv6 header with extensions in pkt.NetworkHeader().
+	hdr, ok = pkt.NetworkHeader().Consume(header.IPv6MinimumSize + extensionsSize)
+	if !ok {
+		panic(fmt.Sprintf("pkt.Data should have at least %d bytes, but only has %d.", header.IPv6MinimumSize+extensionsSize, pkt.Data.Size()))
+	}
+	ipHdr = header.IPv6(hdr)
+	pkt.Data.CapLength(int(ipHdr.PayloadLength()))
+	pkt.NetworkProtocolNumber = header.IPv6ProtocolNumber
+
+	return nextHdr, foundNext, true
+}
+
 // calculateMTU calculates the network-layer payload MTU based on the link-layer
 // payload mtu.
 func calculateMTU(mtu uint32) uint32 {
@@ -268,5 +585,8 @@ func calculateMTU(mtu uint32) uint32 {
 
 // NewProtocol returns an IPv6 network protocol.
 func NewProtocol() stack.NetworkProtocol {
-	return &protocol{defaultTTL: DefaultTTL}
+	return &protocol{
+		defaultTTL:    DefaultTTL,
+		fragmentation: fragmentation.NewFragmentation(header.IPv6FragmentExtHdrFragmentOffsetBytesPerUnit, fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, fragmentation.DefaultReassembleTimeout),
+	}
 }