Queue packets in WritePackets when resolving link address

Test: integration_test.TestWritePacketsLinkResolution

Fixes #4458.

PiperOrigin-RevId: 353108826

2 files changed, 74 insertions, 14 deletions

diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go
index d19150a20..447b5c99d 100644
--- a/pkg/tcpip/stack/nic.go
+++ b/pkg/tcpip/stack/nic.go
@@ -358,16 +358,43 @@ func (n *NIC) writePacket(r RouteInfo, gso *GSO, protocol tcpip.NetworkProtocolN
 
 // WritePackets implements NetworkLinkEndpoint.
 func (n *NIC) WritePackets(r *Route, gso *GSO, pkts PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) {
-	// TODO(gvisor.dev/issue/4458): Queue packets whie link address resolution
-	// is being peformed like WritePacket.
-	routeInfo := r.Fields()
+	// As per relevant RFCs, we should queue packets while we wait for link
+	// resolution to complete.
+	//
+	// RFC 1122 section 2.3.2.2 (for IPv4):
+	//   The link layer SHOULD save (rather than discard) at least
+	//   one (the latest) packet of each set of packets destined to
+	//   the same unresolved IP address, and transmit the saved
+	//   packet when the address has been resolved.
+	//
+	// RFC 4861 section 7.2.2 (for IPv6):
+	//   While waiting for address resolution to complete, the sender MUST, for
+	//   each neighbor, retain a small queue of packets waiting for address
+	//   resolution to complete. The queue MUST hold at least one packet, and MAY
+	//   contain more. However, the number of queued packets per neighbor SHOULD
+	//   be limited to some small value. When a queue overflows, the new arrival
+	//   SHOULD replace the oldest entry. Once address resolution completes, the
+	//   node transmits any queued packets.
+	if ch, err := r.Resolve(nil); err != nil {
+		if err == tcpip.ErrWouldBlock {
+			r.Acquire()
+			n.linkResQueue.enqueue(ch, r, protocol, &pkts)
+			return pkts.Len(), nil
+		}
+		return 0, err
+	}
+
+	return n.writePackets(r.Fields(), gso, protocol, pkts)
+}
+
+func (n *NIC) writePackets(r RouteInfo, gso *GSO, protocol tcpip.NetworkProtocolNumber, pkts PacketBufferList) (int, *tcpip.Error) {
 	for pkt := pkts.Front(); pkt != nil; pkt = pkt.Next() {
-		pkt.EgressRoute = routeInfo
+		pkt.EgressRoute = r
 		pkt.GSOOptions = gso
 		pkt.NetworkProtocolNumber = protocol
 	}
 
-	writtenPackets, err := n.LinkEndpoint.WritePackets(routeInfo, gso, pkts, protocol)
+	writtenPackets, err := n.LinkEndpoint.WritePackets(r, gso, pkts, protocol)
 	n.stats.Tx.Packets.IncrementBy(uint64(writtenPackets))
 	writtenBytes := 0
 	for i, pb := 0, pkts.Front(); i < writtenPackets && pb != nil; i, pb = i+1, pb.Next() {
diff --git a/pkg/tcpip/stack/pending_packets.go b/pkg/tcpip/stack/pending_packets.go
index 81d8ff6e8..3ac039c7d 100644
--- a/pkg/tcpip/stack/pending_packets.go
+++ b/pkg/tcpip/stack/pending_packets.go
@@ -28,10 +28,26 @@ const (
 	maxPendingPacketsPerResolution = 256
 )
 
+// pendingPacketBuffer is a pending packet buffer.
+//
+// TODO(gvisor.dev/issue/5331): Drop this when we drop WritePacket and only use
+// WritePackets so we can use a PacketBufferList everywhere.
+type pendingPacketBuffer interface {
+	len() int
+}
+
+func (*PacketBuffer) len() int {
+	return 1
+}
+
+func (p *PacketBufferList) len() int {
+	return p.Len()
+}
+
 type pendingPacket struct {
 	route *Route
 	proto tcpip.NetworkProtocolNumber
-	pkt   *PacketBuffer
+	pkt   pendingPacketBuffer
 }
 
 // packetsPendingLinkResolution is a queue of packets pending link resolution.
@@ -54,16 +70,17 @@ func (f *packetsPendingLinkResolution) init() {
 	f.packets = make(map[<-chan struct{}][]pendingPacket)
 }
 
-func incrementOutgoingPacketErrors(r *Route, proto tcpip.NetworkProtocolNumber) {
-	r.Stats().IP.OutgoingPacketErrors.Increment()
+func incrementOutgoingPacketErrors(r *Route, proto tcpip.NetworkProtocolNumber, pkt pendingPacketBuffer) {
+	n := uint64(pkt.len())
+	r.Stats().IP.OutgoingPacketErrors.IncrementBy(n)
 
 	// ok may be false if the endpoint's stats do not collect IP-related data.
 	if ipEndpointStats, ok := r.outgoingNIC.getNetworkEndpoint(proto).Stats().(IPNetworkEndpointStats); ok {
-		ipEndpointStats.IPStats().OutgoingPacketErrors.Increment()
+		ipEndpointStats.IPStats().OutgoingPacketErrors.IncrementBy(n)
 	}
 }
 
-func (f *packetsPendingLinkResolution) enqueue(ch <-chan struct{}, r *Route, proto tcpip.NetworkProtocolNumber, pkt *PacketBuffer) {
+func (f *packetsPendingLinkResolution) enqueue(ch <-chan struct{}, r *Route, proto tcpip.NetworkProtocolNumber, pkt pendingPacketBuffer) {
 	f.Lock()
 	defer f.Unlock()
 
@@ -73,7 +90,7 @@ func (f *packetsPendingLinkResolution) enqueue(ch <-chan struct{}, r *Route, pro
 		packets[0] = pendingPacket{}
 		packets = packets[1:]
 
-		incrementOutgoingPacketErrors(r, proto)
+		incrementOutgoingPacketErrors(r, proto, p.pkt)
 
 		p.route.Release()
 	}
@@ -113,13 +130,29 @@ func (f *packetsPendingLinkResolution) enqueue(ch <-chan struct{}, r *Route, pro
 
 		for _, p := range packets {
 			if cancelled || p.route.IsResolutionRequired() {
-				incrementOutgoingPacketErrors(r, proto)
+				incrementOutgoingPacketErrors(r, proto, p.pkt)
 
 				if linkResolvableEP, ok := p.route.outgoingNIC.getNetworkEndpoint(p.route.NetProto).(LinkResolvableNetworkEndpoint); ok {
-					linkResolvableEP.HandleLinkResolutionFailure(pkt)
+					switch pkt := p.pkt.(type) {
+					case *PacketBuffer:
+						linkResolvableEP.HandleLinkResolutionFailure(pkt)
+					case *PacketBufferList:
+						for pb := pkt.Front(); pb != nil; pb = pb.Next() {
+							linkResolvableEP.HandleLinkResolutionFailure(pb)
+						}
+					default:
+						panic(fmt.Sprintf("unrecognized pending packet buffer type = %T", p.pkt))
+					}
 				}
 			} else {
-				p.route.outgoingNIC.writePacket(p.route.Fields(), nil /* gso */, p.proto, p.pkt)
+				switch pkt := p.pkt.(type) {
+				case *PacketBuffer:
+					p.route.outgoingNIC.writePacket(p.route.Fields(), nil /* gso */, p.proto, pkt)
+				case *PacketBufferList:
+					p.route.outgoingNIC.writePackets(p.route.Fields(), nil /* gso */, p.proto, *pkt)
+				default:
+					panic(fmt.Sprintf("unrecognized pending packet buffer type = %T", p.pkt))
+				}
 			}
 			p.route.Release()
 		}