cloudflared-mirror/ingress/packet_router.go

package ingress

import (
	"context"
	"fmt"
	"net/netip"

	"github.com/rs/zerolog"
	"go.opentelemetry.io/otel/attribute"
	"go.opentelemetry.io/otel/trace"

	"github.com/cloudflare/cloudflared/packet"
	quicpogs "github.com/cloudflare/cloudflared/quic"
	"github.com/cloudflare/cloudflared/tracing"
)

// Upstream of raw packets
type muxer interface {
	SendPacket(pk quicpogs.Packet) error
	// ReceivePacket waits for the next raw packet from upstream
	ReceivePacket(ctx context.Context) (quicpogs.Packet, error)
}

// PacketRouter routes packets between Upstream and ICMPRouter. Currently it rejects all other type of ICMP packets
type PacketRouter struct {
	globalConfig *GlobalRouterConfig
	muxer        muxer
	logger       *zerolog.Logger
	icmpDecoder  *packet.ICMPDecoder
	encoder      *packet.Encoder
}

// GlobalRouterConfig is the configuration shared by all instance of Router.
type GlobalRouterConfig struct {
	ICMPRouter *icmpRouter
	IPv4Src    netip.Addr
	IPv6Src    netip.Addr
	Zone       string
}

// NewPacketRouter creates a PacketRouter that handles ICMP packets. Packets are read from muxer but dropped if globalConfig is nil.
func NewPacketRouter(globalConfig *GlobalRouterConfig, muxer muxer, logger *zerolog.Logger) *PacketRouter {
	return &PacketRouter{
		globalConfig: globalConfig,
		muxer:        muxer,
		logger:       logger,
		icmpDecoder:  packet.NewICMPDecoder(),
		encoder:      packet.NewEncoder(),
	}
}

func (r *PacketRouter) Serve(ctx context.Context) error {
	for {
		rawPacket, responder, err := r.nextPacket(ctx)
		if err != nil {
			return err
		}
		r.handlePacket(ctx, rawPacket, responder)
	}
}

func (r *PacketRouter) nextPacket(ctx context.Context) (packet.RawPacket, *packetResponder, error) {
	pk, err := r.muxer.ReceivePacket(ctx)
	if err != nil {
		return packet.RawPacket{}, nil, err
	}
	responder := &packetResponder{
		datagramMuxer: r.muxer,
	}
	switch pk.Type() {
	case quicpogs.DatagramTypeIP:
		return packet.RawPacket{Data: pk.Payload()}, responder, nil
	case quicpogs.DatagramTypeIPWithTrace:
		var identity tracing.Identity
		if err := identity.UnmarshalBinary(pk.Metadata()); err != nil {
			r.logger.Err(err).Bytes("tracingIdentity", pk.Metadata()).Msg("Failed to unmarshal tracing identity")
		} else {
			responder.tracedCtx = tracing.NewTracedContext(ctx, identity.String(), r.logger)
			responder.serializedIdentity = pk.Metadata()
		}
		return packet.RawPacket{Data: pk.Payload()}, responder, nil
	default:
		return packet.RawPacket{}, nil, fmt.Errorf("unexpected datagram type %d", pk.Type())
	}
}

func (r *PacketRouter) handlePacket(ctx context.Context, rawPacket packet.RawPacket, responder *packetResponder) {
	// ICMP Proxy feature is disabled, drop packets
	if r.globalConfig == nil {
		return
	}

	icmpPacket, err := r.icmpDecoder.Decode(rawPacket)
	if err != nil {
		r.logger.Err(err).Msg("Failed to decode ICMP packet from quic datagram")
		return
	}

	if icmpPacket.TTL <= 1 {
		if err := r.sendTTLExceedMsg(ctx, icmpPacket, rawPacket, r.encoder); err != nil {
			r.logger.Err(err).Msg("Failed to return ICMP TTL exceed error")
		}
		return
	}
	icmpPacket.TTL--

	if err := r.globalConfig.ICMPRouter.Request(ctx, icmpPacket, responder); err != nil {
		r.logger.Err(err).
			Str("src", icmpPacket.Src.String()).
			Str("dst", icmpPacket.Dst.String()).
			Interface("type", icmpPacket.Type).
			Msg("Failed to send ICMP packet")
	}
}

func (r *PacketRouter) sendTTLExceedMsg(ctx context.Context, pk *packet.ICMP, rawPacket packet.RawPacket, encoder *packet.Encoder) error {
	var srcIP netip.Addr
	if pk.Dst.Is4() {
		srcIP = r.globalConfig.IPv4Src
	} else {
		srcIP = r.globalConfig.IPv6Src
	}
	ttlExceedPacket := packet.NewICMPTTLExceedPacket(pk.IP, rawPacket, srcIP)

	encodedTTLExceed, err := encoder.Encode(ttlExceedPacket)
	if err != nil {
		return err
	}
	return r.muxer.SendPacket(quicpogs.RawPacket(encodedTTLExceed))
}

// packetResponder should not be used concurrently. This assumption is upheld because reply packets are ready one-by-one
type packetResponder struct {
	datagramMuxer      muxer
	tracedCtx          *tracing.TracedContext
	serializedIdentity []byte
	// hadReply tracks if there has been any reply for this flow
	hadReply bool
}

func (pr *packetResponder) tracingEnabled() bool {
	return pr.tracedCtx != nil
}

func (pr *packetResponder) returnPacket(rawPacket packet.RawPacket) error {
	pr.hadReply = true
	return pr.datagramMuxer.SendPacket(quicpogs.RawPacket(rawPacket))
}

func (pr *packetResponder) requestSpan(ctx context.Context, pk *packet.ICMP) (context.Context, trace.Span) {
	if !pr.tracingEnabled() {
		return ctx, tracing.NewNoopSpan()
	}
	return pr.tracedCtx.Tracer().Start(pr.tracedCtx, "icmp-echo-request", trace.WithAttributes(
		attribute.String("src", pk.Src.String()),
		attribute.String("dst", pk.Dst.String()),
	))
}

func (pr *packetResponder) replySpan(ctx context.Context, logger *zerolog.Logger) (context.Context, trace.Span) {
	if !pr.tracingEnabled() || pr.hadReply {
		return ctx, tracing.NewNoopSpan()
	}
	return pr.tracedCtx.Tracer().Start(pr.tracedCtx, "icmp-echo-reply")
}

func (pr *packetResponder) exportSpan() {
	if !pr.tracingEnabled() {
		return
	}
	spans := pr.tracedCtx.GetProtoSpans()
	if len(spans) > 0 {
		pr.datagramMuxer.SendPacket(&quicpogs.TracingSpanPacket{
			Spans:           spans,
			TracingIdentity: pr.serializedIdentity,
		})
	}
}