Add support for AF_UNIX SOCK_DGRAM

Unfortunately these aren't supported by libuv directly, so we need to
implement our own separate system. It doesn't need to be particularly
scalable as the expected use-case is to send and receive ethernet frames.

The server will listen on a Unix domain socket and receive connected
SOCK_DGRAM sockets which will have ethernet frames on them.

Signed-off-by: David Scott <[email protected]>

go/cmd/vmnet-example/main.go

@@ -0,0 +1,41 @@
+package main
+
+import (
+	"context"
+	"flag"
+	"fmt"
+	"log"
+	"os"
+
+	"github.com/google/uuid"
+	"github.com/moby/vpnkit/go/pkg/vmnet"
+)
+
+var path string
+
+func main() {
+	flag.StringVar(&path, "path", "", "path to vmnet socket")
+	flag.Parse()
+	if path == "" {
+		fmt.Fprintf(os.Stderr, "Please supply a --path argument\n")
+	}
+	vm, err := vmnet.Connect(context.Background(), vmnet.Config{
+		Path: path,
+	})
+	if err != nil {
+		log.Fatal(err)
+	}
+	defer vm.Close()
+	log.Println("connected to vmnet service")
+	u, err := uuid.NewRandom()
+	if err != nil {
+		log.Fatal(err)
+	}
+	vif, err := vm.ConnectVif(u)
+	if err != nil {
+		log.Fatal(err)
+	}
+	defer vif.Close()
+	log.Printf("VIF has IP %s", vif.IP)
+	log.Printf("SOCK_DGRAM fd: %d", vif.Ethernet.Fd)
+}

go/pkg/vmnet/datagram.go

@@ -0,0 +1,50 @@
+package vmnet
+
+/*
+// FIXME: Needed because we call C.send. Perhaps we could use syscall instead?
+#include <stdlib.h>
+#include <sys/socket.h>
+
+*/
+import "C"
+
+import (
+	"syscall"
+
+	"github.com/pkg/errors"
+)
+
+// Datagram sends and receives ethernet frames via send/recv over a SOCK_DGRAM fd.
+type Datagram struct {
+	Fd   int // Underlying SOCK_DGRAM file descriptor.
+	pcap *PcapWriter
+}
+
+func (e Datagram) Recv(buf []byte) (int, error) {
+	num, _, err := syscall.Recvfrom(e.Fd, buf, 0)
+	if e.pcap != nil {
+		if err := e.pcap.Write(buf[0:num]); err != nil {
+			return 0, errors.Wrap(err, "writing to pcap")
+		}
+	}
+	return num, err
+}
+
+func (e Datagram) Send(packet []byte) (int, error) {
+	if e.pcap != nil {
+		if err := e.pcap.Write(packet); err != nil {
+			return 0, errors.Wrap(err, "writing to pcap")
+		}
+	}
+	result, err := C.send(C.int(e.Fd), C.CBytes(packet), C.size_t(len(packet)), 0)
+	if result == -1 {
+		return 0, err
+	}
+	return len(packet), nil
+}
+
+func (e Datagram) Close() error {
+	return syscall.Close(e.Fd)
+}
+
+var _ sendReceiver = Datagram{}

go/pkg/vmnet/dhcp.go

@@ -0,0 +1,114 @@
+package vmnet
+
+import (
+	"net"
+	"time"
+)
+
+// dhcp queries the IP by DHCP
+func dhcpRequest(packet sendReceiver, clientMAC net.HardwareAddr) (net.IP, error) {
+	broadcastMAC := []byte{0xff, 0xff, 0xff, 0xff, 0xff, 0xff}
+	broadcastIP := []byte{0xff, 0xff, 0xff, 0xff}
+	unknownIP := []byte{0, 0, 0, 0}
+
+	dhcpRequest := NewDhcpRequest(clientMAC).Bytes()
+	ipv4 := NewIpv4(broadcastIP, unknownIP)
+
+	udpv4 := NewUdpv4(ipv4, 68, 67, dhcpRequest)
+	ipv4.setData(udpv4.Bytes())
+
+	ethernet := NewEthernetFrame(broadcastMAC, clientMAC, 0x800)
+	ethernet.setData(ipv4.Bytes())
+	finished := false
+	go func() {
+		for !finished {
+			if _, err := packet.Send(ethernet.Bytes()); err != nil {
+				panic(err)
+			}
+			time.Sleep(time.Second)
+		}
+	}()
+
+	buf := make([]byte, 1500)
+	for {
+		n, err := packet.Recv(buf)
+		if err != nil {
+			return nil, err
+		}
+		response := buf[0:n]
+		ethernet, err = ParseEthernetFrame(response)
+		if err != nil {
+			continue
+		}
+		for i, x := range ethernet.Dst {
+			if i > len(clientMAC) || clientMAC[i] != x {
+				// intended for someone else
+				continue
+			}
+		}
+		ipv4, err = ParseIpv4(ethernet.Data)
+		if err != nil {
+			// probably not an IPv4 packet
+			continue
+		}
+		udpv4, err = ParseUdpv4(ipv4.Data)
+		if err != nil {
+			// probably not a UDPv4 packet
+			continue
+		}
+		if udpv4.Src != 67 || udpv4.Dst != 68 {
+			// not a DHCP response
+			continue
+		}
+		if len(udpv4.Data) < 243 {
+			// truncated
+			continue
+		}
+		if udpv4.Data[240] != 53 || udpv4.Data[241] != 1 || udpv4.Data[242] != 2 {
+			// not a DHCP offer
+			continue
+		}
+		var ip net.IP
+		ip = udpv4.Data[16:20]
+		finished = true // will terminate sending goroutine
+		return ip, nil
+	}
+}
+
+// DhcpRequest is a simple DHCP request
+type DhcpRequest struct {
+	MAC net.HardwareAddr
+}
+
+// NewDhcpRequest constructs a DHCP request
+func NewDhcpRequest(MAC net.HardwareAddr) *DhcpRequest {
+	if len(MAC) != 6 {
+		panic("MAC address must be 6 bytes")
+	}
+	return &DhcpRequest{MAC}
+}
+
+// Bytes returns the marshalled DHCP request
+func (d *DhcpRequest) Bytes() []byte {
+	bs := []byte{
+		0x01,                   // OP
+		0x01,                   // HTYPE
+		0x06,                   // HLEN
+		0x00,                   // HOPS
+		0x01, 0x00, 0x00, 0x00, // XID
+		0x00, 0x00, // SECS
+		0x80, 0x00, // FLAGS
+		0x00, 0x00, 0x00, 0x00, // CIADDR
+		0x00, 0x00, 0x00, 0x00, // YIADDR
+		0x00, 0x00, 0x00, 0x00, // SIADDR
+		0x00, 0x00, 0x00, 0x00, // GIADDR
+		d.MAC[0], d.MAC[1], d.MAC[2], d.MAC[3], d.MAC[4], d.MAC[5],
+	}
+	bs = append(bs, make([]byte, 202)...)
+	bs = append(bs, []byte{
+		0x63, 0x82, 0x53, 0x63, // Magic cookie
+		0x35, 0x01, 0x01, // DHCP discover
+		0xff, // Endmark
+	}...)
+	return bs
+}

go/pkg/vmnet/ethernet.go

@@ -0,0 +1,66 @@
+package vmnet
+
+import (
+	"bytes"
+	"encoding/binary"
+	"io"
+	"net"
+
+	"github.com/pkg/errors"
+)
+
+// EthernetFrame is an ethernet frame
+type EthernetFrame struct {
+	Dst  net.HardwareAddr
+	Src  net.HardwareAddr
+	Type uint16
+	Data []byte
+}
+
+// NewEthernetFrame constructs an Ethernet frame
+func NewEthernetFrame(Dst, Src net.HardwareAddr, Type uint16) *EthernetFrame {
+	Data := make([]byte, 0)
+	return &EthernetFrame{Dst, Src, Type, Data}
+}
+
+func (e *EthernetFrame) setData(data []byte) {
+	e.Data = data
+}
+
+// Write marshals an Ethernet frame
+func (e *EthernetFrame) Write(w io.Writer) error {
+	if err := binary.Write(w, binary.BigEndian, e.Dst); err != nil {
+		return err
+	}
+	if err := binary.Write(w, binary.BigEndian, e.Src); err != nil {
+		return err
+	}
+	if err := binary.Write(w, binary.BigEndian, e.Type); err != nil {
+		return err
+	}
+	if err := binary.Write(w, binary.BigEndian, e.Data); err != nil {
+		return err
+	}
+	return nil
+}
+
+// ParseEthernetFrame parses the ethernet frame
+func ParseEthernetFrame(frame []byte) (*EthernetFrame, error) {
+	if len(frame) < (6 + 6 + 2) {
+		return nil, errors.New("Ethernet frame is too small")
+	}
+	Dst := frame[0:6]
+	Src := frame[6:12]
+	Type := uint16(frame[12])<<8 + uint16(frame[13])
+	Data := frame[14:]
+	return &EthernetFrame{Dst, Src, Type, Data}, nil
+}
+
+// Bytes returns the marshalled ethernet frame
+func (e *EthernetFrame) Bytes() []byte {
+	buf := bytes.NewBufferString("")
+	if err := e.Write(buf); err != nil {
+		panic(err)
+	}
+	return buf.Bytes()
+}

go/pkg/vmnet/framing.go

@@ -0,0 +1,54 @@
+package vmnet
+
+import (
+	"encoding/binary"
+	"io"
+)
+
+// Messages sent to vpnkit can either be
+// - fixed-size, no length prefix
+// - variable-length, with a length prefix
+
+// fixedSizeSendReceiver sends and receives fixed-size control messages with no length prefix.
+type fixedSizeSendReceiver struct {
+	rw io.ReadWriter
+}
+
+var _ sendReceiver = fixedSizeSendReceiver{}
+
+func (f fixedSizeSendReceiver) Recv(buf []byte) (int, error) {
+	return io.ReadFull(f.rw, buf)
+}
+
+func (f fixedSizeSendReceiver) Send(buf []byte) (int, error) {
+	return f.rw.Write(buf)
+}
+
+// lengthPrefixer sends and receives variable-length control messages with a length prefix.
+type lengthPrefixer struct {
+	rw io.ReadWriter
+}
+
+var _ sendReceiver = lengthPrefixer{}
+
+func (e lengthPrefixer) Recv(buf []byte) (int, error) {
+	var len uint16
+	if err := binary.Read(e.rw, binary.LittleEndian, &len); err != nil {
+		return 0, err
+	}
+	if err := binary.Read(e.rw, binary.LittleEndian, &buf); err != nil {
+		return 0, err
+	}
+	return int(len), nil
+}
+
+func (e lengthPrefixer) Send(packet []byte) (int, error) {
+	len := uint16(len(packet))
+	if err := binary.Write(e.rw, binary.LittleEndian, len); err != nil {
+		return 0, err
+	}
+	if err := binary.Write(e.rw, binary.LittleEndian, packet); err != nil {
+		return 0, err
+	}
+	return int(len), nil
+}

go/pkg/vmnet/ipv4.go

@@ -0,0 +1,68 @@
+package vmnet
+
+import (
+	"net"
+
+	"github.com/pkg/errors"
+)
+
+// Ipv4 is an IPv4 frame
+type Ipv4 struct {
+	Dst      net.IP
+	Src      net.IP
+	Data     []byte
+	Checksum uint16
+}
+
+// NewIpv4 constructs a new empty IPv4 packet
+func NewIpv4(Dst, Src net.IP) *Ipv4 {
+	Checksum := uint16(0)
+	Data := make([]byte, 0)
+	return &Ipv4{Dst, Src, Data, Checksum}
+}
+
+// ParseIpv4 parses an IP packet
+func ParseIpv4(packet []byte) (*Ipv4, error) {
+	if len(packet) < 20 {
+		return nil, errors.New("IPv4 packet too small")
+	}
+	ihl := int((packet[0] & 0xf) * 4) // in octets
+	if len(packet) < ihl {
+		return nil, errors.New("IPv4 packet too small")
+	}
+	Dst := packet[12:16]
+	Src := packet[16:20]
+	Data := packet[ihl:]
+	Checksum := uint16(0) // assume offload
+	return &Ipv4{Dst, Src, Data, Checksum}, nil
+}
+
+func (i *Ipv4) setData(data []byte) {
+	i.Data = data
+	i.Checksum = uint16(0) // as if we were using offload
+}
+
+// HeaderBytes returns the marshalled form of the IPv4 header
+func (i *Ipv4) HeaderBytes() []byte {
+	len := len(i.Data) + 20
+	length := [2]byte{byte(len >> 8), byte(len & 0xff)}
+	checksum := [2]byte{byte(i.Checksum >> 8), byte(i.Checksum & 0xff)}
+	return []byte{
+		0x45,                 // version + IHL
+		0x00,                 // DSCP + ECN
+		length[0], length[1], // total length
+		0x7f, 0x61, // Identification
+		0x00, 0x00, // Flags + Fragment offset
+		0x40, // TTL
+		0x11, // Protocol
+		checksum[0], checksum[1],
+		0x00, 0x00, 0x00, 0x00, // source
+		0xff, 0xff, 0xff, 0xff, // destination
+	}
+}
+
+// Bytes returns the marshalled IPv4 packet
+func (i *Ipv4) Bytes() []byte {
+	header := i.HeaderBytes()
+	return append(header, i.Data...)
+}