hakurei/container/netlink.go

package container

import (
	"encoding/binary"
	"errors"
	"net"
	"os"
	. "syscall"
	"unsafe"

	"hakurei.app/container/std"
	"hakurei.app/message"
)

// rtnetlink represents a NETLINK_ROUTE socket.
type rtnetlink struct {
	// Sent as part of rtnetlink messages.
	pid uint32
	// AF_NETLINK socket.
	fd int
	// Whether the socket is open.
	ok bool
	// Message sequence number.
	seq uint32
}

// open creates the underlying NETLINK_ROUTE socket.
func (s *rtnetlink) open() (err error) {
	if s.ok || s.fd < 0 {
		return os.ErrInvalid
	}

	s.pid = uint32(Getpid())
	if s.fd, err = Socket(
		AF_NETLINK,
		SOCK_RAW|SOCK_CLOEXEC,
		NETLINK_ROUTE,
	); err != nil {
		return os.NewSyscallError("socket", err)
	} else if err = Bind(s.fd, &SockaddrNetlink{
		Family: AF_NETLINK,
		Pid:    s.pid,
	}); err != nil {
		_ = s.close()
		return os.NewSyscallError("bind", err)
	} else {
		s.ok = true
		return nil
	}
}

// close closes the underlying NETLINK_ROUTE socket.
func (s *rtnetlink) close() error {
	if !s.ok {
		return os.ErrInvalid
	}

	s.ok = false
	err := Close(s.fd)
	s.fd = -1
	return err
}

// roundtrip sends a netlink message and handles the reply.
func (s *rtnetlink) roundtrip(data []byte) error {
	if !s.ok {
		return os.ErrInvalid
	}

	defer func() { s.seq++ }()

	if err := Sendto(s.fd, data, 0, &SockaddrNetlink{
		Family: AF_NETLINK,
	}); err != nil {
		return os.NewSyscallError("sendto", err)
	}
	buf := make([]byte, Getpagesize())

done:
	for {
		p := buf
		if n, _, err := Recvfrom(s.fd, p, 0); err != nil {
			return os.NewSyscallError("recvfrom", err)
		} else if n < NLMSG_HDRLEN {
			return errors.ErrUnsupported
		} else {
			p = p[:n]
		}

		if msgs, err := ParseNetlinkMessage(p); err != nil {
			return err
		} else {
			for _, m := range msgs {
				if m.Header.Seq != s.seq || m.Header.Pid != s.pid {
					return errors.ErrUnsupported
				}
				if m.Header.Type == NLMSG_DONE {
					break done
				}
				if m.Header.Type == NLMSG_ERROR {
					if len(m.Data) >= 4 {
						errno := Errno(-std.ScmpInt(binary.NativeEndian.Uint32(m.Data)))
						if errno == 0 {
							return nil
						}
						return errno
					}
					return errors.ErrUnsupported
				}
			}
		}
	}

	return nil
}

// mustRoundtrip calls roundtrip and terminates via msg for a non-nil error.
func (s *rtnetlink) mustRoundtrip(msg message.Msg, data []byte) {
	err := s.roundtrip(data)
	if err == nil {
		return
	}
	if closeErr := Close(s.fd); closeErr != nil {
		msg.Verbosef("cannot close: %v", err)
	}

	switch err.(type) {
	case *os.SyscallError:
		msg.GetLogger().Fatalf("cannot %v", err)

	case Errno:
		msg.GetLogger().Fatalf("RTNETLINK answers: %v", err)

	default:
		msg.GetLogger().Fatalln("RTNETLINK answers with unexpected message")
	}
}

// newaddrLo represents a RTM_NEWADDR message with two addresses.
type newaddrLo struct {
	header NlMsghdr
	data   IfAddrmsg

	r0 RtAttr
	a0 [4]byte // in_addr
	r1 RtAttr
	a1 [4]byte // in_addr
}

// sizeofNewaddrLo is the expected size of newaddrLo.
const sizeofNewaddrLo = NLMSG_HDRLEN + SizeofIfAddrmsg + (SizeofRtAttr+4)*2

// newaddrLo returns the address of a populated newaddrLo.
func (s *rtnetlink) newaddrLo(lo int) *newaddrLo {
	return &newaddrLo{NlMsghdr{
		Len:   sizeofNewaddrLo,
		Type:  RTM_NEWADDR,
		Flags: NLM_F_REQUEST | NLM_F_ACK | NLM_F_CREATE | NLM_F_EXCL,
		Seq:   s.seq,
		Pid:   s.pid,
	}, IfAddrmsg{
		Family:    AF_INET,
		Prefixlen: 8,
		Flags:     IFA_F_PERMANENT,
		Scope:     RT_SCOPE_HOST,
		Index:     uint32(lo),
	}, RtAttr{
		Len:  uint16(SizeofRtAttr + len(newaddrLo{}.a0)),
		Type: IFA_LOCAL,
	}, [4]byte{127, 0, 0, 1}, RtAttr{
		Len:  uint16(SizeofRtAttr + len(newaddrLo{}.a1)),
		Type: IFA_ADDRESS,
	}, [4]byte{127, 0, 0, 1}}
}

func (msg *newaddrLo) toWireFormat() []byte {
	var buf [sizeofNewaddrLo]byte

	*(*uint32)(unsafe.Pointer(&buf[0:4][0])) = msg.header.Len
	*(*uint16)(unsafe.Pointer(&buf[4:6][0])) = msg.header.Type
	*(*uint16)(unsafe.Pointer(&buf[6:8][0])) = msg.header.Flags
	*(*uint32)(unsafe.Pointer(&buf[8:12][0])) = msg.header.Seq
	*(*uint32)(unsafe.Pointer(&buf[12:16][0])) = msg.header.Pid

	buf[16] = msg.data.Family
	buf[17] = msg.data.Prefixlen
	buf[18] = msg.data.Flags
	buf[19] = msg.data.Scope
	*(*uint32)(unsafe.Pointer(&buf[20:24][0])) = msg.data.Index

	*(*uint16)(unsafe.Pointer(&buf[24:26][0])) = msg.r0.Len
	*(*uint16)(unsafe.Pointer(&buf[26:28][0])) = msg.r0.Type
	copy(buf[28:32], msg.a0[:])
	*(*uint16)(unsafe.Pointer(&buf[32:34][0])) = msg.r1.Len
	*(*uint16)(unsafe.Pointer(&buf[34:36][0])) = msg.r1.Type
	copy(buf[36:40], msg.a1[:])

	return buf[:]
}

// newlinkLo represents a RTM_NEWLINK message.
type newlinkLo struct {
	header NlMsghdr
	data   IfInfomsg
}

// sizeofNewlinkLo is the expected size of newlinkLo.
const sizeofNewlinkLo = NLMSG_HDRLEN + SizeofIfInfomsg

// newlinkLo returns the address of a populated newlinkLo.
func (s *rtnetlink) newlinkLo(lo int) *newlinkLo {
	return &newlinkLo{NlMsghdr{
		Len:   sizeofNewlinkLo,
		Type:  RTM_NEWLINK,
		Flags: NLM_F_REQUEST | NLM_F_ACK,
		Seq:   s.seq,
		Pid:   s.pid,
	}, IfInfomsg{
		Family: AF_UNSPEC,
		Index:  int32(lo),
		Flags:  IFF_UP,
		Change: IFF_UP,
	}}
}

func (msg *newlinkLo) toWireFormat() []byte {
	var buf [sizeofNewlinkLo]byte

	*(*uint32)(unsafe.Pointer(&buf[0:4][0])) = msg.header.Len
	*(*uint16)(unsafe.Pointer(&buf[4:6][0])) = msg.header.Type
	*(*uint16)(unsafe.Pointer(&buf[6:8][0])) = msg.header.Flags
	*(*uint32)(unsafe.Pointer(&buf[8:12][0])) = msg.header.Seq
	*(*uint32)(unsafe.Pointer(&buf[12:16][0])) = msg.header.Pid

	buf[16] = msg.data.Family
	*(*uint16)(unsafe.Pointer(&buf[18:20][0])) = msg.data.Type
	*(*int32)(unsafe.Pointer(&buf[20:24][0])) = msg.data.Index
	*(*uint32)(unsafe.Pointer(&buf[24:28][0])) = msg.data.Flags
	*(*uint32)(unsafe.Pointer(&buf[28:32][0])) = msg.data.Change

	return buf[:]
}

// mustLoopback creates the loopback address and brings the lo interface up.
// mustLoopback calls a fatal method of the underlying [log.Logger] of m with a
// user-facing error message if RTNETLINK behaves unexpectedly.
func mustLoopback(msg message.Msg) {
	log := msg.GetLogger()

	var lo int
	if ifi, err := net.InterfaceByName("lo"); err != nil {
		log.Fatalln(err)
	} else {
		lo = ifi.Index
	}

	var s rtnetlink
	if err := s.open(); err != nil {
		log.Fatalln(err)
	}
	defer func() {
		if err := s.close(); err != nil {
			msg.Verbosef("cannot close netlink: %v", err)
		}
	}()

	s.mustRoundtrip(msg, s.newaddrLo(lo).toWireFormat())
	s.mustRoundtrip(msg, s.newlinkLo(lo).toWireFormat())
}