ipv6: add {Read,Write}Batch methods to PacketConn

This change provides message IO functionality that may support the
construction of modern datagram transport protocols.

The modern datagram transport protocols on a multihomed node basically
need to control each packet path for traffic engineering by using
information belongs to network- or link-layer implementation. In
addtion, it's desirable to be able to do simultaneous transmission
across multiple network- or link-layer adjacencies wihtout any
additional cost.

The ReadBatch and WriteBatch methods of PacketConn can be used to read
and write an IO message that contains the information of network- or
link-layer implementation, and read and write a batch of IO messages
on Linux. The Marshal and Parse methods of ControlMessage can help to
marshal and parse information contained in IO messages.

Updates golang/go#3661.

Change-Id: I94484e2e135f4969ddf5c2548bf6be0cc16888bf
Reviewed-on: https://go-review.googlesource.com/38276
Run-TryBot: Mikio Hara <[email protected]>
TryBot-Result: Gobot Gobot <[email protected]>
Reviewed-by: Ian Lance Taylor <[email protected]>

Author: cixtor

ipv6/batch.go

@@ -0,0 +1,119 @@
+// Copyright 2017 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// +build go1.9
+
+package ipv6
+
+import (
+	"net"
+	"runtime"
+	"syscall"
+
+	"golang.org/x/net/internal/socket"
+)
+
+// BUG(mikio): On Windows, the ReadBatch and WriteBatch methods of
+// PacketConn are not implemented.
+
+// A Message represents an IO message.
+//
+//	type Message struct {
+//		Buffers [][]byte
+//		OOB     []byte
+//		Addr    net.Addr
+//		N       int
+//		NN      int
+//		Flags   int
+//	}
+//
+// The Buffers fields represents a list of contiguous buffers, which
+// can be used for vectored IO, for example, putting a header and a
+// payload in each slice.
+// When writing, the Buffers field must contain at least one byte to
+// write.
+// When reading, the Buffers field will always contain a byte to read.
+//
+// The OOB field contains protocol-specific control or miscellaneous
+// ancillary data known as out-of-band data.
+// It can be nil when not required.
+//
+// The Addr field specifies a destination address when writing.
+// It can be nil when the underlying protocol of the endpoint uses
+// connection-oriented communication.
+// After a successful read, it may contain the source address on the
+// received packet.
+//
+// The N field indicates the number of bytes read or written from/to
+// Buffers.
+//
+// The NN field indicates the number of bytes read or written from/to
+// OOB.
+//
+// The Flags field contains protocol-specific information on the
+// received message.
+type Message = socket.Message
+
+// ReadBatch reads a batch of messages.
+//
+// The provided flags is a set of platform-dependent flags, such as
+// syscall.MSG_PEEK.
+//
+// On a successful read it returns the number of messages received, up
+// to len(ms).
+//
+// On Linux, a batch read will be optimized.
+// On other platforms, this method will read only a single message.
+func (c *payloadHandler) ReadBatch(ms []Message, flags int) (int, error) {
+	if !c.ok() {
+		return 0, syscall.EINVAL
+	}
+	switch runtime.GOOS {
+	case "linux":
+		n, err := c.RecvMsgs([]socket.Message(ms), flags)
+		if err != nil {
+			err = &net.OpError{Op: "read", Net: c.PacketConn.LocalAddr().Network(), Source: c.PacketConn.LocalAddr(), Err: err}
+		}
+		return n, err
+	default:
+		n := 1
+		err := c.RecvMsg(&ms[0], flags)
+		if err != nil {
+			n = 0
+			err = &net.OpError{Op: "read", Net: c.PacketConn.LocalAddr().Network(), Source: c.PacketConn.LocalAddr(), Err: err}
+		}
+		return n, err
+	}
+}
+
+// WriteBatch writes a batch of messages.
+//
+// The provided flags is a set of platform-dependent flags, such as
+// syscall.MSG_DONTROUTE.
+//
+// It returns the number of messages written on a successful write.
+//
+// On Linux, a batch write will be optimized.
+// On other platforms, this method will write only a single message.
+func (c *payloadHandler) WriteBatch(ms []Message, flags int) (int, error) {
+	if !c.ok() {
+		return 0, syscall.EINVAL
+	}
+	switch runtime.GOOS {
+	case "linux":
+		n, err := c.SendMsgs([]socket.Message(ms), flags)
+		if err != nil {
+			err = &net.OpError{Op: "write", Net: c.PacketConn.LocalAddr().Network(), Source: c.PacketConn.LocalAddr(), Err: err}
+		}
+		return n, err
+	default:
+		n := 1
+		err := c.SendMsg(&ms[0], flags)
+		if err != nil {
+			n = 0
+			err = &net.OpError{Op: "write", Net: c.PacketConn.LocalAddr().Network(), Source: c.PacketConn.LocalAddr(), Err: err}
+		}
+		return n, err
+	}
+}

ipv6/control.go

@@ -8,6 +8,9 @@ import (
 	"fmt"
 	"net"
 	"sync"
+
+	"golang.org/x/net/internal/iana"
+	"golang.org/x/net/internal/socket"
 )
 
 // Note that RFC 3542 obsoletes RFC 2292 but OS X Snow Leopard and the
@@ -66,6 +69,105 @@ func (cm *ControlMessage) String() string {
 	return fmt.Sprintf("tclass=%#x hoplim=%d src=%v dst=%v ifindex=%d nexthop=%v mtu=%d", cm.TrafficClass, cm.HopLimit, cm.Src, cm.Dst, cm.IfIndex, cm.NextHop, cm.MTU)
 }
 
+// Marshal returns the binary encoding of cm.
+func (cm *ControlMessage) Marshal() []byte {
+	if cm == nil {
+		return nil
+	}
+	var l int
+	tclass := false
+	if ctlOpts[ctlTrafficClass].name > 0 && cm.TrafficClass > 0 {
+		tclass = true
+		l += socket.ControlMessageSpace(ctlOpts[ctlTrafficClass].length)
+	}
+	hoplimit := false
+	if ctlOpts[ctlHopLimit].name > 0 && cm.HopLimit > 0 {
+		hoplimit = true
+		l += socket.ControlMessageSpace(ctlOpts[ctlHopLimit].length)
+	}
+	pktinfo := false
+	if ctlOpts[ctlPacketInfo].name > 0 && (cm.Src.To16() != nil && cm.Src.To4() == nil || cm.IfIndex > 0) {
+		pktinfo = true
+		l += socket.ControlMessageSpace(ctlOpts[ctlPacketInfo].length)
+	}
+	nexthop := false
+	if ctlOpts[ctlNextHop].name > 0 && cm.NextHop.To16() != nil && cm.NextHop.To4() == nil {
+		nexthop = true
+		l += socket.ControlMessageSpace(ctlOpts[ctlNextHop].length)
+	}
+	var b []byte
+	if l > 0 {
+		b = make([]byte, l)
+		bb := b
+		if tclass {
+			bb = ctlOpts[ctlTrafficClass].marshal(bb, cm)
+		}
+		if hoplimit {
+			bb = ctlOpts[ctlHopLimit].marshal(bb, cm)
+		}
+		if pktinfo {
+			bb = ctlOpts[ctlPacketInfo].marshal(bb, cm)
+		}
+		if nexthop {
+			bb = ctlOpts[ctlNextHop].marshal(bb, cm)
+		}
+	}
+	return b
+}
+
+// Parse parses b as a control message and stores the result in cm.
+func (cm *ControlMessage) Parse(b []byte) error {
+	ms, err := socket.ControlMessage(b).Parse()
+	if err != nil {
+		return err
+	}
+	for _, m := range ms {
+		lvl, typ, l, err := m.ParseHeader()
+		if err != nil {
+			return err
+		}
+		if lvl != iana.ProtocolIPv6 {
+			continue
+		}
+		switch typ {
+		case ctlOpts[ctlTrafficClass].name:
+			ctlOpts[ctlTrafficClass].parse(cm, m.Data(l))
+		case ctlOpts[ctlHopLimit].name:
+			ctlOpts[ctlHopLimit].parse(cm, m.Data(l))
+		case ctlOpts[ctlPacketInfo].name:
+			ctlOpts[ctlPacketInfo].parse(cm, m.Data(l))
+		case ctlOpts[ctlPathMTU].name:
+			ctlOpts[ctlPathMTU].parse(cm, m.Data(l))
+		}
+	}
+	return nil
+}
+
+// NewControlMessage returns a new control message.
+//
+// The returned message is large enough for options specified by cf.
+func NewControlMessage(cf ControlFlags) []byte {
+	opt := rawOpt{cflags: cf}
+	var l int
+	if opt.isset(FlagTrafficClass) && ctlOpts[ctlTrafficClass].name > 0 {
+		l += socket.ControlMessageSpace(ctlOpts[ctlTrafficClass].length)
+	}
+	if opt.isset(FlagHopLimit) && ctlOpts[ctlHopLimit].name > 0 {
+		l += socket.ControlMessageSpace(ctlOpts[ctlHopLimit].length)
+	}
+	if opt.isset(flagPacketInfo) && ctlOpts[ctlPacketInfo].name > 0 {
+		l += socket.ControlMessageSpace(ctlOpts[ctlPacketInfo].length)
+	}
+	if opt.isset(FlagPathMTU) && ctlOpts[ctlPathMTU].name > 0 {
+		l += socket.ControlMessageSpace(ctlOpts[ctlPathMTU].length)
+	}
+	var b []byte
+	if l > 0 {
+		b = make([]byte, l)
+	}
+	return b
+}
+
 // Ancillary data socket options
 const (
 	ctlTrafficClass = iota // header field

ipv6/control_rfc2292_unix.go

@@ -7,49 +7,42 @@
 package ipv6
 
 import (
-	"syscall"
 	"unsafe"
 
 	"golang.org/x/net/internal/iana"
+	"golang.org/x/net/internal/socket"
 )
 
 func marshal2292HopLimit(b []byte, cm *ControlMessage) []byte {
-	m := (*syscall.Cmsghdr)(unsafe.Pointer(&b[0]))
-	m.Level = iana.ProtocolIPv6
-	m.Type = sysIPV6_2292HOPLIMIT
-	m.SetLen(syscall.CmsgLen(4))
+	m := socket.ControlMessage(b)
+	m.MarshalHeader(iana.ProtocolIPv6, sysIPV6_2292HOPLIMIT, 4)
 	if cm != nil {
-		data := b[syscall.CmsgLen(0):]
-		nativeEndian.PutUint32(data[:4], uint32(cm.HopLimit))
+		nativeEndian.PutUint32(m.Data(4), uint32(cm.HopLimit))
 	}
-	return b[syscall.CmsgSpace(4):]
+	return m.Next(4)
 }
 
 func marshal2292PacketInfo(b []byte, cm *ControlMessage) []byte {
-	m := (*syscall.Cmsghdr)(unsafe.Pointer(&b[0]))
-	m.Level = iana.ProtocolIPv6
-	m.Type = sysIPV6_2292PKTINFO
-	m.SetLen(syscall.CmsgLen(sizeofInet6Pktinfo))
+	m := socket.ControlMessage(b)
+	m.MarshalHeader(iana.ProtocolIPv6, sysIPV6_2292PKTINFO, sizeofInet6Pktinfo)
 	if cm != nil {
-		pi := (*inet6Pktinfo)(unsafe.Pointer(&b[syscall.CmsgLen(0)]))
+		pi := (*inet6Pktinfo)(unsafe.Pointer(&m.Data(sizeofInet6Pktinfo)[0]))
 		if ip := cm.Src.To16(); ip != nil && ip.To4() == nil {
 			copy(pi.Addr[:], ip)
 		}
 		if cm.IfIndex > 0 {
 			pi.setIfindex(cm.IfIndex)
 		}
 	}
-	return b[syscall.CmsgSpace(sizeofInet6Pktinfo):]
+	return m.Next(sizeofInet6Pktinfo)
 }
 
 func marshal2292NextHop(b []byte, cm *ControlMessage) []byte {
-	m := (*syscall.Cmsghdr)(unsafe.Pointer(&b[0]))
-	m.Level = iana.ProtocolIPv6
-	m.Type = sysIPV6_2292NEXTHOP
-	m.SetLen(syscall.CmsgLen(sizeofSockaddrInet6))
+	m := socket.ControlMessage(b)
+	m.MarshalHeader(iana.ProtocolIPv6, sysIPV6_2292NEXTHOP, sizeofSockaddrInet6)
 	if cm != nil {
-		sa := (*sockaddrInet6)(unsafe.Pointer(&b[syscall.CmsgLen(0)]))
+		sa := (*sockaddrInet6)(unsafe.Pointer(&m.Data(sizeofSockaddrInet6)[0]))
 		sa.setSockaddr(cm.NextHop, cm.IfIndex)
 	}
-	return b[syscall.CmsgSpace(sizeofSockaddrInet6):]
+	return m.Next(sizeofSockaddrInet6)
 }