net: implement raw sockets

R=rsc
CC=golang-dev
https://golang.org/cl/684041

8 files changed, 519 insertions, 10 deletions

diff --git a/src/pkg/net/Makefile b/src/pkg/net/Makefile
index 95360539b6..b018d05af3 100644
--- a/src/pkg/net/Makefile
+++ b/src/pkg/net/Makefile
@@ -14,6 +14,7 @@ GOFILES=\
 	hosts.go\
 	ip.go\
 	ipsock.go\
+	iprawsock.go\
 	net.go\
 	parse.go\
 	pipe.go\
diff --git a/src/pkg/net/ipraw_test.go b/src/pkg/net/ipraw_test.go
new file mode 100644
index 0000000000..6d9fb965ab
--- /dev/null
+++ b/src/pkg/net/ipraw_test.go
@@ -0,0 +1,117 @@
+// Copyright 2009 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.
+
+
+// TODO(cw): ListenPacket test, Read() test, ipv6 test &
+// Dial()/Listen() level tests
+
+package net
+
+import (
+	"bytes"
+	"flag"
+	"os"
+	"testing"
+)
+
+const ICMP_ECHO_REQUEST = 8
+const ICMP_ECHO_REPLY = 0
+
+// returns a suitable 'ping request' packet, with id & seq and a
+// payload length of pktlen
+func makePingRequest(id, seq, pktlen int, filler []byte) []byte {
+	p := make([]byte, pktlen)
+	copy(p[8:], bytes.Repeat(filler, (pktlen-8)/len(filler)+1))
+
+	p[0] = ICMP_ECHO_REQUEST // type
+	p[1] = 0                 // code
+	p[2] = 0                 // cksum
+	p[3] = 0                 // cksum
+	p[4] = uint8(id >> 8)    // id
+	p[5] = uint8(id & 0xff)  // id
+	p[6] = uint8(seq >> 8)   // sequence
+	p[7] = uint8(seq & 0xff) // sequence
+
+	// calculate icmp checksum
+	cklen := len(p)
+	s := uint32(0)
+	for i := 0; i < (cklen - 1); i += 2 {
+		s += uint32(p[i+1])<<8 | uint32(p[i])
+	}
+	if cklen&1 == 1 {
+		s += uint32(p[cklen-1])
+	}
+	s = (s >> 16) + (s & 0xffff)
+	s = s + (s >> 16)
+
+	// place checksum back in header; using ^= avoids the
+	// assumption the checksum bytes are zero
+	p[2] ^= uint8(^s & 0xff)
+	p[3] ^= uint8(^s >> 8)
+
+	return p
+}
+
+func parsePingReply(p []byte) (id, seq int) {
+	id = int(p[4])<<8 | int(p[5])
+	seq = int(p[6])<<8 | int(p[7])
+	return
+}
+
+var srchost = flag.String("srchost", "", "Source of the ICMP ECHO request")
+var dsthost = flag.String("dsthost", "localhost", "Destination for the ICMP ECHO request")
+
+// test (raw) IP socket using ICMP
+func TestICMP(t *testing.T) {
+	if os.Getuid() != 0 {
+		t.Logf("test disabled; must be root")
+		return
+	}
+
+	var laddr *IPAddr
+	if *srchost != "" {
+		laddr, err := ResolveIPAddr(*srchost)
+		if err != nil {
+			t.Fatalf(`net.ResolveIPAddr("%v") = %v, %v`, *srchost, laddr, err)
+		}
+	}
+
+	raddr, err := ResolveIPAddr(*dsthost)
+	if err != nil {
+		t.Fatalf(`net.ResolveIPAddr("%v") = %v, %v`, *dsthost, raddr, err)
+	}
+
+	c, err := ListenIP("ip4:icmp", laddr)
+	if err != nil {
+		t.Fatalf(`net.ListenIP("ip4:icmp", %v) = %v, %v`, *srchost, c, err)
+	}
+
+	sendid := os.Getpid()
+	const sendseq = 61455
+	const pingpktlen = 128
+	sendpkt := makePingRequest(sendid, sendseq, pingpktlen, []byte("Go Go Gadget Ping!!!"))
+
+	n, err := c.WriteToIP(sendpkt, raddr)
+	if err != nil || n != pingpktlen {
+		t.Fatalf(`net.WriteToIP(..., %v) = %v, %v`, raddr, n, err)
+	}
+
+	c.SetTimeout(100e6)
+	resp := make([]byte, 1024)
+	for {
+		n, from, err := c.ReadFrom(resp)
+		if err != nil {
+			t.Fatalf(`ReadFrom(...) = %v, %v, %v`, n, from, err)
+		}
+		if resp[0] != ICMP_ECHO_REPLY {
+			continue
+		}
+		rcvid, rcvseq := parsePingReply(resp)
+		if rcvid != sendid || rcvseq != sendseq {
+			t.Fatal(`Ping reply saw id,seq=%v,%v (expected %v, %v)`, rcvid, rcvseq, sendid, sendseq)
+		}
+		return
+	}
+	t.Fatalf("saw no ping return")
+}
diff --git a/src/pkg/net/iprawsock.go b/src/pkg/net/iprawsock.go
new file mode 100644
index 0000000000..e7eee1a4bf
--- /dev/null
+++ b/src/pkg/net/iprawsock.go
@@ -0,0 +1,351 @@
+// Copyright 2010 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.
+
+// (Raw) IP sockets
+
+package net
+
+import (
+	"once"
+	"os"
+	"syscall"
+)
+
+func sockaddrToIP(sa syscall.Sockaddr) Addr {
+	switch sa := sa.(type) {
+	case *syscall.SockaddrInet4:
+		return &IPAddr{&sa.Addr}
+	case *syscall.SockaddrInet6:
+		return &IPAddr{&sa.Addr}
+	}
+	return nil
+}
+
+// IPAddr represents the address of a IP end point.
+type IPAddr struct {
+	IP IP
+}
+
+// Network returns the address's network name, "ip".
+func (a *IPAddr) Network() string { return "ip" }
+
+func (a *IPAddr) String() string { return a.IP.String() }
+
+func (a *IPAddr) family() int {
+	if a == nil || len(a.IP) <= 4 {
+		return syscall.AF_INET
+	}
+	if ip := a.IP.To4(); ip != nil {
+		return syscall.AF_INET
+	}
+	return syscall.AF_INET6
+}
+
+func (a *IPAddr) sockaddr(family int) (syscall.Sockaddr, os.Error) {
+	return ipToSockaddr(family, a.IP, 0)
+}
+
+func (a *IPAddr) toAddr() sockaddr {
+	if a == nil { // nil *IPAddr
+		return nil // nil interface
+	}
+	return a
+}
+
+// ResolveIPAddr parses addr as a IP address and resolves domain
+// names to numeric addresses.  A literal IPv6 host address must be
+// enclosed in square brackets, as in "[::]".
+func ResolveIPAddr(addr string) (*IPAddr, os.Error) {
+	ip, err := hostToIP(addr)
+	if err != nil {
+		return nil, err
+	}
+	return &IPAddr{ip}, nil
+}
+
+// IPConn is the implementation of the Conn and PacketConn
+// interfaces for IP network connections.
+type IPConn struct {
+	fd *netFD
+}
+
+func newIPConn(fd *netFD) *IPConn { return &IPConn{fd} }
+
+func (c *IPConn) ok() bool { return c != nil && c.fd != nil }
+
+// Implementation of the Conn interface - see Conn for documentation.
+
+// Read implements the net.Conn Read method.
+func (c *IPConn) Read(b []byte) (n int, err os.Error) {
+	n, _, err = c.ReadFrom(b)
+	return
+}
+
+// Write implements the net.Conn Write method.
+func (c *IPConn) Write(b []byte) (n int, err os.Error) {
+	if !c.ok() {
+		return 0, os.EINVAL
+	}
+	return c.fd.Write(b)
+}
+
+// Close closes the IP connection.
+func (c *IPConn) Close() os.Error {
+	if !c.ok() {
+		return os.EINVAL
+	}
+	err := c.fd.Close()
+	c.fd = nil
+	return err
+}
+
+// LocalAddr returns the local network address.
+func (c *IPConn) LocalAddr() Addr {
+	if !c.ok() {
+		return nil
+	}
+	return c.fd.laddr
+}
+
+// RemoteAddr returns the remote network address, a *IPAddr.
+func (c *IPConn) RemoteAddr() Addr {
+	if !c.ok() {
+		return nil
+	}
+	return c.fd.raddr
+}
+
+// SetTimeout implements the net.Conn SetTimeout method.
+func (c *IPConn) SetTimeout(nsec int64) os.Error {
+	if !c.ok() {
+		return os.EINVAL
+	}
+	return setTimeout(c.fd, nsec)
+}
+
+// SetReadTimeout implements the net.Conn SetReadTimeout method.
+func (c *IPConn) SetReadTimeout(nsec int64) os.Error {
+	if !c.ok() {
+		return os.EINVAL
+	}
+	return setReadTimeout(c.fd, nsec)
+}
+
+// SetWriteTimeout implements the net.Conn SetWriteTimeout method.
+func (c *IPConn) SetWriteTimeout(nsec int64) os.Error {
+	if !c.ok() {
+		return os.EINVAL
+	}
+	return setWriteTimeout(c.fd, nsec)
+}
+
+// SetReadBuffer sets the size of the operating system's
+// receive buffer associated with the connection.
+func (c *IPConn) SetReadBuffer(bytes int) os.Error {
+	if !c.ok() {
+		return os.EINVAL
+	}
+	return setReadBuffer(c.fd, bytes)
+}
+
+// SetWriteBuffer sets the size of the operating system's
+// transmit buffer associated with the connection.
+func (c *IPConn) SetWriteBuffer(bytes int) os.Error {
+	if !c.ok() {
+		return os.EINVAL
+	}
+	return setWriteBuffer(c.fd, bytes)
+}
+
+// IP-specific methods.
+
+// ReadFromIP reads a IP packet from c, copying the payload into b.
+// It returns the number of bytes copied into b and the return address
+// that was on the packet.
+//
+// ReadFromIP can be made to time out and return an error with
+// Timeout() == true after a fixed time limit; see SetTimeout and
+// SetReadTimeout.
+func (c *IPConn) ReadFromIP(b []byte) (n int, addr *IPAddr, err os.Error) {
+	if !c.ok() {
+		return 0, nil, os.EINVAL
+	}
+	// TODO(cw,rsc): consider using readv if we know the family
+	// type to avoid the header trim/copy
+	n, sa, err := c.fd.ReadFrom(b)
+	switch sa := sa.(type) {
+	case *syscall.SockaddrInet4:
+		addr = &IPAddr{&sa.Addr}
+		if len(b) >= 4 { // discard ipv4 header
+			hsize := (int(b[0]) & 0xf) * 4
+			copy(b, b[hsize:])
+			n -= hsize
+		}
+	case *syscall.SockaddrInet6:
+		addr = &IPAddr{&sa.Addr}
+	}
+	return
+}
+
+// ReadFrom implements the net.PacketConn ReadFrom method.
+func (c *IPConn) ReadFrom(b []byte) (n int, addr Addr, err os.Error) {
+	if !c.ok() {
+		return 0, nil, os.EINVAL
+	}
+	n, uaddr, err := c.ReadFromIP(b)
+	return n, uaddr.toAddr(), err
+}
+
+// WriteToIP writes a IP packet to addr via c, copying the payload from b.
+//
+// WriteToIP can be made to time out and return
+// an error with Timeout() == true after a fixed time limit;
+// see SetTimeout and SetWriteTimeout.
+// On packet-oriented connections, write timeouts are rare.
+func (c *IPConn) WriteToIP(b []byte, addr *IPAddr) (n int, err os.Error) {
+	if !c.ok() {
+		return 0, os.EINVAL
+	}
+	sa, err1 := addr.sockaddr(c.fd.family)
+	if err1 != nil {
+		return 0, &OpError{Op: "write", Net: "ip", Addr: addr, Error: err1}
+	}
+	return c.fd.WriteTo(b, sa)
+}
+
+// WriteTo implements the net.PacketConn WriteTo method.
+func (c *IPConn) WriteTo(b []byte, addr Addr) (n int, err os.Error) {
+	if !c.ok() {
+		return 0, os.EINVAL
+	}
+	a, ok := addr.(*IPAddr)
+	if !ok {
+		return 0, &OpError{"writeto", "ip", addr, os.EINVAL}
+	}
+	return c.WriteToIP(b, a)
+}
+
+// Convert "host" into IP address.
+func hostToIP(host string) (ip IP, err os.Error) {
+	var addr IP
+	// Try as an IP address.
+	addr = ParseIP(host)
+	if addr == nil {
+		// Not an IP address.  Try as a DNS name.
+		_, addrs, err1 := LookupHost(host)
+		if err1 != nil {
+			err = err1
+			goto Error
+		}
+		addr = ParseIP(addrs[0])
+		if addr == nil {
+			// should not happen
+			err = &AddrError{"LookupHost returned invalid address", addrs[0]}
+			goto Error
+		}
+	}
+
+	return addr, nil
+
+Error:
+	return nil, err
+}
+
+
+var protocols map[string]int
+
+func readProtocols() {
+	protocols = make(map[string]int)
+	if file, err := open("/etc/protocols"); err == nil {
+		for line, ok := file.readLine(); ok; line, ok = file.readLine() {
+			// tcp    6   TCP    # transmission control protocol
+			if i := byteIndex(line, '#'); i >= 0 {
+				line = line[0:i]
+			}
+			f := getFields(line)
+			if len(f) < 2 {
+				continue
+			}
+			if proto, _, ok := dtoi(f[1], 0); ok {
+				protocols[f[0]] = proto
+				for _, alias := range f[2:] {
+					protocols[alias] = proto
+				}
+			}
+		}
+		file.close()
+	}
+}
+
+func netProtoSplit(netProto string) (net string, proto int, err os.Error) {
+	once.Do(readProtocols)
+	i := last(netProto, ':')
+	if i+1 >= len(netProto) { // no colon
+		return "", 0, os.ErrorString("no IP protocol specified")
+	}
+	net = netProto[0:i]
+	protostr := netProto[i+1:]
+	proto, i, ok := dtoi(protostr, 0)
+	if !ok || i != len(protostr) {
+		// lookup by name
+		proto, ok = protocols[protostr]
+		if ok {
+			return
+		}
+	}
+	return
+}
+
+// DialIP connects to the remote address raddr on the network net,
+// which must be "ip", "ip4", or "ip6".
+func DialIP(netProto string, laddr, raddr *IPAddr) (c *IPConn, err os.Error) {
+	net, proto, err := netProtoSplit(netProto)
+	if err != nil {
+		return
+	}
+	switch prefixBefore(net, ':') {
+	case "ip", "ip4", "ip6":
+	default:
+		return nil, UnknownNetworkError(net)
+	}
+	if raddr == nil {
+		return nil, &OpError{"dial", "ip", nil, errMissingAddress}
+	}
+	fd, e := internetSocket(net, laddr.toAddr(), raddr.toAddr(), syscall.SOCK_RAW, proto, "dial", sockaddrToIP)
+	if e != nil {
+		return nil, e
+	}
+	return newIPConn(fd), nil
+}
+
+// ListenIP listens for incoming IP packets addressed to the
+// local address laddr.  The returned connection c's ReadFrom
+// and WriteTo methods can be used to receive and send IP
+// packets with per-packet addressing.
+func ListenIP(netProto string, laddr *IPAddr) (c *IPConn, err os.Error) {
+	net, proto, err := netProtoSplit(netProto)
+	if err != nil {
+		return
+	}
+	switch prefixBefore(net, ':') {
+	case "ip", "ip4", "ip6":
+	default:
+		return nil, UnknownNetworkError(net)
+	}
+	fd, e := internetSocket(net, laddr.toAddr(), nil, syscall.SOCK_RAW, proto, "dial", sockaddrToIP)
+	if e != nil {
+		return nil, e
+	}
+	return newIPConn(fd), nil
+}
+
+// BindToDevice binds an IPConn to a network interface.
+func (c *IPConn) BindToDevice(device string) os.Error {
+	if !c.ok() {
+		return os.EINVAL
+	}
+	c.fd.incref()
+	defer c.fd.decref()
+	return os.NewSyscallError("setsockopt", syscall.BindToDevice(c.fd.sysfd, device))
+}
diff --git a/src/pkg/net/ipsock.go b/src/pkg/net/ipsock.go
index 3453aaab16..e4b442e73c 100644
--- a/src/pkg/net/ipsock.go
+++ b/src/pkg/net/ipsock.go
@@ -46,7 +46,7 @@ type sockaddr interface {
 	family() int
 }
 
-func internetSocket(net string, laddr, raddr sockaddr, proto int, mode string, toAddr func(syscall.Sockaddr) Addr) (fd *netFD, err os.Error) {
+func internetSocket(net string, laddr, raddr sockaddr, socktype, proto int, mode string, toAddr func(syscall.Sockaddr) Addr) (fd *netFD, err os.Error) {
 	// Figure out IP version.
 	// If network has a suffix like "tcp4", obey it.
 	var oserr os.Error
@@ -77,7 +77,7 @@ func internetSocket(net string, laddr, raddr sockaddr, proto int, mode string, t
 			goto Error
 		}
 	}
-	fd, oserr = socket(net, family, proto, 0, la, ra, toAddr)
+	fd, oserr = socket(net, family, socktype, proto, la, ra, toAddr)
 	if oserr != nil {
 		goto Error
 	}
diff --git a/src/pkg/net/net.go b/src/pkg/net/net.go
index 3f0b834c24..ba54412e94 100644
--- a/src/pkg/net/net.go
+++ b/src/pkg/net/net.go
@@ -8,7 +8,6 @@
 package net
 
 // TODO(rsc):
-//	support for raw IP sockets
 //	support for raw ethernet sockets
 
 import "os"
@@ -125,7 +124,8 @@ type Listener interface {
 // for the connection.
 //
 // Known networks are "tcp", "tcp4" (IPv4-only), "tcp6" (IPv6-only),
-// "udp", "udp4" (IPv4-only), and "udp6" (IPv6-only).
+// "udp", "udp4" (IPv4-only), "udp6" (IPv6-only), "ip", "ip4"
+// (IPv4-only) and "ip6" IPv6-only).
 //
 // For IP networks, addresses have the form host:port.  If host is
 // a literal IPv6 address, it must be enclosed in square brackets.
@@ -137,7 +137,7 @@ type Listener interface {
 //	Dial("tcp", "127.0.0.1:123", "127.0.0.1:88")
 //
 func Dial(net, laddr, raddr string) (c Conn, err os.Error) {
-	switch net {
+	switch prefixBefore(net, ':') {
 	case "tcp", "tcp4", "tcp6":
 		var la, ra *TCPAddr
 		if laddr != "" {
@@ -189,6 +189,24 @@ func Dial(net, laddr, raddr string) (c Conn, err os.Error) {
 			return nil, err
 		}
 		return c, nil
+	case "ip", "ip4", "ip6":
+		var la, ra *IPAddr
+		if laddr != "" {
+			if la, err = ResolveIPAddr(laddr); err != nil {
+				goto Error
+			}
+		}
+		if raddr != "" {
+			if ra, err = ResolveIPAddr(raddr); err != nil {
+				goto Error
+			}
+		}
+		c, err := DialIP(net, la, ra)
+		if err != nil {
+			return nil, err
+		}
+		return c, nil
+
 	}
 	err = UnknownNetworkError(net)
 Error:
@@ -232,7 +250,7 @@ func Listen(net, laddr string) (l Listener, err os.Error) {
 // The network string net must be a packet-oriented network:
 // "udp", "udp4", "udp6", or "unixgram".
 func ListenPacket(net, laddr string) (c PacketConn, err os.Error) {
-	switch net {
+	switch prefixBefore(net, ':') {
 	case "udp", "udp4", "udp6":
 		var la *UDPAddr
 		if laddr != "" {
@@ -257,6 +275,18 @@ func ListenPacket(net, laddr string) (c PacketConn, err os.Error) {
 			return nil, err
 		}
 		return c, nil
+	case "ip", "ip4", "ip6":
+		var la *IPAddr
+		if laddr != "" {
+			if la, err = ResolveIPAddr(laddr); err != nil {
+				return nil, err
+			}
+		}
+		c, err := ListenIP(net, la)
+		if err != nil {
+			return nil, err
+		}
+		return c, nil
 	}
 	return nil, UnknownNetworkError(net)
 }
diff --git a/src/pkg/net/parse.go b/src/pkg/net/parse.go
index 2bc0db4655..605f3110b7 100644
--- a/src/pkg/net/parse.go
+++ b/src/pkg/net/parse.go
@@ -192,6 +192,16 @@ func count(s string, b byte) int {
 	return n
 }
 
+// Returns the prefix of s up to but not including the character c
+func prefixBefore(s string, c byte) string {
+	for i, v := range s {
+		if v == int(c) {
+			return s[0:i]
+		}
+	}
+	return s
+}
+
 // Index of rightmost occurrence of b in s.
 func last(s string, b byte) int {
 	i := len(s)
diff --git a/src/pkg/net/tcpsock.go b/src/pkg/net/tcpsock.go
index 5b09f2d8c0..2221922325 100644
--- a/src/pkg/net/tcpsock.go
+++ b/src/pkg/net/tcpsock.go
@@ -198,7 +198,7 @@ func DialTCP(net string, laddr, raddr *TCPAddr) (c *TCPConn, err os.Error) {
 	if raddr == nil {
 		return nil, &OpError{"dial", "tcp", nil, errMissingAddress}
 	}
-	fd, e := internetSocket(net, laddr.toAddr(), raddr.toAddr(), syscall.SOCK_STREAM, "dial", sockaddrToTCP)
+	fd, e := internetSocket(net, laddr.toAddr(), raddr.toAddr(), syscall.SOCK_STREAM, 0, "dial", sockaddrToTCP)
 	if e != nil {
 		return nil, e
 	}
@@ -217,7 +217,7 @@ type TCPListener struct {
 // If laddr has a port of 0, it means to listen on some available port.
 // The caller can use l.Addr() to retrieve the chosen address.
 func ListenTCP(net string, laddr *TCPAddr) (l *TCPListener, err os.Error) {
-	fd, err := internetSocket(net, laddr.toAddr(), nil, syscall.SOCK_STREAM, "listen", sockaddrToTCP)
+	fd, err := internetSocket(net, laddr.toAddr(), nil, syscall.SOCK_STREAM, 0, "listen", sockaddrToTCP)
 	if err != nil {
 		return nil, err
 	}
diff --git a/src/pkg/net/udpsock.go b/src/pkg/net/udpsock.go
index f38f52f272..6de69a9c9b 100644
--- a/src/pkg/net/udpsock.go
+++ b/src/pkg/net/udpsock.go
@@ -233,7 +233,7 @@ func DialUDP(net string, laddr, raddr *UDPAddr) (c *UDPConn, err os.Error) {
 	if raddr == nil {
 		return nil, &OpError{"dial", "udp", nil, errMissingAddress}
 	}
-	fd, e := internetSocket(net, laddr.toAddr(), raddr.toAddr(), syscall.SOCK_DGRAM, "dial", sockaddrToUDP)
+	fd, e := internetSocket(net, laddr.toAddr(), raddr.toAddr(), syscall.SOCK_DGRAM, 0, "dial", sockaddrToUDP)
 	if e != nil {
 		return nil, e
 	}
@@ -253,7 +253,7 @@ func ListenUDP(net string, laddr *UDPAddr) (c *UDPConn, err os.Error) {
 	if laddr == nil {
 		return nil, &OpError{"listen", "udp", nil, errMissingAddress}
 	}
-	fd, e := internetSocket(net, laddr.toAddr(), nil, syscall.SOCK_DGRAM, "dial", sockaddrToUDP)
+	fd, e := internetSocket(net, laddr.toAddr(), nil, syscall.SOCK_DGRAM, 0, "dial", sockaddrToUDP)
 	if e != nil {
 		return nil, e
 	}