Issue #33: Move portmapper and portallocator into libnetwork

- As they provide network translation functionalities,
  they should be part of libnetwork
- In driver/bridge/setup_ip_tables.go remove depenency
  on docker/daemon/networkdriver

Signed-off-by: Alessandro Boch <aboch@docker.com>

libnetwork/portmapper/mapper.go

@@ -0,0 +1,182 @@
+package portmapper
+
+import (
+	"errors"
+	"fmt"
+	"net"
+	"sync"
+
+	log "github.com/Sirupsen/logrus"
+	"github.com/docker/docker/pkg/iptables"
+	"github.com/docker/libnetwork/portallocator"
+)
+
+type mapping struct {
+	proto         string
+	userlandProxy userlandProxy
+	host          net.Addr
+	container     net.Addr
+}
+
+var (
+	chain *iptables.Chain
+	lock  sync.Mutex
+
+	// udp:ip:port
+	currentMappings = make(map[string]*mapping)
+
+	newProxy = newProxyCommand
+)
+
+var (
+	// ErrUnknownBackendAddressType refers to an unknown container or unsupported address type
+	ErrUnknownBackendAddressType = errors.New("unknown container address type not supported")
+	// ErrPortMappedForIP refers to a port already mapped to an ip address
+	ErrPortMappedForIP = errors.New("port is already mapped to ip")
+	// ErrPortNotMapped refers to an unampped port
+	ErrPortNotMapped = errors.New("port is not mapped")
+)
+
+// SetIptablesChain sets the specified chain into portmapper
+func SetIptablesChain(c *iptables.Chain) {
+	chain = c
+}
+
+// Map maps the specified container transport address to the host's network address and transport port
+func Map(container net.Addr, hostIP net.IP, hostPort int) (host net.Addr, err error) {
+	lock.Lock()
+	defer lock.Unlock()
+
+	var (
+		m                 *mapping
+		proto             string
+		allocatedHostPort int
+		proxy             userlandProxy
+	)
+
+	switch container.(type) {
+	case *net.TCPAddr:
+		proto = "tcp"
+		if allocatedHostPort, err = portallocator.RequestPort(hostIP, proto, hostPort); err != nil {
+			return nil, err
+		}
+
+		m = &mapping{
+			proto:     proto,
+			host:      &net.TCPAddr{IP: hostIP, Port: allocatedHostPort},
+			container: container,
+		}
+
+		proxy = newProxy(proto, hostIP, allocatedHostPort, container.(*net.TCPAddr).IP, container.(*net.TCPAddr).Port)
+	case *net.UDPAddr:
+		proto = "udp"
+		if allocatedHostPort, err = portallocator.RequestPort(hostIP, proto, hostPort); err != nil {
+			return nil, err
+		}
+
+		m = &mapping{
+			proto:     proto,
+			host:      &net.UDPAddr{IP: hostIP, Port: allocatedHostPort},
+			container: container,
+		}
+
+		proxy = newProxy(proto, hostIP, allocatedHostPort, container.(*net.UDPAddr).IP, container.(*net.UDPAddr).Port)
+	default:
+		return nil, ErrUnknownBackendAddressType
+	}
+
+	// release the allocated port on any further error during return.
+	defer func() {
+		if err != nil {
+			portallocator.ReleasePort(hostIP, proto, allocatedHostPort)
+		}
+	}()
+
+	key := getKey(m.host)
+	if _, exists := currentMappings[key]; exists {
+		return nil, ErrPortMappedForIP
+	}
+
+	containerIP, containerPort := getIPAndPort(m.container)
+	if err := forward(iptables.Append, m.proto, hostIP, allocatedHostPort, containerIP.String(), containerPort); err != nil {
+		return nil, err
+	}
+
+	cleanup := func() error {
+		// need to undo the iptables rules before we return
+		proxy.Stop()
+		forward(iptables.Delete, m.proto, hostIP, allocatedHostPort, containerIP.String(), containerPort)
+		if err := portallocator.ReleasePort(hostIP, m.proto, allocatedHostPort); err != nil {
+			return err
+		}
+
+		return nil
+	}
+
+	if err := proxy.Start(); err != nil {
+		if err := cleanup(); err != nil {
+			return nil, fmt.Errorf("Error during port allocation cleanup: %v", err)
+		}
+		return nil, err
+	}
+	m.userlandProxy = proxy
+	currentMappings[key] = m
+	return m.host, nil
+}
+
+// Unmap removes stored mapping for the specified host transport address
+func Unmap(host net.Addr) error {
+	lock.Lock()
+	defer lock.Unlock()
+
+	key := getKey(host)
+	data, exists := currentMappings[key]
+	if !exists {
+		return ErrPortNotMapped
+	}
+
+	data.userlandProxy.Stop()
+
+	delete(currentMappings, key)
+
+	containerIP, containerPort := getIPAndPort(data.container)
+	hostIP, hostPort := getIPAndPort(data.host)
+	if err := forward(iptables.Delete, data.proto, hostIP, hostPort, containerIP.String(), containerPort); err != nil {
+		log.Errorf("Error on iptables delete: %s", err)
+	}
+
+	switch a := host.(type) {
+	case *net.TCPAddr:
+		return portallocator.ReleasePort(a.IP, "tcp", a.Port)
+	case *net.UDPAddr:
+		return portallocator.ReleasePort(a.IP, "udp", a.Port)
+	}
+	return nil
+}
+
+func getKey(a net.Addr) string {
+	switch t := a.(type) {
+	case *net.TCPAddr:
+		return fmt.Sprintf("%s:%d/%s", t.IP.String(), t.Port, "tcp")
+	case *net.UDPAddr:
+		return fmt.Sprintf("%s:%d/%s", t.IP.String(), t.Port, "udp")
+	}
+	return ""
+}
+
+func getIPAndPort(a net.Addr) (net.IP, int) {
+	switch t := a.(type) {
+	case *net.TCPAddr:
+		return t.IP, t.Port
+	case *net.UDPAddr:
+		return t.IP, t.Port
+	}
+	return nil, 0
+}
+
+func forward(action iptables.Action, proto string, sourceIP net.IP, sourcePort int, containerIP string, containerPort int) error {
+	if chain == nil {
+		return nil
+	}
+	return chain.Forward(action, sourceIP, sourcePort, proto, containerIP, containerPort)
+}

libnetwork/portmapper/mapper_test.go

@@ -0,0 +1,152 @@
+package portmapper
+
+import (
+	"net"
+	"testing"
+
+	"github.com/docker/docker/pkg/iptables"
+	"github.com/docker/libnetwork/portallocator"
+)
+
+func init() {
+	// override this func to mock out the proxy server
+	newProxy = newMockProxyCommand
+}
+
+func reset() {
+	chain = nil
+	currentMappings = make(map[string]*mapping)
+}
+
+func TestSetIptablesChain(t *testing.T) {
+	defer reset()
+
+	c := &iptables.Chain{
+		Name:   "TEST",
+		Bridge: "192.168.1.1",
+	}
+
+	if chain != nil {
+		t.Fatal("chain should be nil at init")
+	}
+
+	SetIptablesChain(c)
+	if chain == nil {
+		t.Fatal("chain should not be nil after set")
+	}
+}
+
+func TestMapPorts(t *testing.T) {
+	dstIP1 := net.ParseIP("192.168.0.1")
+	dstIP2 := net.ParseIP("192.168.0.2")
+	dstAddr1 := &net.TCPAddr{IP: dstIP1, Port: 80}
+	dstAddr2 := &net.TCPAddr{IP: dstIP2, Port: 80}
+
+	srcAddr1 := &net.TCPAddr{Port: 1080, IP: net.ParseIP("172.16.0.1")}
+	srcAddr2 := &net.TCPAddr{Port: 1080, IP: net.ParseIP("172.16.0.2")}
+
+	addrEqual := func(addr1, addr2 net.Addr) bool {
+		return (addr1.Network() == addr2.Network()) && (addr1.String() == addr2.String())
+	}
+
+	if host, err := Map(srcAddr1, dstIP1, 80); err != nil {
+		t.Fatalf("Failed to allocate port: %s", err)
+	} else if !addrEqual(dstAddr1, host) {
+		t.Fatalf("Incorrect mapping result: expected %s:%s, got %s:%s",
+			dstAddr1.String(), dstAddr1.Network(), host.String(), host.Network())
+	}
+
+	if _, err := Map(srcAddr1, dstIP1, 80); err == nil {
+		t.Fatalf("Port is in use - mapping should have failed")
+	}
+
+	if _, err := Map(srcAddr2, dstIP1, 80); err == nil {
+		t.Fatalf("Port is in use - mapping should have failed")
+	}
+
+	if _, err := Map(srcAddr2, dstIP2, 80); err != nil {
+		t.Fatalf("Failed to allocate port: %s", err)
+	}
+
+	if Unmap(dstAddr1) != nil {
+		t.Fatalf("Failed to release port")
+	}
+
+	if Unmap(dstAddr2) != nil {
+		t.Fatalf("Failed to release port")
+	}
+
+	if Unmap(dstAddr2) == nil {
+		t.Fatalf("Port already released, but no error reported")
+	}
+}
+
+func TestGetUDPKey(t *testing.T) {
+	addr := &net.UDPAddr{IP: net.ParseIP("192.168.1.5"), Port: 53}
+
+	key := getKey(addr)
+
+	if expected := "192.168.1.5:53/udp"; key != expected {
+		t.Fatalf("expected key %s got %s", expected, key)
+	}
+}
+
+func TestGetTCPKey(t *testing.T) {
+	addr := &net.TCPAddr{IP: net.ParseIP("192.168.1.5"), Port: 80}
+
+	key := getKey(addr)
+
+	if expected := "192.168.1.5:80/tcp"; key != expected {
+		t.Fatalf("expected key %s got %s", expected, key)
+	}
+}
+
+func TestGetUDPIPAndPort(t *testing.T) {
+	addr := &net.UDPAddr{IP: net.ParseIP("192.168.1.5"), Port: 53}
+
+	ip, port := getIPAndPort(addr)
+	if expected := "192.168.1.5"; ip.String() != expected {
+		t.Fatalf("expected ip %s got %s", expected, ip)
+	}
+
+	if ep := 53; port != ep {
+		t.Fatalf("expected port %d got %d", ep, port)
+	}
+}
+
+func TestMapAllPortsSingleInterface(t *testing.T) {
+	dstIP1 := net.ParseIP("0.0.0.0")
+	srcAddr1 := &net.TCPAddr{Port: 1080, IP: net.ParseIP("172.16.0.1")}
+
+	hosts := []net.Addr{}
+	var host net.Addr
+	var err error
+
+	defer func() {
+		for _, val := range hosts {
+			Unmap(val)
+		}
+	}()
+
+	for i := 0; i < 10; i++ {
+		for i := portallocator.BeginPortRange; i < portallocator.EndPortRange; i++ {
+			if host, err = Map(srcAddr1, dstIP1, 0); err != nil {
+				t.Fatal(err)
+			}
+
+			hosts = append(hosts, host)
+		}
+
+		if _, err := Map(srcAddr1, dstIP1, portallocator.BeginPortRange); err == nil {
+			t.Fatalf("Port %d should be bound but is not", portallocator.BeginPortRange)
+		}
+
+		for _, val := range hosts {
+			if err := Unmap(val); err != nil {
+				t.Fatal(err)
+			}
+		}
+
+		hosts = []net.Addr{}
+	}
+}

libnetwork/portmapper/mock_proxy.go

@@ -0,0 +1,18 @@
+package portmapper
+
+import "net"
+
+func newMockProxyCommand(proto string, hostIP net.IP, hostPort int, containerIP net.IP, containerPort int) userlandProxy {
+	return &mockProxyCommand{}
+}
+
+type mockProxyCommand struct {
+}
+
+func (p *mockProxyCommand) Start() error {
+	return nil
+}
+
+func (p *mockProxyCommand) Stop() error {
+	return nil
+}

libnetwork/portmapper/proxy.go

@@ -0,0 +1,161 @@
+package portmapper
+
+import (
+	"flag"
+	"fmt"
+	"io/ioutil"
+	"log"
+	"net"
+	"os"
+	"os/exec"
+	"os/signal"
+	"strconv"
+	"syscall"
+	"time"
+
+	"github.com/docker/docker/pkg/proxy"
+	"github.com/docker/docker/pkg/reexec"
+)
+
+const userlandProxyCommandName = "docker-proxy"
+
+func init() {
+	reexec.Register(userlandProxyCommandName, execProxy)
+}
+
+type userlandProxy interface {
+	Start() error
+	Stop() error
+}
+
+// proxyCommand wraps an exec.Cmd to run the userland TCP and UDP
+// proxies as separate processes.
+type proxyCommand struct {
+	cmd *exec.Cmd
+}
+
+// execProxy is the reexec function that is registered to start the userland proxies
+func execProxy() {
+	f := os.NewFile(3, "signal-parent")
+	host, container := parseHostContainerAddrs()
+
+	p, err := proxy.NewProxy(host, container)
+	if err != nil {
+		fmt.Fprintf(f, "1\n%s", err)
+		f.Close()
+		os.Exit(1)
+	}
+	go handleStopSignals(p)
+	fmt.Fprint(f, "0\n")
+	f.Close()
+
+	// Run will block until the proxy stops
+	p.Run()
+}
+
+// parseHostContainerAddrs parses the flags passed on reexec to create the TCP or UDP
+// net.Addrs to map the host and container ports
+func parseHostContainerAddrs() (host net.Addr, container net.Addr) {
+	var (
+		proto         = flag.String("proto", "tcp", "proxy protocol")
+		hostIP        = flag.String("host-ip", "", "host ip")
+		hostPort      = flag.Int("host-port", -1, "host port")
+		containerIP   = flag.String("container-ip", "", "container ip")
+		containerPort = flag.Int("container-port", -1, "container port")
+	)
+
+	flag.Parse()
+
+	switch *proto {
+	case "tcp":
+		host = &net.TCPAddr{IP: net.ParseIP(*hostIP), Port: *hostPort}
+		container = &net.TCPAddr{IP: net.ParseIP(*containerIP), Port: *containerPort}
+	case "udp":
+		host = &net.UDPAddr{IP: net.ParseIP(*hostIP), Port: *hostPort}
+		container = &net.UDPAddr{IP: net.ParseIP(*containerIP), Port: *containerPort}
+	default:
+		log.Fatalf("unsupported protocol %s", *proto)
+	}
+
+	return host, container
+}
+
+func handleStopSignals(p proxy.Proxy) {
+	s := make(chan os.Signal, 10)
+	signal.Notify(s, os.Interrupt, syscall.SIGTERM, syscall.SIGSTOP)
+
+	for _ = range s {
+		p.Close()
+
+		os.Exit(0)
+	}
+}
+
+func newProxyCommand(proto string, hostIP net.IP, hostPort int, containerIP net.IP, containerPort int) userlandProxy {
+	args := []string{
+		userlandProxyCommandName,
+		"-proto", proto,
+		"-host-ip", hostIP.String(),
+		"-host-port", strconv.Itoa(hostPort),
+		"-container-ip", containerIP.String(),
+		"-container-port", strconv.Itoa(containerPort),
+	}
+
+	return &proxyCommand{
+		cmd: &exec.Cmd{
+			Path: reexec.Self(),
+			Args: args,
+			SysProcAttr: &syscall.SysProcAttr{
+				Pdeathsig: syscall.SIGTERM, // send a sigterm to the proxy if the daemon process dies
+			},
+		},
+	}
+}
+
+func (p *proxyCommand) Start() error {
+	r, w, err := os.Pipe()
+	if err != nil {
+		return fmt.Errorf("proxy unable to open os.Pipe %s", err)
+	}
+	defer r.Close()
+	p.cmd.ExtraFiles = []*os.File{w}
+	if err := p.cmd.Start(); err != nil {
+		return err
+	}
+	w.Close()
+
+	errchan := make(chan error, 1)
+	go func() {
+		buf := make([]byte, 2)
+		r.Read(buf)
+
+		if string(buf) != "0\n" {
+			errStr, err := ioutil.ReadAll(r)
+			if err != nil {
+				errchan <- fmt.Errorf("Error reading exit status from userland proxy: %v", err)
+				return
+			}
+
+			errchan <- fmt.Errorf("Error starting userland proxy: %s", errStr)
+			return
+		}
+		errchan <- nil
+	}()
+
+	select {
+	case err := <-errchan:
+		return err
+	case <-time.After(16 * time.Second):
+		return fmt.Errorf("Timed out proxy starting the userland proxy")
+	}
+}
+
+func (p *proxyCommand) Stop() error {
+	if p.cmd.Process != nil {
+		if err := p.cmd.Process.Signal(os.Interrupt); err != nil {
+			return err
+		}
+		return p.cmd.Wait()
+	}
+	return nil
+}