cr: add network support

We need to enumirate all veth devices on restore and handle
{un,}lock-network notifications.

Signed-off-by: Andrey Vagin <[email protected]>

Author: avagin

Diff for: container_linux.go

@@ -417,6 +417,18 @@ func (c *linuxContainer) Restore(process *Process, imagePath string) error {
 			req.Opts.ExtMnt = append(req.Opts.ExtMnt, extMnt)
 		}
 	}
+	for _, iface := range c.config.Networks {
+		switch iface.Type {
+		case "veth":
+			veth := new(criurpc.CriuVethPair)
+			veth.IfOut = proto.String(iface.HostInterfaceName)
+			veth.IfIn = proto.String(iface.Name)
+			req.Opts.Veths = append(req.Opts.Veths, veth)
+			break
+		case "loopback":
+			break
+		}
+	}
 	// Pipes that were previously set up for std{in,out,err}
 	// were removed after checkpoint.  Use the new ones.
 	var i int32
@@ -555,6 +567,34 @@ func (c *linuxContainer) criuSwrk(process *Process, req *criurpc.CriuReq, imageP
 	return nil
 }
 
+// block any external network activity
+func lockNetwork(config *configs.Config) error {
+	for _, config := range config.Networks {
+		strategy, err := getStrategy(config.Type)
+		if err != nil {
+			return err
+		}
+
+		if err := strategy.detach(config); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+func unlockNetwork(config *configs.Config) error {
+	for _, config := range config.Networks {
+		strategy, err := getStrategy(config.Type)
+		if err != nil {
+			return err
+		}
+		if err = strategy.attach(config); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
 func (c *linuxContainer) criuNotifications(resp *criurpc.CriuResp, process *Process, imagePath string) error {
 	notify := resp.GetNotify()
 	if notify == nil {
@@ -570,6 +610,18 @@ func (c *linuxContainer) criuNotifications(resp *criurpc.CriuResp, process *Proc
 		f.Close()
 		break
 
+	case notify.GetScript() == "network-unlock":
+		if err := unlockNetwork(c.config); err != nil {
+			return err
+		}
+		break
+
+	case notify.GetScript() == "network-lock":
+		if err := lockNetwork(c.config); err != nil {
+			return err
+		}
+		break
+
 	case notify.GetScript() == "post-restore":
 		pid := notify.GetPid()
 		r, err := newRestoredProcess(int(pid))

Diff for: netlink/netlink_linux.go

@@ -27,6 +27,7 @@ const (
 	SIOC_BRADDBR      = 0x89a0
 	SIOC_BRDELBR      = 0x89a1
 	SIOC_BRADDIF      = 0x89a2
+	SIOC_BRDELIF      = 0x89a3
 )
 
 const (
@@ -1191,9 +1192,7 @@ func DeleteBridge(name string) error {
 	return nil
 }
 
-// Add a slave to abridge device.  This is more backward-compatible than
-// netlink.NetworkSetMaster and works on RHEL 6.
-func AddToBridge(iface, master *net.Interface) error {
+func ifIoctBridge(iface, master *net.Interface, op uintptr) error {
 	if len(master.Name) >= IFNAMSIZ {
 		return fmt.Errorf("Interface name %s too long", master.Name)
 	}
@@ -1208,13 +1207,25 @@ func AddToBridge(iface, master *net.Interface) error {
 	copy(ifr.IfrnName[:len(ifr.IfrnName)-1], master.Name)
 	ifr.IfruIndex = int32(iface.Index)
 
-	if _, _, err := syscall.Syscall(syscall.SYS_IOCTL, uintptr(s), SIOC_BRADDIF, uintptr(unsafe.Pointer(&ifr))); err != 0 {
+	if _, _, err := syscall.Syscall(syscall.SYS_IOCTL, uintptr(s), op, uintptr(unsafe.Pointer(&ifr))); err != 0 {
 		return err
 	}
 
 	return nil
 }
 
+// Add a slave to a bridge device.  This is more backward-compatible than
+// netlink.NetworkSetMaster and works on RHEL 6.
+func AddToBridge(iface, master *net.Interface) error {
+	return ifIoctBridge(iface, master, SIOC_BRADDIF)
+}
+
+// Detach a slave from a bridge device.  This is more backward-compatible than
+// netlink.NetworkSetMaster and works on RHEL 6.
+func DelFromBridge(iface, master *net.Interface) error {
+	return ifIoctBridge(iface, master, SIOC_BRDELIF)
+}
+
 func randMacAddr() string {
 	hw := make(net.HardwareAddr, 6)
 	for i := 0; i < 6; i++ {

Diff for: network_linux.go

@@ -10,6 +10,7 @@ import (
 	"strconv"
 	"strings"
 
+	"github.com/docker/libcontainer/configs"
 	"github.com/docker/libcontainer/netlink"
 	"github.com/docker/libcontainer/utils"
 )
@@ -24,6 +25,8 @@ var strategies = map[string]networkStrategy{
 type networkStrategy interface {
 	create(*network, int) error
 	initialize(*network) error
+	detach(*configs.Network) error
+	attach(*configs.Network) error
 }
 
 // getStrategy returns the specific network strategy for the
@@ -97,32 +100,39 @@ func (l *loopback) initialize(config *network) error {
 	return netlink.NetworkLinkUp(iface)
 }
 
+func (l *loopback) attach(n *configs.Network) (err error) {
+	return nil
+}
+
+func (l *loopback) detach(n *configs.Network) (err error) {
+	return nil
+}
+
 // veth is a network strategy that uses a bridge and creates
 // a veth pair, one that is attached to the bridge on the host and the other
 // is placed inside the container's namespace
 type veth struct {
 }
 
-func (v *veth) create(n *network, nspid int) (err error) {
-	tmpName, err := v.generateTempPeerName()
+func (v *veth) detach(n *configs.Network) (err error) {
+	bridge, err := net.InterfaceByName(n.Bridge)
 	if err != nil {
 		return err
 	}
-	n.TempVethPeerName = tmpName
-	defer func() {
-		if err != nil {
-			netlink.NetworkLinkDel(n.HostInterfaceName)
-			netlink.NetworkLinkDel(n.TempVethPeerName)
-		}
-	}()
-	if n.Bridge == "" {
-		return fmt.Errorf("bridge is not specified")
-	}
-	bridge, err := net.InterfaceByName(n.Bridge)
+	host, err := net.InterfaceByName(n.HostInterfaceName)
 	if err != nil {
 		return err
 	}
-	if err := netlink.NetworkCreateVethPair(n.HostInterfaceName, n.TempVethPeerName, n.TxQueueLen); err != nil {
+	if err := netlink.DelFromBridge(host, bridge); err != nil {
+		return err
+	}
+	return nil
+}
+
+// attach a container network interface to an external network
+func (v *veth) attach(n *configs.Network) (err error) {
+	bridge, err := net.InterfaceByName(n.Bridge)
+	if err != nil {
 		return err
 	}
 	host, err := net.InterfaceByName(n.HostInterfaceName)
@@ -143,6 +153,31 @@ func (v *veth) create(n *network, nspid int) (err error) {
 	if err := netlink.NetworkLinkUp(host); err != nil {
 		return err
 	}
+
+	return nil
+}
+
+func (v *veth) create(n *network, nspid int) (err error) {
+	tmpName, err := v.generateTempPeerName()
+	if err != nil {
+		return err
+	}
+	n.TempVethPeerName = tmpName
+	defer func() {
+		if err != nil {
+			netlink.NetworkLinkDel(n.HostInterfaceName)
+			netlink.NetworkLinkDel(n.TempVethPeerName)
+		}
+	}()
+	if n.Bridge == "" {
+		return fmt.Errorf("bridge is not specified")
+	}
+	if err := netlink.NetworkCreateVethPair(n.HostInterfaceName, n.TempVethPeerName, n.TxQueueLen); err != nil {
+		return err
+	}
+	if err := v.attach(&n.Network); err != nil {
+		return err
+	}
 	child, err := net.InterfaceByName(n.TempVethPeerName)
 	if err != nil {
 		return err