github.com/mirantis/virtlet@v1.5.2-0.20191204181327-1659b8a48e9b/pkg/nettools/nettools.go

/*
Copyright 2016 Mirantis

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
*/

// Some of the code is based on CNI's plugins/main/bridge/bridge.go, pkg/ip/link.go
// Original copyright notice:
//
// Copyright 2014 CNI authors
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

package nettools

import (
	"crypto/rand"
	"errors"
	"fmt"
	"log"
	"net"
	"os"
	"os/exec"
	"sort"

	"github.com/containernetworking/cni/pkg/ns"
	cnitypes "github.com/containernetworking/cni/pkg/types"
	cnicurrent "github.com/containernetworking/cni/pkg/types/current"
	"github.com/davecgh/go-spew/spew"
	"github.com/golang/glog"
	"github.com/vishvananda/netlink"

	"github.com/Mirantis/virtlet/pkg/cni"
	"github.com/Mirantis/virtlet/pkg/network"
)

const (
	defaultMTU                  = 1500
	tapInterfaceNameTemplate    = "tap%d"
	containerBridgeNameTemplate = "br%d"
	loopbackInterfaceName       = "lo"
	// Address for dhcp server internal interface
	internalDhcpAddr = "169.254.254.2/24"
)

func makeVethPair(name, peer string, mtu int) (netlink.Link, error) {
	veth := &netlink.Veth{
		LinkAttrs: netlink.LinkAttrs{
			Name:  name,
			Flags: net.FlagUp,
			MTU:   mtu,
		},
		PeerName: peer,
	}
	if err := netlink.LinkAdd(veth); err != nil {
		return nil, err
	}

	return veth, nil
}

func peerExists(name string) bool {
	if _, err := netlink.LinkByName(name); err != nil {
		return false
	}
	return true
}

func makeVeth(name string, mtu int) (peerName string, veth netlink.Link, err error) {
	for i := 0; i < 10; i++ {
		peerName, err = RandomVethName()
		if err != nil {
			return
		}

		veth, err = makeVethPair(name, peerName, mtu)
		switch {
		case err == nil:
			return

		case os.IsExist(err):
			if peerExists(peerName) {
				continue
			}
			err = fmt.Errorf("container veth name provided (%v) already exists", name)
			return

		default:
			err = fmt.Errorf("failed to make veth pair: %v", err)
			return
		}
	}

	// should really never be hit
	err = fmt.Errorf("failed to find a unique veth name")
	return
}

// RandomVethName returns string "veth" with random prefix (hashed from entropy)
func RandomVethName() (string, error) {
	entropy := make([]byte, 4)
	_, err := rand.Reader.Read(entropy)
	if err != nil {
		return "", fmt.Errorf("failed to generate random veth name: %v", err)
	}

	// NetworkManager (recent versions) will ignore veth devices that start with "veth"
	return fmt.Sprintf("veth%x", entropy), nil
}

// SetupVeth sets up a pair of virtual ethernet devices.
// Call SetupVeth from inside the container netns.  It will create both veth
// devices and move the host-side veth into the provided hostNS namespace.
// On success, SetupVeth returns (hostVeth, containerVeth, nil)
func SetupVeth(contVethName string, mtu int, hostNS ns.NetNS) (netlink.Link, netlink.Link, error) {
	hostVethName, contVeth, err := makeVeth(contVethName, mtu)
	if err != nil {
		return nil, nil, err
	}

	if err = netlink.LinkSetUp(contVeth); err != nil {
		return nil, nil, fmt.Errorf("failed to set %q up: %v", contVethName, err)
	}

	hostVeth, err := netlink.LinkByName(hostVethName)
	if err != nil {
		return nil, nil, fmt.Errorf("failed to lookup %q: %v", hostVethName, err)
	}

	if err = netlink.LinkSetNsFd(hostVeth, int(hostNS.Fd())); err != nil {
		return nil, nil, fmt.Errorf("failed to move veth to host netns: %v", err)
	}

	err = hostNS.Do(func(_ ns.NetNS) error {
		hostVeth, err = netlink.LinkByName(hostVethName)
		if err != nil {
			return fmt.Errorf("failed to lookup %q in %q: %v", hostVethName, hostNS.Path(), err)
		}

		if err = netlink.LinkSetUp(hostVeth); err != nil {
			return fmt.Errorf("failed to set %q up: %v", hostVethName, err)
		}
		return nil
	})
	if err != nil {
		return nil, nil, err
	}
	return hostVeth, contVeth, nil
}

// CreateEscapeVethPair creates a veth pair with innerVeth residing in
// the specified network namespace innerNS and outerVeth residing in
// the 'outer' (current) namespace.
// TBD: move this to test tools
func CreateEscapeVethPair(innerNS ns.NetNS, ifName string, mtu int) (outerVeth, innerVeth netlink.Link, err error) {
	var outerVethName string

	err = innerNS.Do(func(outerNS ns.NetNS) error {
		// create the veth pair in the inner ns and move outer end into the outer netns
		outerVeth, innerVeth, err = SetupVeth(ifName, mtu, outerNS)
		if err != nil {
			return err
		}

		// need to lookup innerVeth again to get its attrs
		innerVeth, err = netlink.LinkByName(innerVeth.Attrs().Name)
		if err != nil {
			return err
		}

		outerVethName = outerVeth.Attrs().Name
		return nil
	})
	if err != nil {
		return
	}

	// need to lookup outerVeth again as its index has changed during ns move
	outerVeth, err = netlink.LinkByName(outerVethName)
	if err != nil {
		return nil, nil, fmt.Errorf("failed to lookup %q: %v", outerVethName, err)
	}

	return
}

func createBridge(brName string, mtu int) (*netlink.Bridge, error) {
	br := &netlink.Bridge{
		LinkAttrs: netlink.LinkAttrs{
			Name: brName,
			MTU:  mtu,
			// Let kernel use default txqueuelen; leaving it unset
			// means 0, and a zero-length TX queue messes up FIFO
			// traffic shapers which use TX queue length as the
			// default packet limit
			TxQLen: -1,
		},
	}

	if err := netlink.LinkAdd(br); err != nil {
		return nil, fmt.Errorf("could not add %q: %v", brName, err)
	}

	if err := netlink.LinkSetUp(br); err != nil {
		return nil, err
	}

	return br, nil
}

// SetupBridge creates a bridge and adds specified links to it.
// It sets bridge's MTU to MTU value of the first link.
func SetupBridge(bridgeName string, links []netlink.Link) (*netlink.Bridge, error) {
	if len(links) == 0 {
		return nil, errors.New("no links provided")
	}

	br, err := createBridge(bridgeName, links[0].Attrs().MTU)
	if err != nil {
		return nil, fmt.Errorf("failed to create bridge %q: %v", bridgeName, err)
	}

	for _, link := range links {
		if err = linkSetMaster(link, br); err != nil {
			delMessage := ""
			if delErr := netlink.LinkDel(br); delErr != nil {
				delMessage = fmt.Sprintf(" (and failed to delete the bridge: %v", err)
			}
			return nil, fmt.Errorf("failed to connect %q to bridge %v: %v%s", link.Attrs().Name, br.Attrs().Name, err, delMessage)
		}
	}

	return br, nil
}

// FindVeth locates single veth link in the list of provided links.
// There must be exactly one veth interface in the list.
func FindVeth(links []netlink.Link) (netlink.Link, error) {
	var veth netlink.Link
	for _, link := range links {
		if link.Type() != "veth" {
			continue
		}
		if veth != nil {
			return nil, errors.New("multiple veth links detected in container namespace")
		}
		veth = link
	}
	if veth == nil {
		return nil, errors.New("no veth interface found")
	}
	return veth, nil
}

func findLinkByAddress(links []netlink.Link, address net.IPNet) (netlink.Link, error) {
	for _, link := range links {
		addresses, err := netlink.AddrList(link, FAMILY_ALL)
		if err != nil {
			return nil, err
		}
		for _, addr := range addresses {
			if addr.IPNet.String() == address.String() {
				return link, nil
			}
		}
	}
	return nil, fmt.Errorf("interface with address %q not found in the container namespace", address.String())
}

// ValidateAndFixCNIResult verifies that netConfig contains proper list of
// ips, routes, interfaces and if something is missing it tries to complement
// that using patch for Weave or for plugins which return their netConfig
// in v0.2.0 version of CNI SPEC
func ValidateAndFixCNIResult(netConfig *cnicurrent.Result, nsPath string, allLinks []netlink.Link) (*cnicurrent.Result, error) {
	// If there are no routes provided, we consider it a broken
	// config and extract interface config instead. That's the
	// case with Weave CNI plugin. We don't do this for multiple CNI
	// at this point.
	if len(netConfig.IPs) == 1 && (cni.GetPodIP(netConfig) == "" || len(netConfig.Routes) == 0) {
		dnsInfo := netConfig.DNS

		veth, err := FindVeth(allLinks)
		if err != nil {
			return nil, err
		}
		if netConfig, err = ExtractLinkInfo(veth, nsPath); err != nil {
			return nil, err
		}

		// extracted netConfig doesn't have DNS information, so
		// still try to extract it from CNI-provided data
		netConfig.DNS = dnsInfo

		return netConfig, nil
	}

	if len(netConfig.IPs) == 0 {
		return nil, fmt.Errorf("cni result does not have any IP addresses")
	}

	// Interfaces contain broken info more often than not, so we
	// replace them here with what we can deduce from the network
	// links in the container netns
	for _, ipConfig := range netConfig.IPs {
		link, err := findLinkByAddress(allLinks, ipConfig.Address)
		if err != nil {
			return nil, err
		}

		found := false
		for i, iface := range netConfig.Interfaces {
			if iface.Name == link.Attrs().Name {
				ipConfig.Interface = i
				found = true
				break
			}
		}
		if !found {
			ipConfig.Interface = len(netConfig.Interfaces)
			netConfig.Interfaces = append(netConfig.Interfaces, &cnicurrent.Interface{
				Name:    link.Attrs().Name,
				Mac:     link.Attrs().HardwareAddr.String(),
				Sandbox: nsPath,
			})
		}
	}

	return netConfig, nil
}

// getContainerLinks finds links that correspond to interfaces in the current
// network namespace
func getContainerLinks(info *cnicurrent.Result) ([]netlink.Link, error) {
	// info.Interfaces is omitted by some CNI implementations and
	// the order may not be correct there after Virtlet adds the
	// missing ones, so we use interface indexes from info.IPs for
	// ordering.
	var links []netlink.Link
	order := make([]int, len(info.Interfaces))
	for n, ip := range info.IPs {
		if ip.Interface >= 0 && ip.Interface < len(order) {
			order[ip.Interface] = n + 1
		}
	}
	ifaces := make([]*cnicurrent.Interface, len(info.Interfaces))
	copy(ifaces, info.Interfaces)
	sort.SliceStable(ifaces, func(i, j int) bool { return order[i] < order[j] })
	for _, iface := range ifaces {
		// empty Sandbox means this interface belongs to the host
		// network namespace, so we skip it
		if iface.Sandbox == "" {
			continue
		}
		// If link is unavailable - simply add nil to slice
		link, err := netlink.LinkByName(iface.Name)
		if err != nil {
			if _, ok := err.(netlink.LinkNotFoundError); !ok {
				return nil, err
			}
		}
		links = append(links, link)
	}
	return links, nil
}

// StripLink removes addresses from the link
// along with any routes related to the link, except
// those created by the kernel
func StripLink(link netlink.Link) error {
	routes, err := netlink.RouteList(link, FAMILY_V4)
	if err != nil {
		return fmt.Errorf("failed to list routes: %v", err)
	}

	addrs, err := netlink.AddrList(link, FAMILY_V4)
	if err != nil {
		return fmt.Errorf("failed to get addresses for link: %v", err)
	}

	for _, route := range routes {
		if route.Protocol == RTPROT_KERNEL {
			// route created by the kernel
			continue
		}
		if err = netlink.RouteDel(&route); err != nil {
			return fmt.Errorf("error deleting route: %v", err)
		}
	}

	for _, addr := range addrs {
		if err = netlink.AddrDel(link, &addr); err != nil {
			return fmt.Errorf("error deleting address from the route: %v", err)
		}
	}

	return nil
}

// ExtractLinkInfo extracts ip address and netmask from veth
// interface in the current namespace, together with routes for this
// interface.
// There must be exactly one veth interface in the namespace
// and exactly one address associated with veth.
// Returns interface info struct and error, if any.
func ExtractLinkInfo(link netlink.Link, nsPath string) (*cnicurrent.Result, error) {
	addrs, err := netlink.AddrList(link, FAMILY_V4)
	if err != nil {
		return nil, fmt.Errorf("failed to get addresses for link: %v", err)
	}
	if len(addrs) != 1 {
		return nil, fmt.Errorf("expected exactly one address for link, but got %v", addrs)
	}

	result := &cnicurrent.Result{
		Interfaces: []*cnicurrent.Interface{
			{
				Name:    link.Attrs().Name,
				Mac:     link.Attrs().HardwareAddr.String(),
				Sandbox: nsPath,
			},
		},
		IPs: []*cnicurrent.IPConfig{
			{
				Version:   "4",
				Interface: 0,
				Address:   *addrs[0].IPNet,
			},
		},
	}

	routes, err := netlink.RouteList(link, FAMILY_V4)
	if err != nil {
		return nil, fmt.Errorf("failed to list routes: %v", err)
	}
	for _, route := range routes {
		switch {
		case route.Protocol == RTPROT_KERNEL:
			// route created by kernel
		case route.Gw == nil:
			// these routes can't be represented properly
			// by CNI result because CNI will consider
			// them having IP's default Gateway value as
			// Gw
		case (route.Dst == nil || route.Dst.IP == nil) && route.Gw == nil:
			// route has only Src
		case (route.Dst == nil || route.Dst.IP == nil):
			result.IPs[0].Gateway = route.Gw
			result.Routes = append(result.Routes, &cnitypes.Route{
				Dst: net.IPNet{
					IP:   net.IP{0, 0, 0, 0},
					Mask: net.IPMask{0, 0, 0, 0},
				},
				GW: route.Gw,
			})
		default:
			result.Routes = append(result.Routes, &cnitypes.Route{
				Dst: *route.Dst,
				GW:  route.Gw,
			})
		}
	}

	return result, nil
}

func mustParseAddr(addr string) *netlink.Addr {
	r, err := netlink.ParseAddr(addr)
	if err != nil {
		log.Panicf("Failed to parse address %q: %v", addr, err)
	}
	return r
}

func bringUpLoopback() error {
	// lo interface is already there in the new ns but it's down
	lo, err := netlink.LinkByName(loopbackInterfaceName)
	if err != nil {
		return fmt.Errorf("failed to find link %q: %v", loopbackInterfaceName, err)
	}
	if err := netlink.LinkSetUp(lo); err != nil {
		return fmt.Errorf("failed to bring up link %q: %v", loopbackInterfaceName, err)
	}
	return nil
}

func updateEbTables(nsPath, interfaceName, command string) error {
	// block/unblock DHCP traffic from/to CNI-provided link
	for _, item := range []struct{ chain, opt string }{
		// dhcp responses originate from bridge itself
		{"OUTPUT", "--ip-source-port"},
		// dhcp requests originate from the VM
		{"FORWARD", "--ip-destination-port"},
	} {
		if out, err := exec.Command(
			"nsenter", "--net="+nsPath,
			"ebtables", command, item.chain, "-p", "IPV4", "--ip-protocol", "UDP",
			item.opt, "67", "--out-if", interfaceName, "-j", "DROP").CombinedOutput(); err != nil {
			return fmt.Errorf("[netns %q] ebtables failed: %v\nOut:\n%s", nsPath, err, out)
		}
	}

	return nil
}

func disableMacLearning(nsPath string, bridgeName string) error {
	if out, err := exec.Command("nsenter", "--net="+nsPath, "brctl", "setageing", bridgeName, "0").CombinedOutput(); err != nil {
		return fmt.Errorf("[netns %q] brctl failed: %v\nOut:\n%s", nsPath, err, out)
	}

	return nil
}

// SetHardwareAddr sets hardware address on provided link.
func SetHardwareAddr(link netlink.Link, hwAddr net.HardwareAddr) error {
	if err := netlink.LinkSetDown(link); err != nil {
		return fmt.Errorf("can't bring down the link: %v", err)
	}
	if err := netlink.LinkSetHardwareAddr(link, hwAddr); err != nil {
		return fmt.Errorf("can't set hardware address for the link: %v", err)
	}
	if err := netlink.LinkSetUp(link); err != nil {
		return fmt.Errorf("can't bring up the link: %v", err)
	}

	return nil
}

func setupTapAndGetInterfaceDescription(link netlink.Link, nsPath string, ifaceNo int) (*network.InterfaceDescription, error) {
	hwAddr := link.Attrs().HardwareAddr
	ifaceName := link.Attrs().Name

	mtu := link.Attrs().MTU

	newHwAddr, err := GenerateMacAddress()
	if err == nil {
		err = SetHardwareAddr(link, newHwAddr)
	}
	if err != nil {
		return nil, err
	}

	tapInterfaceName := fmt.Sprintf(tapInterfaceNameTemplate, ifaceNo)
	tap, err := CreateTAP(tapInterfaceName, mtu)
	if err != nil {
		return nil, err
	}

	containerBridgeName := fmt.Sprintf(containerBridgeNameTemplate, ifaceNo)
	br, err := SetupBridge(containerBridgeName, []netlink.Link{link, tap})
	if err != nil {
		return nil, fmt.Errorf("failed to create bridge: %v", err)
	}

	if err := netlink.AddrAdd(br, mustParseAddr(internalDhcpAddr)); err != nil {
		return nil, fmt.Errorf("failed to set address for the bridge: %v", err)
	}

	// Add ebtables DHCP blocking rules
	if err := updateEbTables(nsPath, ifaceName, "-A"); err != nil {
		return nil, err
	}

	// Work around bridge MAC learning problem
	// https://ubuntuforums.org/showthread.php?t=2329373&s=cf580a41179e0f186ad4e625834a1d61&p=13511965#post13511965
	// (affects Flannel)
	if err := disableMacLearning(nsPath, containerBridgeName); err != nil {
		return nil, err
	}

	if err := bringUpLoopback(); err != nil {
		return nil, err
	}

	glog.V(3).Infof("Opening tap interface %q for link %q", tapInterfaceName, ifaceName)
	fo, err := OpenTAP(tapInterfaceName)
	if err != nil {
		return nil, fmt.Errorf("failed to open tap: %v", err)
	}
	glog.V(3).Infof("Adding interface %q as %q", ifaceName, tapInterfaceName)

	return &network.InterfaceDescription{
		Type:         network.InterfaceTypeTap,
		Name:         ifaceName,
		Fo:           fo,
		HardwareAddr: hwAddr,
		MTU:          uint16(mtu),
	}, nil
}

// SetupContainerSideNetwork sets up networking in container
// namespace.  It does so by preparing the following
// network interfaces in container ns:
//     tapX      - tap interface for the each interface to pass to VM
//     brX       - a bridge that joins above tapX and original CNI interface
// with X denoting an link index in info.Interfaces list.
// Each bridge gets assigned a link-local address to be used
// for dhcp server.
// In case of SR-IOV VFs this function only sets up a device to be passed to VM.
// The function should be called from within container namespace.
// Returns container network struct and an error, if any.
func SetupContainerSideNetwork(info *cnicurrent.Result, nsPath string, allLinks []netlink.Link, enableSriov bool, hostNS ns.NetNS) (*network.ContainerSideNetwork, error) {
	contLinks, err := getContainerLinks(info)
	if err != nil {
		return nil, err
	}

	var interfaces []*network.InterfaceDescription
	for i, link := range contLinks {
		if link == nil {
			return nil, fmt.Errorf("missing link #%d in the container network namespace (Virtlet pod restarted?)", i)
		}

		if err := StripLink(link); err != nil {
			return nil, err
		}

		var ifDesc *network.InterfaceDescription

		if isSriovVf(link) {
			if !enableSriov {
				return nil, fmt.Errorf("SR-IOV device configured in container network namespace while Virtlet is configured with disabled SR-IOV support")
			}
			if ifDesc, err = setupSriovAndGetInterfaceDescription(link, hostNS); err != nil {
				return nil, err
			}
		} else {
			if ifDesc, err = setupTapAndGetInterfaceDescription(link, nsPath, i); err != nil {
				return nil, err
			}
		}

		interfaces = append(interfaces, ifDesc)
	}

	return &network.ContainerSideNetwork{info, nsPath, interfaces}, nil
}

// RecoverContainerSideNetwork tries to populate ContainerSideNetwork
// structure based on a network namespace that was already adjusted for Virtlet
func RecoverContainerSideNetwork(csn *network.ContainerSideNetwork, nsPath string, allLinks []netlink.Link, hostNS ns.NetNS) error {
	if len(csn.Result.Interfaces) == 0 {
		return fmt.Errorf("wrong cni configuration: no interfaces defined: %s", spew.Sdump(csn.Result))
	}

	contLinks, err := getContainerLinks(csn.Result)
	if err != nil {
		return err
	}

	oldDescs := map[string]*network.InterfaceDescription{}
	for _, desc := range csn.Interfaces {
		oldDescs[desc.Name] = desc
	}

	for _, link := range contLinks {
		// Skip missing link which is already used by running VM
		if link == nil {
			continue
		}
		ifaceName := link.Attrs().Name
		desc, found := oldDescs[ifaceName]
		if !found {
			glog.Warningf("Recovering container side network: missing description for interface %q", ifaceName)
		}
		delete(oldDescs, ifaceName)
		var ifaceType network.InterfaceType

		if isSriovVf(link) {
			ifaceType = network.InterfaceTypeVF

			// device should be already unbound, but after machine reboot that can be necessary
			unbindDriverFromDevice(desc.PCIAddress)

			devIdentifier, err := getDeviceIdentifier(desc.PCIAddress)
			if err != nil {
				return err
			}

			// this can be problematic in case of machine reboot - we are trying to use the same
			// devices as was used before reboot, but in meantime there is small chance that they
			// were used already by sriov cni plugin (for which reboot means it's starting everything
			// from clean situation) for some other pods, before even virtlet was started
			// also in case of virtlet pod restart - device can be already bound to vfio-pci, so we
			// are ignoring any error there)
			bindDeviceToVFIO(devIdentifier)
		} else {
			ifaceType = network.InterfaceTypeTap
			// It's OK if OpenTAP failed as the device is busy and used by running VM
			if fo, err := OpenTAP(link.Attrs().Name); err == nil {
				desc.Fo = fo
			}
		}
		if desc.Type != ifaceType {
			return fmt.Errorf("bad interface type for %q", desc.Name)
		}
	}

	return nil
}

// TeardownBridge removes links from bridge and sets it down
func TeardownBridge(bridge netlink.Link, links []netlink.Link) error {
	for _, link := range links {
		if err := netlink.LinkSetNoMaster(link); err != nil {
			return err
		}
	}

	return netlink.LinkSetDown(bridge)
}

// ConfigureLink configures a link according to the CNI result
func ConfigureLink(link netlink.Link, info *cnicurrent.Result) error {
	ifaceNo := -1
	linkMAC := link.Attrs().HardwareAddr.String()
	for i, iface := range info.Interfaces {
		if iface.Mac == linkMAC {
			ifaceNo = i
			break
		}
	}
	if ifaceNo == -1 {
		return fmt.Errorf("can't find link with MAC %q in saved cni result: %s", linkMAC, spew.Sdump(info))
	}

	for _, addr := range info.IPs {
		if addr.Interface == ifaceNo {
			linkAddr := &netlink.Addr{IPNet: &addr.Address}
			if err := netlink.AddrAdd(link, linkAddr); err != nil {
				return fmt.Errorf("error adding address %v to link %q: %v", addr.Address, link.Attrs().Name, err)
			}

			for _, route := range info.Routes {
				// TODO: that's too naive - if there are more than one interfaces which have this gw address
				// in their subnet - same gw will be added on both of them
				// in theory this should be ok, but there is can lead to configuration other than prepared
				// by cni plugins
				if linkAddr.Contains(route.GW) {
					err := netlink.RouteAdd(&netlink.Route{
						LinkIndex: link.Attrs().Index,
						Scope:     SCOPE_UNIVERSE,
						Dst:       &route.Dst,
						Gw:        route.GW,
					})
					if err != nil {
						return fmt.Errorf("error adding route (dst %v gw %v): %v", route.Dst, route.GW, err)
					}
				}
			}
		}
	}

	return nil
}

// Teardown cleans up container network configuration.
// It does so by invoking teardown sequence which removes ebtables rules, links
// and addresses in an order opposite to that of their creation in SetupContainerSideNetwork.
// The end result is the same network configuration in the container network namespace
// as it was before SetupContainerSideNetwork() call.
func Teardown(csn *network.ContainerSideNetwork) error {
	for _, i := range csn.Interfaces {
		i.Fo.Close()
	}

	contLinks, err := getContainerLinks(csn.Result)
	if err != nil {
		return err
	}

	for i, contLink := range contLinks {
		if contLink == nil {
			return fmt.Errorf("missing %d link during teardown", i)
		}

		// Remove ebtables DHCP rules
		if err := updateEbTables(csn.NsPath, contLink.Attrs().Name, "-D"); err != nil {
			return nil
		}

		if !isSriovVf(contLink) {
			tapInterfaceName := fmt.Sprintf(tapInterfaceNameTemplate, i)
			tap, err := netlink.LinkByName(tapInterfaceName)
			if err != nil {
				return err
			}

			containerBridgeName := fmt.Sprintf(containerBridgeNameTemplate, i)
			br, err := netlink.LinkByName(containerBridgeName)
			if err != nil {
				return err
			}

			if err := netlink.AddrDel(br, mustParseAddr(internalDhcpAddr)); err != nil {
				return err
			}

			if err := TeardownBridge(br, []netlink.Link{contLink, tap}); err != nil {
				return err
			}

			if err := netlink.LinkDel(br); err != nil {
				return err
			}

			if err := netlink.LinkSetDown(tap); err != nil {
				return err
			}

			if err := netlink.LinkDel(tap); err != nil {
				return err
			}

			if err := SetHardwareAddr(contLink, csn.Interfaces[i].HardwareAddr); err != nil {
				return err
			}
		}

		rereadLink, err := netlink.LinkByName(contLink.Attrs().Name)
		if err != nil {
			return err
		}
		if err := ConfigureLink(rereadLink, csn.Result); err != nil {
			return err
		}
	}

	return nil
}

// GenerateMacAddress returns a random locally administrated unicast
// hardware address.
// Copied from:
// https://github.com/coreos/rkt/blob/56564bac090b44788684040f2ffd66463f29d5d0/stage1/init/kvm/network.go#L71
func GenerateMacAddress() (net.HardwareAddr, error) {
	mac := net.HardwareAddr{
		2,          // locally administred unicast
		0x65, 0x02, // OUI (randomly chosen by jell)
		0, 0, 0, // bytes to randomly overwrite
	}

	_, err := rand.Reader.Read(mac[3:6])
	if err != nil {
		return nil, fmt.Errorf("cannot generate random mac address: %v", err)
	}

	return mac, nil
}