netlink/link_linux.go

package netlink

import (
	"bytes"
	"encoding/binary"
	"fmt"
	"net"
	"syscall"

	"github.com/vishvananda/netlink/nl"
)

var native = nl.NativeEndian()
var lookupByDump = false

func ensureIndex(link *LinkAttrs) {
	if link != nil && link.Index == 0 {
		newlink, _ := LinkByName(link.Name)
		if newlink != nil {
			link.Index = newlink.Attrs().Index
		}
	}
}

// LinkSetUp enables the link device.
// Equivalent to: `ip link set $link up`
func LinkSetUp(link Link) error {
	base := link.Attrs()
	ensureIndex(base)
	req := nl.NewNetlinkRequest(syscall.RTM_NEWLINK, syscall.NLM_F_ACK)

	msg := nl.NewIfInfomsg(syscall.AF_UNSPEC)
	msg.Change = syscall.IFF_UP
	msg.Flags = syscall.IFF_UP
	msg.Index = int32(base.Index)
	req.AddData(msg)

	_, err := req.Execute(syscall.NETLINK_ROUTE, 0)
	return err
}

// LinkSetUp disables link device.
// Equivalent to: `ip link set $link down`
func LinkSetDown(link Link) error {
	base := link.Attrs()
	ensureIndex(base)
	req := nl.NewNetlinkRequest(syscall.RTM_NEWLINK, syscall.NLM_F_ACK)

	msg := nl.NewIfInfomsg(syscall.AF_UNSPEC)
	msg.Change = syscall.IFF_UP
	msg.Flags = 0 & ^syscall.IFF_UP
	msg.Index = int32(base.Index)
	req.AddData(msg)

	_, err := req.Execute(syscall.NETLINK_ROUTE, 0)
	return err
}

// LinkSetMTU sets the mtu of the link device.
// Equivalent to: `ip link set $link mtu $mtu`
func LinkSetMTU(link Link, mtu int) error {
	base := link.Attrs()
	ensureIndex(base)
	req := nl.NewNetlinkRequest(syscall.RTM_SETLINK, syscall.NLM_F_ACK)

	msg := nl.NewIfInfomsg(syscall.AF_UNSPEC)
	msg.Type = syscall.RTM_SETLINK
	msg.Flags = syscall.NLM_F_REQUEST
	msg.Index = int32(base.Index)
	msg.Change = nl.DEFAULT_CHANGE
	req.AddData(msg)

	b := make([]byte, 4)
	native.PutUint32(b, uint32(mtu))

	data := nl.NewRtAttr(syscall.IFLA_MTU, b)
	req.AddData(data)

	_, err := req.Execute(syscall.NETLINK_ROUTE, 0)
	return err
}

// LinkSetName sets the name of the link device.
// Equivalent to: `ip link set $link name $name`
func LinkSetName(link Link, name string) error {
	base := link.Attrs()
	ensureIndex(base)
	req := nl.NewNetlinkRequest(syscall.RTM_SETLINK, syscall.NLM_F_ACK)

	msg := nl.NewIfInfomsg(syscall.AF_UNSPEC)
	msg.Type = syscall.RTM_SETLINK
	msg.Flags = syscall.NLM_F_REQUEST
	msg.Index = int32(base.Index)
	msg.Change = nl.DEFAULT_CHANGE
	req.AddData(msg)

	data := nl.NewRtAttr(syscall.IFLA_IFNAME, []byte(name))
	req.AddData(data)

	_, err := req.Execute(syscall.NETLINK_ROUTE, 0)
	return err
}

// LinkSetHardwareAddr sets the hardware address of the link device.
// Equivalent to: `ip link set $link address $hwaddr`
func LinkSetHardwareAddr(link Link, hwaddr net.HardwareAddr) error {
	base := link.Attrs()
	ensureIndex(base)
	req := nl.NewNetlinkRequest(syscall.RTM_SETLINK, syscall.NLM_F_ACK)

	msg := nl.NewIfInfomsg(syscall.AF_UNSPEC)
	msg.Type = syscall.RTM_SETLINK
	msg.Flags = syscall.NLM_F_REQUEST
	msg.Index = int32(base.Index)
	msg.Change = nl.DEFAULT_CHANGE
	req.AddData(msg)

	data := nl.NewRtAttr(syscall.IFLA_ADDRESS, []byte(hwaddr))
	req.AddData(data)

	_, err := req.Execute(syscall.NETLINK_ROUTE, 0)
	return err
}

// LinkSetMaster sets the master of the link device.
// Equivalent to: `ip link set $link master $master`
func LinkSetMaster(link Link, master *Bridge) error {
	index := 0
	if master != nil {
		masterBase := master.Attrs()
		ensureIndex(masterBase)
		index = masterBase.Index
	}
	return LinkSetMasterByIndex(link, index)
}

// LinkSetMasterByIndex sets the master of the link device.
// Equivalent to: `ip link set $link master $master`
func LinkSetMasterByIndex(link Link, masterIndex int) error {
	base := link.Attrs()
	ensureIndex(base)
	req := nl.NewNetlinkRequest(syscall.RTM_SETLINK, syscall.NLM_F_ACK)

	msg := nl.NewIfInfomsg(syscall.AF_UNSPEC)
	msg.Type = syscall.RTM_SETLINK
	msg.Flags = syscall.NLM_F_REQUEST
	msg.Index = int32(base.Index)
	msg.Change = nl.DEFAULT_CHANGE
	req.AddData(msg)

	b := make([]byte, 4)
	native.PutUint32(b, uint32(masterIndex))

	data := nl.NewRtAttr(syscall.IFLA_MASTER, b)
	req.AddData(data)

	_, err := req.Execute(syscall.NETLINK_ROUTE, 0)
	return err
}

// LinkSetNsPid puts the device into a new network namespace. The
// pid must be a pid of a running process.
// Equivalent to: `ip link set $link netns $pid`
func LinkSetNsPid(link Link, nspid int) error {
	base := link.Attrs()
	ensureIndex(base)
	req := nl.NewNetlinkRequest(syscall.RTM_SETLINK, syscall.NLM_F_ACK)

	msg := nl.NewIfInfomsg(syscall.AF_UNSPEC)
	msg.Type = syscall.RTM_SETLINK
	msg.Flags = syscall.NLM_F_REQUEST
	msg.Index = int32(base.Index)
	msg.Change = nl.DEFAULT_CHANGE
	req.AddData(msg)

	b := make([]byte, 4)
	native.PutUint32(b, uint32(nspid))

	data := nl.NewRtAttr(syscall.IFLA_NET_NS_PID, b)
	req.AddData(data)

	_, err := req.Execute(syscall.NETLINK_ROUTE, 0)
	return err
}

// LinkSetNsPid puts the device into a new network namespace. The
// fd must be an open file descriptor to a network namespace.
// Similar to: `ip link set $link netns $ns`
func LinkSetNsFd(link Link, fd int) error {
	base := link.Attrs()
	ensureIndex(base)
	req := nl.NewNetlinkRequest(syscall.RTM_SETLINK, syscall.NLM_F_ACK)

	msg := nl.NewIfInfomsg(syscall.AF_UNSPEC)
	msg.Type = syscall.RTM_SETLINK
	msg.Flags = syscall.NLM_F_REQUEST
	msg.Index = int32(base.Index)
	msg.Change = nl.DEFAULT_CHANGE
	req.AddData(msg)

	b := make([]byte, 4)
	native.PutUint32(b, uint32(fd))

	data := nl.NewRtAttr(nl.IFLA_NET_NS_FD, b)
	req.AddData(data)

	_, err := req.Execute(syscall.NETLINK_ROUTE, 0)
	return err
}

func boolAttr(val bool) []byte {
	var v uint8
	if val {
		v = 1
	}
	return nl.Uint8Attr(v)
}

type vxlanPortRange struct {
	Lo, Hi uint16
}

func addVxlanAttrs(vxlan *Vxlan, linkInfo *nl.RtAttr) {
	data := nl.NewRtAttrChild(linkInfo, nl.IFLA_INFO_DATA, nil)
	nl.NewRtAttrChild(data, nl.IFLA_VXLAN_ID, nl.Uint32Attr(uint32(vxlan.VxlanId)))
	if vxlan.VtepDevIndex != 0 {
		nl.NewRtAttrChild(data, nl.IFLA_VXLAN_LINK, nl.Uint32Attr(uint32(vxlan.VtepDevIndex)))
	}
	if vxlan.SrcAddr != nil {
		ip := vxlan.SrcAddr.To4()
		if ip != nil {
			nl.NewRtAttrChild(data, nl.IFLA_VXLAN_LOCAL, []byte(ip))
		} else {
			ip = vxlan.SrcAddr.To16()
			if ip != nil {
				nl.NewRtAttrChild(data, nl.IFLA_VXLAN_LOCAL6, []byte(ip))
			}
		}
	}
	if vxlan.Group != nil {
		group := vxlan.Group.To4()
		if group != nil {
			nl.NewRtAttrChild(data, nl.IFLA_VXLAN_GROUP, []byte(group))
		} else {
			group = vxlan.Group.To16()
			if group != nil {
				nl.NewRtAttrChild(data, nl.IFLA_VXLAN_GROUP6, []byte(group))
			}
		}
	}

	nl.NewRtAttrChild(data, nl.IFLA_VXLAN_TTL, nl.Uint8Attr(uint8(vxlan.TTL)))
	nl.NewRtAttrChild(data, nl.IFLA_VXLAN_TOS, nl.Uint8Attr(uint8(vxlan.TOS)))
	nl.NewRtAttrChild(data, nl.IFLA_VXLAN_LEARNING, boolAttr(vxlan.Learning))
	nl.NewRtAttrChild(data, nl.IFLA_VXLAN_PROXY, boolAttr(vxlan.Proxy))
	nl.NewRtAttrChild(data, nl.IFLA_VXLAN_RSC, boolAttr(vxlan.RSC))
	nl.NewRtAttrChild(data, nl.IFLA_VXLAN_L2MISS, boolAttr(vxlan.L2miss))
	nl.NewRtAttrChild(data, nl.IFLA_VXLAN_L3MISS, boolAttr(vxlan.L3miss))

	if vxlan.NoAge {
		nl.NewRtAttrChild(data, nl.IFLA_VXLAN_AGEING, nl.Uint32Attr(0))
	} else if vxlan.Age > 0 {
		nl.NewRtAttrChild(data, nl.IFLA_VXLAN_AGEING, nl.Uint32Attr(uint32(vxlan.Age)))
	}
	if vxlan.Limit > 0 {
		nl.NewRtAttrChild(data, nl.IFLA_VXLAN_LIMIT, nl.Uint32Attr(uint32(vxlan.Limit)))
	}
	if vxlan.Port > 0 {
		nl.NewRtAttrChild(data, nl.IFLA_VXLAN_PORT, nl.Uint16Attr(uint16(vxlan.Port)))
	}
	if vxlan.PortLow > 0 || vxlan.PortHigh > 0 {
		pr := vxlanPortRange{uint16(vxlan.PortLow), uint16(vxlan.PortHigh)}

		buf := new(bytes.Buffer)
		binary.Write(buf, binary.BigEndian, &pr)

		nl.NewRtAttrChild(data, nl.IFLA_VXLAN_PORT_RANGE, buf.Bytes())
	}
}

// LinkAdd adds a new link device. The type and features of the device
// are taken fromt the parameters in the link object.
// Equivalent to: `ip link add $link`
func LinkAdd(link Link) error {
	// TODO: set mtu and hardware address
	// TODO: support extra data for macvlan
	base := link.Attrs()

	if base.Name == "" {
		return fmt.Errorf("LinkAttrs.Name cannot be empty!")
	}

	req := nl.NewNetlinkRequest(syscall.RTM_NEWLINK, syscall.NLM_F_CREATE|syscall.NLM_F_EXCL|syscall.NLM_F_ACK)

	msg := nl.NewIfInfomsg(syscall.AF_UNSPEC)
	req.AddData(msg)

	if base.ParentIndex != 0 {
		b := make([]byte, 4)
		native.PutUint32(b, uint32(base.ParentIndex))
		data := nl.NewRtAttr(syscall.IFLA_LINK, b)
		req.AddData(data)
	} else if link.Type() == "ipvlan" {
		return fmt.Errorf("Can't create ipvlan link without ParentIndex")
	}

	nameData := nl.NewRtAttr(syscall.IFLA_IFNAME, nl.ZeroTerminated(base.Name))
	req.AddData(nameData)

	if base.MTU > 0 {
		mtu := nl.NewRtAttr(syscall.IFLA_MTU, nl.Uint32Attr(uint32(base.MTU)))
		req.AddData(mtu)
	}

	linkInfo := nl.NewRtAttr(syscall.IFLA_LINKINFO, nil)
	nl.NewRtAttrChild(linkInfo, nl.IFLA_INFO_KIND, nl.NonZeroTerminated(link.Type()))

	nl.NewRtAttrChild(linkInfo, syscall.IFLA_TXQLEN, nl.Uint32Attr(base.TxQLen))

	if vlan, ok := link.(*Vlan); ok {
		b := make([]byte, 2)
		native.PutUint16(b, uint16(vlan.VlanId))
		data := nl.NewRtAttrChild(linkInfo, nl.IFLA_INFO_DATA, nil)
		nl.NewRtAttrChild(data, nl.IFLA_VLAN_ID, b)
	} else if veth, ok := link.(*Veth); ok {
		data := nl.NewRtAttrChild(linkInfo, nl.IFLA_INFO_DATA, nil)
		peer := nl.NewRtAttrChild(data, nl.VETH_INFO_PEER, nil)
		nl.NewIfInfomsgChild(peer, syscall.AF_UNSPEC)
		nl.NewRtAttrChild(peer, syscall.IFLA_IFNAME, nl.ZeroTerminated(veth.PeerName))
		nl.NewRtAttrChild(peer, syscall.IFLA_TXQLEN, nl.Uint32Attr(base.TxQLen))
		if base.MTU > 0 {
			nl.NewRtAttrChild(peer, syscall.IFLA_MTU, nl.Uint32Attr(uint32(base.MTU)))
		}
	} else if vxlan, ok := link.(*Vxlan); ok {
		addVxlanAttrs(vxlan, linkInfo)
	} else if ipv, ok := link.(*IPVlan); ok {
		data := nl.NewRtAttrChild(linkInfo, nl.IFLA_INFO_DATA, nil)
		nl.NewRtAttrChild(data, nl.IFLA_IPVLAN_MODE, nl.Uint16Attr(uint16(ipv.Mode)))
	}

	req.AddData(linkInfo)

	_, err := req.Execute(syscall.NETLINK_ROUTE, 0)
	if err != nil {
		return err
	}

	ensureIndex(base)

	// can't set master during create, so set it afterwards
	if base.MasterIndex != 0 {
		// TODO: verify MasterIndex is actually a bridge?
		return LinkSetMasterByIndex(link, base.MasterIndex)
	}
	return nil
}

// LinkAdd adds a new link device. Either Index or Name must be set in
// the link object for it to be deleted. The other values are ignored.
// Equivalent to: `ip link del $link`
func LinkDel(link Link) error {
	base := link.Attrs()

	ensureIndex(base)

	req := nl.NewNetlinkRequest(syscall.RTM_DELLINK, syscall.NLM_F_ACK)

	msg := nl.NewIfInfomsg(syscall.AF_UNSPEC)
	msg.Index = int32(base.Index)
	req.AddData(msg)

	_, err := req.Execute(syscall.NETLINK_ROUTE, 0)
	return err
}

func linkByNameDump(name string) (Link, error) {
	links, err := LinkList()
	if err != nil {
		return nil, err
	}

	for _, link := range links {
		if link.Attrs().Name == name {
			return link, nil
		}
	}
	return nil, fmt.Errorf("Link %s not found", name)
}

// LinkByName finds a link by name and returns a pointer to the object.
func LinkByName(name string) (Link, error) {
	if lookupByDump {
		return linkByNameDump(name)
	}

	req := nl.NewNetlinkRequest(syscall.RTM_GETLINK, syscall.NLM_F_ACK)

	msg := nl.NewIfInfomsg(syscall.AF_UNSPEC)
	req.AddData(msg)

	nameData := nl.NewRtAttr(syscall.IFLA_IFNAME, nl.ZeroTerminated(name))
	req.AddData(nameData)

	link, err := execGetLink(req)
	if err == syscall.EINVAL {
		// older kernels don't support looking up via IFLA_IFNAME
		// so fall back to dumping all links
		lookupByDump = true
		return linkByNameDump(name)
	}

	return link, err
}

// LinkByIndex finds a link by index and returns a pointer to the object.
func LinkByIndex(index int) (Link, error) {
	req := nl.NewNetlinkRequest(syscall.RTM_GETLINK, syscall.NLM_F_ACK)

	msg := nl.NewIfInfomsg(syscall.AF_UNSPEC)
	msg.Index = int32(index)
	req.AddData(msg)

	return execGetLink(req)
}

func execGetLink(req *nl.NetlinkRequest) (Link, error) {
	msgs, err := req.Execute(syscall.NETLINK_ROUTE, 0)
	if err != nil {
		if errno, ok := err.(syscall.Errno); ok {
			if errno == syscall.ENODEV {
				return nil, fmt.Errorf("Link not found")
			}
		}
		return nil, err
	}

	switch {
	case len(msgs) == 0:
		return nil, fmt.Errorf("Link not found")

	case len(msgs) == 1:
		return linkDeserialize(msgs[0])

	default:
		return nil, fmt.Errorf("More than one link found")
	}
}

// linkDeserialize deserializes a raw message received from netlink into
// a link object.
func linkDeserialize(m []byte) (Link, error) {
	msg := nl.DeserializeIfInfomsg(m)

	attrs, err := nl.ParseRouteAttr(m[msg.Len():])
	if err != nil {
		return nil, err
	}

	base := LinkAttrs{Index: int(msg.Index), Flags: linkFlags(msg.Flags)}
	var link Link
	linkType := ""
	for _, attr := range attrs {
		switch attr.Attr.Type {
		case syscall.IFLA_LINKINFO:
			infos, err := nl.ParseRouteAttr(attr.Value)
			if err != nil {
				return nil, err
			}
			for _, info := range infos {
				switch info.Attr.Type {
				case nl.IFLA_INFO_KIND:
					linkType = string(info.Value[:len(info.Value)-1])
					switch linkType {
					case "dummy":
						link = &Dummy{}
					case "bridge":
						link = &Bridge{}
					case "vlan":
						link = &Vlan{}
					case "veth":
						link = &Veth{}
					case "vxlan":
						link = &Vxlan{}
					case "ipvlan":
						link = &IPVlan{}
					default:
						link = &Generic{LinkType: linkType}
					}
				case nl.IFLA_INFO_DATA:
					data, err := nl.ParseRouteAttr(info.Value)
					if err != nil {
						return nil, err
					}
					switch linkType {
					case "vlan":
						parseVlanData(link, data)
					case "vxlan":
						parseVxlanData(link, data)
					case "ipvlan":
						parseIPVlanData(link, data)
					}
				}
			}
		case syscall.IFLA_ADDRESS:
			var nonzero bool
			for _, b := range attr.Value {
				if b != 0 {
					nonzero = true
				}
			}
			if nonzero {
				base.HardwareAddr = attr.Value[:]
			}
		case syscall.IFLA_IFNAME:
			base.Name = string(attr.Value[:len(attr.Value)-1])
		case syscall.IFLA_MTU:
			base.MTU = int(native.Uint32(attr.Value[0:4]))
		case syscall.IFLA_LINK:
			base.ParentIndex = int(native.Uint32(attr.Value[0:4]))
		case syscall.IFLA_MASTER:
			base.MasterIndex = int(native.Uint32(attr.Value[0:4]))
		case syscall.IFLA_TXQLEN:
			base.TxQLen = native.Uint32(attr.Value[0:4])
		}
	}
	// Links that don't have IFLA_INFO_KIND are hardware devices
	if link == nil {
		link = &Device{}
	}
	*link.Attrs() = base

	return link, nil
}

// LinkList gets a list of link devices.
// Equivalent to: `ip link show`
func LinkList() ([]Link, error) {
	// NOTE(vish): This duplicates functionality in net/iface_linux.go, but we need
	//             to get the message ourselves to parse link type.
	req := nl.NewNetlinkRequest(syscall.RTM_GETLINK, syscall.NLM_F_DUMP)

	msg := nl.NewIfInfomsg(syscall.AF_UNSPEC)
	req.AddData(msg)

	msgs, err := req.Execute(syscall.NETLINK_ROUTE, syscall.RTM_NEWLINK)
	if err != nil {
		return nil, err
	}

	res := make([]Link, 0)

	for _, m := range msgs {
		link, err := linkDeserialize(m)
		if err != nil {
			return nil, err
		}
		res = append(res, link)
	}

	return res, nil
}

func parseVlanData(link Link, data []syscall.NetlinkRouteAttr) {
	vlan := link.(*Vlan)
	for _, datum := range data {
		switch datum.Attr.Type {
		case nl.IFLA_VLAN_ID:
			vlan.VlanId = int(native.Uint16(datum.Value[0:2]))
		}
	}
}

func parseVxlanData(link Link, data []syscall.NetlinkRouteAttr) {
	vxlan := link.(*Vxlan)
	for _, datum := range data {
		switch datum.Attr.Type {
		case nl.IFLA_VXLAN_ID:
			vxlan.VxlanId = int(native.Uint32(datum.Value[0:4]))
		case nl.IFLA_VXLAN_LINK:
			vxlan.VtepDevIndex = int(native.Uint32(datum.Value[0:4]))
		case nl.IFLA_VXLAN_LOCAL:
			vxlan.SrcAddr = net.IP(datum.Value[0:4])
		case nl.IFLA_VXLAN_LOCAL6:
			vxlan.SrcAddr = net.IP(datum.Value[0:16])
		case nl.IFLA_VXLAN_GROUP:
			vxlan.Group = net.IP(datum.Value[0:4])
		case nl.IFLA_VXLAN_GROUP6:
			vxlan.Group = net.IP(datum.Value[0:16])
		case nl.IFLA_VXLAN_TTL:
			vxlan.TTL = int(datum.Value[0])
		case nl.IFLA_VXLAN_TOS:
			vxlan.TOS = int(datum.Value[0])
		case nl.IFLA_VXLAN_LEARNING:
			vxlan.Learning = int8(datum.Value[0]) != 0
		case nl.IFLA_VXLAN_PROXY:
			vxlan.Proxy = int8(datum.Value[0]) != 0
		case nl.IFLA_VXLAN_RSC:
			vxlan.RSC = int8(datum.Value[0]) != 0
		case nl.IFLA_VXLAN_L2MISS:
			vxlan.L2miss = int8(datum.Value[0]) != 0
		case nl.IFLA_VXLAN_L3MISS:
			vxlan.L3miss = int8(datum.Value[0]) != 0
		case nl.IFLA_VXLAN_AGEING:
			vxlan.Age = int(native.Uint32(datum.Value[0:4]))
			vxlan.NoAge = vxlan.Age == 0
		case nl.IFLA_VXLAN_LIMIT:
			vxlan.Limit = int(native.Uint32(datum.Value[0:4]))
		case nl.IFLA_VXLAN_PORT:
			vxlan.Port = int(native.Uint16(datum.Value[0:2]))
		case nl.IFLA_VXLAN_PORT_RANGE:
			buf := bytes.NewBuffer(datum.Value[0:4])
			var pr vxlanPortRange
			if binary.Read(buf, binary.BigEndian, &pr) != nil {
				vxlan.PortLow = int(pr.Lo)
				vxlan.PortHigh = int(pr.Hi)
			}
		}
	}
}

func parseIPVlanData(link Link, data []syscall.NetlinkRouteAttr) {
	ipv := link.(*IPVlan)
	for _, datum := range data {
		if datum.Attr.Type == nl.IFLA_IPVLAN_MODE {
			ipv.Mode = IPVlanMode(native.Uint32(datum.Value[0:4]))
			return
		}
	}
}

// copied from pkg/net_linux.go
func linkFlags(rawFlags uint32) net.Flags {
	var f net.Flags
	if rawFlags&syscall.IFF_UP != 0 {
		f |= net.FlagUp
	}
	if rawFlags&syscall.IFF_BROADCAST != 0 {
		f |= net.FlagBroadcast
	}
	if rawFlags&syscall.IFF_LOOPBACK != 0 {
		f |= net.FlagLoopback
	}
	if rawFlags&syscall.IFF_POINTOPOINT != 0 {
		f |= net.FlagPointToPoint
	}
	if rawFlags&syscall.IFF_MULTICAST != 0 {
		f |= net.FlagMulticast
	}
	return f
}