func encodeActions(attr *nl.RtAttr, actions []Action) error {
tabIndex := int(nl.TCA_ACT_TAB)
for _, action := range actions {
switch action := action.(type) {
default:
return fmt.Errorf("unknown action type %s", action.Type())
case *MirredAction:
table := nl.NewRtAttrChild(attr, tabIndex, nil)
tabIndex++
nl.NewRtAttrChild(table, nl.TCA_ACT_KIND, nl.ZeroTerminated("mirred"))
aopts := nl.NewRtAttrChild(table, nl.TCA_ACT_OPTIONS, nil)
nl.NewRtAttrChild(aopts, nl.TCA_MIRRED_PARMS, action.Serialize())
case *BpfAction:
table := nl.NewRtAttrChild(attr, tabIndex, nil)
tabIndex++
nl.NewRtAttrChild(table, nl.TCA_ACT_KIND, nl.ZeroTerminated("bpf"))
aopts := nl.NewRtAttrChild(table, nl.TCA_ACT_OPTIONS, nil)
nl.NewRtAttrChild(aopts, nl.TCA_ACT_BPF_PARMS, action.Serialize())
nl.NewRtAttrChild(aopts, nl.TCA_ACT_BPF_FD, nl.Uint32Attr(uint32(action.Fd)))
nl.NewRtAttrChild(aopts, nl.TCA_ACT_BPF_NAME, nl.ZeroTerminated(action.Name))
}
}
return nil
}
func fillService(s *Service) nl.NetlinkRequestData {
cmdAttr := nl.NewRtAttr(ipvsCmdAttrService, nil)
nl.NewRtAttrChild(cmdAttr, ipvsSvcAttrAddressFamily, nl.Uint16Attr(s.AddressFamily))
if s.FWMark != 0 {
nl.NewRtAttrChild(cmdAttr, ipvsSvcAttrFWMark, nl.Uint32Attr(s.FWMark))
} else {
nl.NewRtAttrChild(cmdAttr, ipvsSvcAttrProtocol, nl.Uint16Attr(s.Protocol))
nl.NewRtAttrChild(cmdAttr, ipvsSvcAttrAddress, rawIPData(s.Address))
// Port needs to be in network byte order.
portBuf := new(bytes.Buffer)
binary.Write(portBuf, binary.BigEndian, s.Port)
nl.NewRtAttrChild(cmdAttr, ipvsSvcAttrPort, portBuf.Bytes())
}
nl.NewRtAttrChild(cmdAttr, ipvsSvcAttrSchedName, nl.ZeroTerminated(s.SchedName))
if s.PEName != "" {
nl.NewRtAttrChild(cmdAttr, ipvsSvcAttrPEName, nl.ZeroTerminated(s.PEName))
}
f := &ipvsFlags{
flags: s.Flags,
mask: 0xFFFFFFFF,
}
nl.NewRtAttrChild(cmdAttr, ipvsSvcAttrFlags, f.Serialize())
nl.NewRtAttrChild(cmdAttr, ipvsSvcAttrTimeout, nl.Uint32Attr(s.Timeout))
nl.NewRtAttrChild(cmdAttr, ipvsSvcAttrNetmask, nl.Uint32Attr(s.Netmask))
return cmdAttr
}
func getIPVSFamily() (int, error) {
sock, err := nl.GetNetlinkSocketAt(netns.None(), netns.None(), syscall.NETLINK_GENERIC)
if err != nil {
return 0, err
}
req := newGenlRequest(genlCtrlID, genlCtrlCmdGetFamily)
req.AddData(nl.NewRtAttr(genlCtrlAttrFamilyName, nl.ZeroTerminated("IPVS")))
msgs, err := execute(sock, req, 0)
if err != nil {
return 0, err
}
for _, m := range msgs {
hdr := deserializeGenlMsg(m)
attrs, err := nl.ParseRouteAttr(m[hdr.Len():])
if err != nil {
return 0, err
}
for _, attr := range attrs {
switch int(attr.Attr.Type) {
case genlCtrlAttrFamilyID:
return int(native.Uint16(attr.Value[0:2])), nil
}
}
}
return 0, fmt.Errorf("no family id in the netlink response")
}
// ensureBridgeTxQueueLen() ensures that the bridge interface's TX queue
// length is greater than zero. Due to a CNI <= 0.3.0 'bridge' plugin bug,
// the bridge is initially created with a TX queue length of 0, which gets
// used as the packet limit for FIFO traffic shapers, which drops packets.
// TODO: remove when we can depend on a fixed CNI
func (plugin *kubenetNetworkPlugin) ensureBridgeTxQueueLen() {
bridge, err := netlink.LinkByName(BridgeName)
if err != nil {
return
}
if bridge.Attrs().TxQLen > 0 {
return
}
req := nl.NewNetlinkRequest(syscall.RTM_NEWLINK, syscall.NLM_F_ACK)
msg := nl.NewIfInfomsg(syscall.AF_UNSPEC)
req.AddData(msg)
nameData := nl.NewRtAttr(syscall.IFLA_IFNAME, nl.ZeroTerminated(BridgeName))
req.AddData(nameData)
qlen := nl.NewRtAttr(syscall.IFLA_TXQLEN, nl.Uint32Attr(1000))
req.AddData(qlen)
_, err = req.Execute(syscall.NETLINK_ROUTE, 0)
if err != nil {
glog.V(5).Infof("Failed to set bridge tx queue length: %v", err)
}
}
// ClassAdd will add a class to the system.
// Equivalent to: `tc class add $class`
func ClassAdd(class Class) error {
req := nl.NewNetlinkRequest(syscall.RTM_NEWTCLASS, syscall.NLM_F_CREATE|syscall.NLM_F_EXCL|syscall.NLM_F_ACK)
base := class.Attrs()
msg := &nl.TcMsg{
Family: nl.FAMILY_ALL,
Ifindex: int32(base.LinkIndex),
Handle: base.Handle,
Parent: base.Parent,
}
req.AddData(msg)
req.AddData(nl.NewRtAttr(nl.TCA_KIND, nl.ZeroTerminated(class.Type())))
options := nl.NewRtAttr(nl.TCA_OPTIONS, nil)
if htb, ok := class.(*HtbClass); ok {
opt := nl.TcHtbCopt{}
opt.Rate.Rate = uint32(htb.Rate)
opt.Ceil.Rate = uint32(htb.Ceil)
opt.Buffer = htb.Buffer
opt.Cbuffer = htb.Cbuffer
opt.Quantum = htb.Quantum
opt.Level = htb.Level
opt.Prio = htb.Prio
// TODO: Handle Debug properly. For now default to 0
nl.NewRtAttrChild(options, nl.TCA_HTB_PARMS, opt.Serialize())
}
req.AddData(options)
_, err := req.Execute(syscall.NETLINK_ROUTE, 0)
return err
}
func (h *Handle) addrHandle(link Link, addr *Addr, req *nl.NetlinkRequest) error {
base := link.Attrs()
if addr.Label != "" && !strings.HasPrefix(addr.Label, base.Name) {
return fmt.Errorf("label must begin with interface name")
}
h.ensureIndex(base)
family := nl.GetIPFamily(addr.IP)
msg := nl.NewIfAddrmsg(family)
msg.Index = uint32(base.Index)
msg.Scope = uint8(addr.Scope)
prefixlen, _ := addr.Mask.Size()
msg.Prefixlen = uint8(prefixlen)
req.AddData(msg)
var localAddrData []byte
if family == FAMILY_V4 {
localAddrData = addr.IP.To4()
} else {
localAddrData = addr.IP.To16()
}
localData := nl.NewRtAttr(syscall.IFA_LOCAL, localAddrData)
req.AddData(localData)
var peerAddrData []byte
if addr.Peer != nil {
if family == FAMILY_V4 {
peerAddrData = addr.Peer.IP.To4()
} else {
peerAddrData = addr.Peer.IP.To16()
}
} else {
peerAddrData = localAddrData
}
addressData := nl.NewRtAttr(syscall.IFA_ADDRESS, peerAddrData)
req.AddData(addressData)
if addr.Flags != 0 {
if addr.Flags <= 0xff {
msg.IfAddrmsg.Flags = uint8(addr.Flags)
} else {
b := make([]byte, 4)
native.PutUint32(b, uint32(addr.Flags))
flagsData := nl.NewRtAttr(IFA_FLAGS, b)
req.AddData(flagsData)
}
}
if addr.Label != "" {
labelData := nl.NewRtAttr(syscall.IFA_LABEL, nl.ZeroTerminated(addr.Label))
req.AddData(labelData)
}
_, err := req.Execute(syscall.NETLINK_ROUTE, 0)
return err
}
func encodeActions(attr *nl.RtAttr, actions []Action) error {
tabIndex := int(nl.TCA_ACT_TAB)
for _, action := range actions {
switch action := action.(type) {
default:
return fmt.Errorf("unknown action type %s", action.Type())
case *MirredAction:
table := nl.NewRtAttrChild(attr, tabIndex, nil)
tabIndex++
nl.NewRtAttrChild(table, nl.TCA_ACT_KIND, nl.ZeroTerminated("mirred"))
aopts := nl.NewRtAttrChild(table, nl.TCA_ACT_OPTIONS, nil)
mirred := nl.TcMirred{
Eaction: int32(action.MirredAction),
Ifindex: uint32(action.Ifindex),
}
toTcGen(action.Attrs(), &mirred.TcGen)
nl.NewRtAttrChild(aopts, nl.TCA_MIRRED_PARMS, mirred.Serialize())
case *BpfAction:
table := nl.NewRtAttrChild(attr, tabIndex, nil)
tabIndex++
nl.NewRtAttrChild(table, nl.TCA_ACT_KIND, nl.ZeroTerminated("bpf"))
aopts := nl.NewRtAttrChild(table, nl.TCA_ACT_OPTIONS, nil)
gen := nl.TcGen{}
toTcGen(action.Attrs(), &gen)
nl.NewRtAttrChild(aopts, nl.TCA_ACT_BPF_PARMS, gen.Serialize())
nl.NewRtAttrChild(aopts, nl.TCA_ACT_BPF_FD, nl.Uint32Attr(uint32(action.Fd)))
nl.NewRtAttrChild(aopts, nl.TCA_ACT_BPF_NAME, nl.ZeroTerminated(action.Name))
case *GenericAction:
table := nl.NewRtAttrChild(attr, tabIndex, nil)
tabIndex++
nl.NewRtAttrChild(table, nl.TCA_ACT_KIND, nl.ZeroTerminated("gact"))
aopts := nl.NewRtAttrChild(table, nl.TCA_ACT_OPTIONS, nil)
gen := nl.TcGen{}
toTcGen(action.Attrs(), &gen)
nl.NewRtAttrChild(aopts, nl.TCA_GACT_PARMS, gen.Serialize())
}
}
return nil
}
// FilterAdd will add a filter to the system.
// Equivalent to: `tc filter add $filter`
func FilterAdd(filter Filter) error {
req := nl.NewNetlinkRequest(syscall.RTM_NEWTFILTER, syscall.NLM_F_CREATE|syscall.NLM_F_EXCL|syscall.NLM_F_ACK)
base := filter.Attrs()
msg := &nl.TcMsg{
Family: nl.FAMILY_ALL,
Ifindex: int32(base.LinkIndex),
Handle: base.Handle,
Parent: base.Parent,
Info: MakeHandle(base.Priority, nl.Swap16(base.Protocol)),
}
req.AddData(msg)
req.AddData(nl.NewRtAttr(nl.TCA_KIND, nl.ZeroTerminated(filter.Type())))
options := nl.NewRtAttr(nl.TCA_OPTIONS, nil)
if u32, ok := filter.(*U32); ok {
// match all
sel := nl.TcU32Sel{
Nkeys: 1,
Flags: nl.TC_U32_TERMINAL,
}
sel.Keys = append(sel.Keys, nl.TcU32Key{})
nl.NewRtAttrChild(options, nl.TCA_U32_SEL, sel.Serialize())
actions := nl.NewRtAttrChild(options, nl.TCA_U32_ACT, nil)
table := nl.NewRtAttrChild(actions, nl.TCA_ACT_TAB, nil)
nl.NewRtAttrChild(table, nl.TCA_KIND, nl.ZeroTerminated("mirred"))
// redirect to other interface
mir := nl.TcMirred{
Action: nl.TC_ACT_STOLEN,
Eaction: nl.TCA_EGRESS_REDIR,
Ifindex: uint32(u32.RedirIndex),
}
aopts := nl.NewRtAttrChild(table, nl.TCA_OPTIONS, nil)
nl.NewRtAttrChild(aopts, nl.TCA_MIRRED_PARMS, mir.Serialize())
}
req.AddData(options)
_, err := req.Execute(syscall.NETLINK_ROUTE, 0)
return err
}
// QdiscAdd will add a qdisc to the system.
// Equivalent to: `tc qdisc add $qdisc`
func QdiscAdd(qdisc Qdisc) error {
req := nl.NewNetlinkRequest(syscall.RTM_NEWQDISC, syscall.NLM_F_CREATE|syscall.NLM_F_EXCL|syscall.NLM_F_ACK)
base := qdisc.Attrs()
msg := &nl.TcMsg{
Family: nl.FAMILY_ALL,
Ifindex: int32(base.LinkIndex),
Handle: base.Handle,
Parent: base.Parent,
}
req.AddData(msg)
req.AddData(nl.NewRtAttr(nl.TCA_KIND, nl.ZeroTerminated(qdisc.Type())))
options := nl.NewRtAttr(nl.TCA_OPTIONS, nil)
if prio, ok := qdisc.(*Prio); ok {
tcmap := nl.TcPrioMap{
Bands: int32(prio.Bands),
Priomap: prio.PriorityMap,
}
options = nl.NewRtAttr(nl.TCA_OPTIONS, tcmap.Serialize())
} else if tbf, ok := qdisc.(*Tbf); ok {
opt := nl.TcTbfQopt{}
// TODO: handle rate > uint32
opt.Rate.Rate = uint32(tbf.Rate)
opt.Limit = tbf.Limit
opt.Buffer = tbf.Buffer
nl.NewRtAttrChild(options, nl.TCA_TBF_PARMS, opt.Serialize())
} else if htb, ok := qdisc.(*Htb); ok {
opt := nl.TcHtbGlob{}
opt.Version = htb.Version
opt.Rate2Quantum = htb.Rate2Quantum
opt.Defcls = htb.Defcls
// TODO: Handle Debug properly. For now default to 0
opt.Debug = htb.Debug
opt.DirectPkts = htb.DirectPkts
nl.NewRtAttrChild(options, nl.TCA_HTB_INIT, opt.Serialize())
// nl.NewRtAttrChild(options, nl.TCA_HTB_DIRECT_QLEN, opt.Serialize())
} else if _, ok := qdisc.(*Ingress); ok {
// ingress filters must use the proper handle
if msg.Parent != HANDLE_INGRESS {
return fmt.Errorf("Ingress filters must set Parent to HANDLE_INGRESS")
}
}
req.AddData(options)
_, err := req.Execute(syscall.NETLINK_ROUTE, 0)
return err
}
func classPayload(req *nl.NetlinkRequest, class Class) error {
req.AddData(nl.NewRtAttr(nl.TCA_KIND, nl.ZeroTerminated(class.Type())))
options := nl.NewRtAttr(nl.TCA_OPTIONS, nil)
if htb, ok := class.(*HtbClass); ok {
opt := nl.TcHtbCopt{}
opt.Rate.Rate = uint32(htb.Rate)
opt.Ceil.Rate = uint32(htb.Ceil)
opt.Buffer = htb.Buffer
opt.Cbuffer = htb.Cbuffer
opt.Quantum = htb.Quantum
opt.Level = htb.Level
opt.Prio = htb.Prio
// TODO: Handle Debug properly. For now default to 0
nl.NewRtAttrChild(options, nl.TCA_HTB_PARMS, opt.Serialize())
}
req.AddData(options)
return nil
}
// LinkByAlias finds a link by its alias and returns a pointer to the object.
// If there are multiple links with the alias it returns the first one
func (h *Handle) LinkByAlias(alias string) (Link, error) {
if h.lookupByDump {
return h.linkByAliasDump(alias)
}
req := h.newNetlinkRequest(syscall.RTM_GETLINK, syscall.NLM_F_ACK)
msg := nl.NewIfInfomsg(syscall.AF_UNSPEC)
req.AddData(msg)
nameData := nl.NewRtAttr(syscall.IFLA_IFALIAS, nl.ZeroTerminated(alias))
req.AddData(nameData)
link, err := execGetLink(req)
if err == syscall.EINVAL {
// older kernels don't support looking up via IFLA_IFALIAS
// so fall back to dumping all links
h.lookupByDump = true
return h.linkByAliasDump(alias)
}
return link, err
}
// 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
}
func addrHandle(link Link, addr *Addr, req *nl.NetlinkRequest) error {
base := link.Attrs()
if addr.Label != "" && !strings.HasPrefix(addr.Label, base.Name) {
return fmt.Errorf("label must begin with interface name")
}
ensureIndex(base)
family := nl.GetIPFamily(addr.IP)
msg := nl.NewIfAddrmsg(family)
msg.Index = uint32(base.Index)
prefixlen, _ := addr.Mask.Size()
msg.Prefixlen = uint8(prefixlen)
req.AddData(msg)
var addrData []byte
if family == FAMILY_V4 {
addrData = addr.IP.To4()
} else {
addrData = addr.IP.To16()
}
localData := nl.NewRtAttr(syscall.IFA_LOCAL, addrData)
req.AddData(localData)
addressData := nl.NewRtAttr(syscall.IFA_ADDRESS, addrData)
req.AddData(addressData)
if addr.Label != "" {
labelData := nl.NewRtAttr(syscall.IFA_LABEL, nl.ZeroTerminated(addr.Label))
req.AddData(labelData)
}
_, err := req.Execute(syscall.NETLINK_ROUTE, 0)
return err
}
func classPayload(req *nl.NetlinkRequest, class Class) error {
req.AddData(nl.NewRtAttr(nl.TCA_KIND, nl.ZeroTerminated(class.Type())))
options := nl.NewRtAttr(nl.TCA_OPTIONS, nil)
if htb, ok := class.(*HtbClass); ok {
opt := nl.TcHtbCopt{}
opt.Buffer = htb.Buffer
opt.Cbuffer = htb.Cbuffer
opt.Quantum = htb.Quantum
opt.Level = htb.Level
opt.Prio = htb.Prio
// TODO: Handle Debug properly. For now default to 0
/* Calculate {R,C}Tab and set Rate and Ceil */
cellLog := -1
ccellLog := -1
linklayer := nl.LINKLAYER_ETHERNET
mtu := 1600
var rtab [256]uint32
var ctab [256]uint32
tcrate := nl.TcRateSpec{Rate: uint32(htb.Rate)}
if CalcRtable(&tcrate, rtab, cellLog, uint32(mtu), linklayer) < 0 {
return errors.New("HTB: failed to calculate rate table")
}
opt.Rate = tcrate
tcceil := nl.TcRateSpec{Rate: uint32(htb.Ceil)}
if CalcRtable(&tcceil, ctab, ccellLog, uint32(mtu), linklayer) < 0 {
return errors.New("HTB: failed to calculate ceil rate table")
}
opt.Ceil = tcceil
nl.NewRtAttrChild(options, nl.TCA_HTB_PARMS, opt.Serialize())
nl.NewRtAttrChild(options, nl.TCA_HTB_RTAB, SerializeRtab(rtab))
nl.NewRtAttrChild(options, nl.TCA_HTB_CTAB, SerializeRtab(ctab))
}
req.AddData(options)
return nil
}
// 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 (h *Handle) 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!")
}
if tuntap, ok := link.(*Tuntap); ok {
// TODO: support user
// TODO: support group
// TODO: multi_queue
// TODO: support non- persistent
if tuntap.Mode < syscall.IFF_TUN || tuntap.Mode > syscall.IFF_TAP {
return fmt.Errorf("Tuntap.Mode %v unknown!", tuntap.Mode)
}
file, err := os.OpenFile("/dev/net/tun", os.O_RDWR, 0)
if err != nil {
return err
}
defer file.Close()
var req ifReq
if tuntap.Flags == 0 {
req.Flags = uint16(TUNTAP_DEFAULTS)
} else {
req.Flags = uint16(tuntap.Flags)
}
req.Flags |= uint16(tuntap.Mode)
copy(req.Name[:15], base.Name)
_, _, errno := syscall.Syscall(syscall.SYS_IOCTL, file.Fd(), uintptr(syscall.TUNSETIFF), uintptr(unsafe.Pointer(&req)))
if errno != 0 {
return fmt.Errorf("Tuntap IOCTL TUNSETIFF failed, errno %v", errno)
}
_, _, errno = syscall.Syscall(syscall.SYS_IOCTL, file.Fd(), uintptr(syscall.TUNSETPERSIST), 1)
if errno != 0 {
return fmt.Errorf("Tuntap IOCTL TUNSETPERSIST failed, errno %v", errno)
}
h.ensureIndex(base)
// can't set master during create, so set it afterwards
if base.MasterIndex != 0 {
// TODO: verify MasterIndex is actually a bridge?
return h.LinkSetMasterByIndex(link, base.MasterIndex)
}
return nil
}
req := h.newNetlinkRequest(syscall.RTM_NEWLINK, syscall.NLM_F_CREATE|syscall.NLM_F_EXCL|syscall.NLM_F_ACK)
msg := nl.NewIfInfomsg(syscall.AF_UNSPEC)
// TODO: make it shorter
if base.Flags&net.FlagUp != 0 {
msg.Change = syscall.IFF_UP
msg.Flags = syscall.IFF_UP
}
if base.Flags&net.FlagBroadcast != 0 {
msg.Change |= syscall.IFF_BROADCAST
msg.Flags |= syscall.IFF_BROADCAST
}
if base.Flags&net.FlagLoopback != 0 {
msg.Change |= syscall.IFF_LOOPBACK
msg.Flags |= syscall.IFF_LOOPBACK
}
if base.Flags&net.FlagPointToPoint != 0 {
msg.Change |= syscall.IFF_POINTOPOINT
msg.Flags |= syscall.IFF_POINTOPOINT
}
if base.Flags&net.FlagMulticast != 0 {
msg.Change |= syscall.IFF_MULTICAST
msg.Flags |= syscall.IFF_MULTICAST
}
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)
}
if base.TxQLen >= 0 {
qlen := nl.NewRtAttr(syscall.IFLA_TXQLEN, nl.Uint32Attr(uint32(base.TxQLen)))
req.AddData(qlen)
}
if base.Namespace != nil {
var attr *nl.RtAttr
switch base.Namespace.(type) {
case NsPid:
val := nl.Uint32Attr(uint32(base.Namespace.(NsPid)))
attr = nl.NewRtAttr(syscall.IFLA_NET_NS_PID, val)
case NsFd:
val := nl.Uint32Attr(uint32(base.Namespace.(NsFd)))
attr = nl.NewRtAttr(nl.IFLA_NET_NS_FD, val)
}
req.AddData(attr)
}
if base.Xdp != nil {
addXdpAttrs(base.Xdp, req)
}
linkInfo := nl.NewRtAttr(syscall.IFLA_LINKINFO, nil)
nl.NewRtAttrChild(linkInfo, nl.IFLA_INFO_KIND, nl.NonZeroTerminated(link.Type()))
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))
if base.TxQLen >= 0 {
nl.NewRtAttrChild(peer, syscall.IFLA_TXQLEN, nl.Uint32Attr(uint32(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 bond, ok := link.(*Bond); ok {
addBondAttrs(bond, 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)))
} else if macv, ok := link.(*Macvlan); ok {
if macv.Mode != MACVLAN_MODE_DEFAULT {
data := nl.NewRtAttrChild(linkInfo, nl.IFLA_INFO_DATA, nil)
nl.NewRtAttrChild(data, nl.IFLA_MACVLAN_MODE, nl.Uint32Attr(macvlanModes[macv.Mode]))
}
} else if macv, ok := link.(*Macvtap); ok {
if macv.Mode != MACVLAN_MODE_DEFAULT {
data := nl.NewRtAttrChild(linkInfo, nl.IFLA_INFO_DATA, nil)
nl.NewRtAttrChild(data, nl.IFLA_MACVLAN_MODE, nl.Uint32Attr(macvlanModes[macv.Mode]))
}
} else if gretap, ok := link.(*Gretap); ok {
addGretapAttrs(gretap, linkInfo)
} else if iptun, ok := link.(*Iptun); ok {
addIptunAttrs(iptun, linkInfo)
} else if vti, ok := link.(*Vti); ok {
addVtiAttrs(vti, linkInfo)
}
req.AddData(linkInfo)
_, err := req.Execute(syscall.NETLINK_ROUTE, 0)
if err != nil {
return err
}
h.ensureIndex(base)
// can't set master during create, so set it afterwards
if base.MasterIndex != 0 {
// TODO: verify MasterIndex is actually a bridge?
return h.LinkSetMasterByIndex(link, base.MasterIndex)
}
return nil
}
// FilterAdd will add a filter to the system.
// Equivalent to: `tc filter add $filter`
func FilterAdd(filter Filter) error {
native = nl.NativeEndian()
req := nl.NewNetlinkRequest(syscall.RTM_NEWTFILTER, syscall.NLM_F_CREATE|syscall.NLM_F_EXCL|syscall.NLM_F_ACK)
base := filter.Attrs()
msg := &nl.TcMsg{
Family: nl.FAMILY_ALL,
Ifindex: int32(base.LinkIndex),
Handle: base.Handle,
Parent: base.Parent,
Info: MakeHandle(base.Priority, nl.Swap16(base.Protocol)),
}
req.AddData(msg)
req.AddData(nl.NewRtAttr(nl.TCA_KIND, nl.ZeroTerminated(filter.Type())))
options := nl.NewRtAttr(nl.TCA_OPTIONS, nil)
if u32, ok := filter.(*U32); ok {
// match all
sel := nl.TcU32Sel{
Nkeys: 1,
Flags: nl.TC_U32_TERMINAL,
}
sel.Keys = append(sel.Keys, nl.TcU32Key{})
nl.NewRtAttrChild(options, nl.TCA_U32_SEL, sel.Serialize())
if u32.ClassId != 0 {
nl.NewRtAttrChild(options, nl.TCA_U32_CLASSID, nl.Uint32Attr(u32.ClassId))
}
actionsAttr := nl.NewRtAttrChild(options, nl.TCA_U32_ACT, nil)
// backwards compatibility
if u32.RedirIndex != 0 {
u32.Actions = append([]Action{NewMirredAction(u32.RedirIndex)}, u32.Actions...)
}
if err := encodeActions(actionsAttr, u32.Actions); err != nil {
return err
}
} else if fw, ok := filter.(*Fw); ok {
if fw.Mask != 0 {
b := make([]byte, 4)
native.PutUint32(b, fw.Mask)
nl.NewRtAttrChild(options, nl.TCA_FW_MASK, b)
}
if fw.InDev != "" {
nl.NewRtAttrChild(options, nl.TCA_FW_INDEV, nl.ZeroTerminated(fw.InDev))
}
if (fw.Police != nl.TcPolice{}) {
police := nl.NewRtAttrChild(options, nl.TCA_FW_POLICE, nil)
nl.NewRtAttrChild(police, nl.TCA_POLICE_TBF, fw.Police.Serialize())
if (fw.Police.Rate != nl.TcRateSpec{}) {
payload := SerializeRtab(fw.Rtab)
nl.NewRtAttrChild(police, nl.TCA_POLICE_RATE, payload)
}
if (fw.Police.PeakRate != nl.TcRateSpec{}) {
payload := SerializeRtab(fw.Ptab)
nl.NewRtAttrChild(police, nl.TCA_POLICE_PEAKRATE, payload)
}
}
if fw.ClassId != 0 {
b := make([]byte, 4)
native.PutUint32(b, fw.ClassId)
nl.NewRtAttrChild(options, nl.TCA_FW_CLASSID, b)
}
} else if bpf, ok := filter.(*BpfFilter); ok {
var bpf_flags uint32
if bpf.ClassId != 0 {
nl.NewRtAttrChild(options, nl.TCA_BPF_CLASSID, nl.Uint32Attr(bpf.ClassId))
}
if bpf.Fd >= 0 {
nl.NewRtAttrChild(options, nl.TCA_BPF_FD, nl.Uint32Attr((uint32(bpf.Fd))))
}
if bpf.Name != "" {
nl.NewRtAttrChild(options, nl.TCA_BPF_NAME, nl.ZeroTerminated(bpf.Name))
}
if bpf.DirectAction {
bpf_flags |= nl.TCA_BPF_FLAG_ACT_DIRECT
}
nl.NewRtAttrChild(options, nl.TCA_BPF_FLAGS, nl.Uint32Attr(bpf_flags))
}
req.AddData(options)
_, err := req.Execute(syscall.NETLINK_ROUTE, 0)
return err
}
// QdiscAdd will add a qdisc to the system.
// Equivalent to: `tc qdisc add $qdisc`
func QdiscAdd(qdisc Qdisc) error {
req := nl.NewNetlinkRequest(syscall.RTM_NEWQDISC, syscall.NLM_F_CREATE|syscall.NLM_F_EXCL|syscall.NLM_F_ACK)
base := qdisc.Attrs()
msg := &nl.TcMsg{
Family: nl.FAMILY_ALL,
Ifindex: int32(base.LinkIndex),
Handle: base.Handle,
Parent: base.Parent,
}
req.AddData(msg)
req.AddData(nl.NewRtAttr(nl.TCA_KIND, nl.ZeroTerminated(qdisc.Type())))
options := nl.NewRtAttr(nl.TCA_OPTIONS, nil)
if prio, ok := qdisc.(*Prio); ok {
tcmap := nl.TcPrioMap{
Bands: int32(prio.Bands),
Priomap: prio.PriorityMap,
}
options = nl.NewRtAttr(nl.TCA_OPTIONS, tcmap.Serialize())
} else if tbf, ok := qdisc.(*Tbf); ok {
opt := nl.TcTbfQopt{}
// TODO: handle rate > uint32
opt.Rate.Rate = uint32(tbf.Rate)
opt.Limit = tbf.Limit
opt.Buffer = tbf.Buffer
nl.NewRtAttrChild(options, nl.TCA_TBF_PARMS, opt.Serialize())
} else if htb, ok := qdisc.(*Htb); ok {
opt := nl.TcHtbGlob{}
opt.Version = htb.Version
opt.Rate2Quantum = htb.Rate2Quantum
opt.Defcls = htb.Defcls
// TODO: Handle Debug properly. For now default to 0
opt.Debug = htb.Debug
opt.DirectPkts = htb.DirectPkts
nl.NewRtAttrChild(options, nl.TCA_HTB_INIT, opt.Serialize())
// nl.NewRtAttrChild(options, nl.TCA_HTB_DIRECT_QLEN, opt.Serialize())
} else if netem, ok := qdisc.(*Netem); ok {
opt := nl.TcNetemQopt{}
opt.Latency = netem.Latency
opt.Limit = netem.Limit
opt.Loss = netem.Loss
opt.Gap = netem.Gap
opt.Duplicate = netem.Duplicate
opt.Jitter = netem.Jitter
options = nl.NewRtAttr(nl.TCA_OPTIONS, opt.Serialize())
// Correlation
corr := nl.TcNetemCorr{}
corr.DelayCorr = netem.DelayCorr
corr.LossCorr = netem.LossCorr
corr.DupCorr = netem.DuplicateCorr
if corr.DelayCorr > 0 || corr.LossCorr > 0 || corr.DupCorr > 0 {
nl.NewRtAttrChild(options, nl.TCA_NETEM_CORR, corr.Serialize())
}
// Corruption
corruption := nl.TcNetemCorrupt{}
corruption.Probability = netem.CorruptProb
corruption.Correlation = netem.CorruptCorr
if corruption.Probability > 0 {
nl.NewRtAttrChild(options, nl.TCA_NETEM_CORRUPT, corruption.Serialize())
}
// Reorder
reorder := nl.TcNetemReorder{}
reorder.Probability = netem.ReorderProb
reorder.Correlation = netem.ReorderCorr
if reorder.Probability > 0 {
nl.NewRtAttrChild(options, nl.TCA_NETEM_REORDER, reorder.Serialize())
}
} else if _, ok := qdisc.(*Ingress); ok {
// ingress filters must use the proper handle
if msg.Parent != HANDLE_INGRESS {
return fmt.Errorf("Ingress filters must set Parent to HANDLE_INGRESS")
}
}
req.AddData(options)
_, err := req.Execute(syscall.NETLINK_ROUTE, 0)
return err
}
// 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)
}
if base.TxQLen >= 0 {
qlen := nl.NewRtAttr(syscall.IFLA_TXQLEN, nl.Uint32Attr(uint32(base.TxQLen)))
req.AddData(qlen)
}
if base.Namespace != nil {
var attr *nl.RtAttr
switch base.Namespace.(type) {
case NsPid:
val := nl.Uint32Attr(uint32(base.Namespace.(NsPid)))
attr = nl.NewRtAttr(syscall.IFLA_NET_NS_PID, val)
case NsFd:
val := nl.Uint32Attr(uint32(base.Namespace.(NsFd)))
attr = nl.NewRtAttr(nl.IFLA_NET_NS_FD, val)
}
req.AddData(attr)
}
linkInfo := nl.NewRtAttr(syscall.IFLA_LINKINFO, nil)
nl.NewRtAttrChild(linkInfo, nl.IFLA_INFO_KIND, nl.NonZeroTerminated(link.Type()))
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))
if base.TxQLen >= 0 {
nl.NewRtAttrChild(peer, syscall.IFLA_TXQLEN, nl.Uint32Attr(uint32(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)))
} else if macv, ok := link.(*Macvlan); ok {
if macv.Mode != MACVLAN_MODE_DEFAULT {
data := nl.NewRtAttrChild(linkInfo, nl.IFLA_INFO_DATA, nil)
nl.NewRtAttrChild(data, nl.IFLA_MACVLAN_MODE, nl.Uint32Attr(macvlanModes[macv.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
}
// FilterAdd will add a filter to the system.
// Equivalent to: `tc filter add $filter`
func FilterAdd(filter Filter) error {
native = nl.NativeEndian()
req := nl.NewNetlinkRequest(syscall.RTM_NEWTFILTER, syscall.NLM_F_CREATE|syscall.NLM_F_EXCL|syscall.NLM_F_ACK)
base := filter.Attrs()
msg := &nl.TcMsg{
Family: nl.FAMILY_ALL,
Ifindex: int32(base.LinkIndex),
Handle: base.Handle,
Parent: base.Parent,
Info: MakeHandle(base.Priority, nl.Swap16(base.Protocol)),
}
req.AddData(msg)
req.AddData(nl.NewRtAttr(nl.TCA_KIND, nl.ZeroTerminated(filter.Type())))
options := nl.NewRtAttr(nl.TCA_OPTIONS, nil)
if u32, ok := filter.(*U32); ok {
// match all
sel := nl.TcU32Sel{
Nkeys: 1,
Flags: nl.TC_U32_TERMINAL,
}
sel.Keys = append(sel.Keys, nl.TcU32Key{})
nl.NewRtAttrChild(options, nl.TCA_U32_SEL, sel.Serialize())
actions := nl.NewRtAttrChild(options, nl.TCA_U32_ACT, nil)
table := nl.NewRtAttrChild(actions, nl.TCA_ACT_TAB, nil)
nl.NewRtAttrChild(table, nl.TCA_KIND, nl.ZeroTerminated("mirred"))
// redirect to other interface
mir := nl.TcMirred{
Action: nl.TC_ACT_STOLEN,
Eaction: nl.TCA_EGRESS_REDIR,
Ifindex: uint32(u32.RedirIndex),
}
aopts := nl.NewRtAttrChild(table, nl.TCA_OPTIONS, nil)
nl.NewRtAttrChild(aopts, nl.TCA_MIRRED_PARMS, mir.Serialize())
} else if fw, ok := filter.(*Fw); ok {
if fw.Mask != 0 {
b := make([]byte, 4)
native.PutUint32(b, fw.Mask)
nl.NewRtAttrChild(options, nl.TCA_FW_MASK, b)
}
if fw.InDev != "" {
nl.NewRtAttrChild(options, nl.TCA_FW_INDEV, nl.ZeroTerminated(fw.InDev))
}
if (fw.Police != nl.TcPolice{}) {
police := nl.NewRtAttrChild(options, nl.TCA_FW_POLICE, nil)
nl.NewRtAttrChild(police, nl.TCA_POLICE_TBF, fw.Police.Serialize())
if (fw.Police.Rate != nl.TcRateSpec{}) {
payload := SerializeRtab(fw.Rtab)
nl.NewRtAttrChild(police, nl.TCA_POLICE_RATE, payload)
}
if (fw.Police.PeakRate != nl.TcRateSpec{}) {
payload := SerializeRtab(fw.Ptab)
nl.NewRtAttrChild(police, nl.TCA_POLICE_PEAKRATE, payload)
}
}
if fw.ClassId != 0 {
b := make([]byte, 4)
native.PutUint32(b, fw.ClassId)
nl.NewRtAttrChild(options, nl.TCA_FW_CLASSID, b)
}
}
req.AddData(options)
_, err := req.Execute(syscall.NETLINK_ROUTE, 0)
return err
}
func qdiscPayload(req *nl.NetlinkRequest, qdisc Qdisc) error {
req.AddData(nl.NewRtAttr(nl.TCA_KIND, nl.ZeroTerminated(qdisc.Type())))
options := nl.NewRtAttr(nl.TCA_OPTIONS, nil)
if prio, ok := qdisc.(*Prio); ok {
tcmap := nl.TcPrioMap{
Bands: int32(prio.Bands),
Priomap: prio.PriorityMap,
}
options = nl.NewRtAttr(nl.TCA_OPTIONS, tcmap.Serialize())
} else if tbf, ok := qdisc.(*Tbf); ok {
opt := nl.TcTbfQopt{}
// TODO: handle rate > uint32
opt.Rate.Rate = uint32(tbf.Rate)
opt.Limit = tbf.Limit
opt.Buffer = tbf.Buffer
nl.NewRtAttrChild(options, nl.TCA_TBF_PARMS, opt.Serialize())
} else if htb, ok := qdisc.(*Htb); ok {
opt := nl.TcHtbGlob{}
opt.Version = htb.Version
opt.Rate2Quantum = htb.Rate2Quantum
opt.Defcls = htb.Defcls
// TODO: Handle Debug properly. For now default to 0
opt.Debug = htb.Debug
opt.DirectPkts = htb.DirectPkts
nl.NewRtAttrChild(options, nl.TCA_HTB_INIT, opt.Serialize())
// nl.NewRtAttrChild(options, nl.TCA_HTB_DIRECT_QLEN, opt.Serialize())
} else if netem, ok := qdisc.(*Netem); ok {
opt := nl.TcNetemQopt{}
opt.Latency = netem.Latency
opt.Limit = netem.Limit
opt.Loss = netem.Loss
opt.Gap = netem.Gap
opt.Duplicate = netem.Duplicate
opt.Jitter = netem.Jitter
options = nl.NewRtAttr(nl.TCA_OPTIONS, opt.Serialize())
// Correlation
corr := nl.TcNetemCorr{}
corr.DelayCorr = netem.DelayCorr
corr.LossCorr = netem.LossCorr
corr.DupCorr = netem.DuplicateCorr
if corr.DelayCorr > 0 || corr.LossCorr > 0 || corr.DupCorr > 0 {
nl.NewRtAttrChild(options, nl.TCA_NETEM_CORR, corr.Serialize())
}
// Corruption
corruption := nl.TcNetemCorrupt{}
corruption.Probability = netem.CorruptProb
corruption.Correlation = netem.CorruptCorr
if corruption.Probability > 0 {
nl.NewRtAttrChild(options, nl.TCA_NETEM_CORRUPT, corruption.Serialize())
}
// Reorder
reorder := nl.TcNetemReorder{}
reorder.Probability = netem.ReorderProb
reorder.Correlation = netem.ReorderCorr
if reorder.Probability > 0 {
nl.NewRtAttrChild(options, nl.TCA_NETEM_REORDER, reorder.Serialize())
}
} else if _, ok := qdisc.(*Ingress); ok {
// ingress filters must use the proper handle
if qdisc.Attrs().Parent != HANDLE_INGRESS {
return fmt.Errorf("Ingress filters must set Parent to HANDLE_INGRESS")
}
}
req.AddData(options)
return nil
}