device.go 8.0 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313
  1. // Copyright 2015 flannel authors
  2. //
  3. // Licensed under the Apache License, Version 2.0 (the "License");
  4. // you may not use this file except in compliance with the License.
  5. // You may obtain a copy of the License at
  6. //
  7. // http://www.apache.org/licenses/LICENSE-2.0
  8. //
  9. // Unless required by applicable law or agreed to in writing, software
  10. // distributed under the License is distributed on an "AS IS" BASIS,
  11. // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  12. // See the License for the specific language governing permissions and
  13. // limitations under the License.
  14. package vxlan
  15. import (
  16. "fmt"
  17. "net"
  18. "os"
  19. "syscall"
  20. "time"
  21. log "github.com/golang/glog"
  22. "github.com/vishvananda/netlink"
  23. "github.com/vishvananda/netlink/nl"
  24. "github.com/coreos/flannel/pkg/ip"
  25. )
  26. type vxlanDeviceAttrs struct {
  27. vni uint32
  28. name string
  29. vtepIndex int
  30. vtepAddr net.IP
  31. vtepPort int
  32. gbp bool
  33. }
  34. type vxlanDevice struct {
  35. link *netlink.Vxlan
  36. }
  37. func sysctlSet(path, value string) error {
  38. f, err := os.Create(path)
  39. if err != nil {
  40. return err
  41. }
  42. defer f.Close()
  43. _, err = f.Write([]byte(value))
  44. return err
  45. }
  46. func newVXLANDevice(devAttrs *vxlanDeviceAttrs) (*vxlanDevice, error) {
  47. link := &netlink.Vxlan{
  48. LinkAttrs: netlink.LinkAttrs{
  49. Name: devAttrs.name,
  50. },
  51. VxlanId: int(devAttrs.vni),
  52. VtepDevIndex: devAttrs.vtepIndex,
  53. SrcAddr: devAttrs.vtepAddr,
  54. Port: devAttrs.vtepPort,
  55. Learning: false,
  56. GBP: devAttrs.gbp,
  57. }
  58. link, err := ensureLink(link)
  59. if err != nil {
  60. return nil, err
  61. }
  62. // this enables ARP requests being sent to userspace via netlink
  63. sysctlPath := fmt.Sprintf("/proc/sys/net/ipv4/neigh/%s/app_solicit", devAttrs.name)
  64. sysctlSet(sysctlPath, "3")
  65. return &vxlanDevice{
  66. link: link,
  67. }, nil
  68. }
  69. func ensureLink(vxlan *netlink.Vxlan) (*netlink.Vxlan, error) {
  70. err := netlink.LinkAdd(vxlan)
  71. if err == syscall.EEXIST {
  72. // it's ok if the device already exists as long as config is similar
  73. existing, err := netlink.LinkByName(vxlan.Name)
  74. if err != nil {
  75. return nil, err
  76. }
  77. incompat := vxlanLinksIncompat(vxlan, existing)
  78. if incompat == "" {
  79. return existing.(*netlink.Vxlan), nil
  80. }
  81. // delete existing
  82. log.Warningf("%q already exists with incompatable configuration: %v; recreating device", vxlan.Name, incompat)
  83. if err = netlink.LinkDel(existing); err != nil {
  84. return nil, fmt.Errorf("failed to delete interface: %v", err)
  85. }
  86. // create new
  87. if err = netlink.LinkAdd(vxlan); err != nil {
  88. return nil, fmt.Errorf("failed to create vxlan interface: %v", err)
  89. }
  90. } else if err != nil {
  91. return nil, err
  92. }
  93. ifindex := vxlan.Index
  94. link, err := netlink.LinkByIndex(vxlan.Index)
  95. if err != nil {
  96. return nil, fmt.Errorf("can't locate created vxlan device with index %v", ifindex)
  97. }
  98. var ok bool
  99. if vxlan, ok = link.(*netlink.Vxlan); !ok {
  100. return nil, fmt.Errorf("created vxlan device with index %v is not vxlan", ifindex)
  101. }
  102. return vxlan, nil
  103. }
  104. func (dev *vxlanDevice) Configure(ipn ip.IP4Net) error {
  105. setAddr4(dev.link, ipn.ToIPNet())
  106. if err := netlink.LinkSetUp(dev.link); err != nil {
  107. return fmt.Errorf("failed to set interface %s to UP state: %s", dev.link.Attrs().Name, err)
  108. }
  109. // explicitly add a route since there might be a route for a subnet already
  110. // installed by Docker and then it won't get auto added
  111. route := netlink.Route{
  112. LinkIndex: dev.link.Attrs().Index,
  113. Scope: netlink.SCOPE_UNIVERSE,
  114. Dst: ipn.Network().ToIPNet(),
  115. }
  116. if err := netlink.RouteAdd(&route); err != nil && err != syscall.EEXIST {
  117. return fmt.Errorf("failed to add route (%s -> %s): %v", ipn.Network().String(), dev.link.Attrs().Name, err)
  118. }
  119. return nil
  120. }
  121. func (dev *vxlanDevice) Destroy() {
  122. netlink.LinkDel(dev.link)
  123. }
  124. func (dev *vxlanDevice) MACAddr() net.HardwareAddr {
  125. return dev.link.HardwareAddr
  126. }
  127. func (dev *vxlanDevice) MTU() int {
  128. return dev.link.MTU
  129. }
  130. type neigh struct {
  131. MAC net.HardwareAddr
  132. IP ip.IP4
  133. }
  134. func (dev *vxlanDevice) GetL2List() ([]netlink.Neigh, error) {
  135. log.Infof("calling GetL2List() dev.link.Index: %d ", dev.link.Index)
  136. return netlink.NeighList(dev.link.Index, syscall.AF_BRIDGE)
  137. }
  138. func (dev *vxlanDevice) AddL2(n neigh) error {
  139. log.Infof("calling NeighAdd: %v, %v", n.IP, n.MAC)
  140. return netlink.NeighAdd(&netlink.Neigh{
  141. LinkIndex: dev.link.Index,
  142. State: netlink.NUD_PERMANENT,
  143. Family: syscall.AF_BRIDGE,
  144. Flags: netlink.NTF_SELF,
  145. IP: n.IP.ToIP(),
  146. HardwareAddr: n.MAC,
  147. })
  148. }
  149. func (dev *vxlanDevice) DelL2(n neigh) error {
  150. log.Infof("calling NeighDel: %v, %v", n.IP, n.MAC)
  151. return netlink.NeighDel(&netlink.Neigh{
  152. LinkIndex: dev.link.Index,
  153. Family: syscall.AF_BRIDGE,
  154. Flags: netlink.NTF_SELF,
  155. IP: n.IP.ToIP(),
  156. HardwareAddr: n.MAC,
  157. })
  158. }
  159. func (dev *vxlanDevice) AddL3(n neigh) error {
  160. log.Infof("calling NeighSet: %v, %v", n.IP, n.MAC)
  161. return netlink.NeighSet(&netlink.Neigh{
  162. LinkIndex: dev.link.Index,
  163. State: netlink.NUD_REACHABLE,
  164. Type: syscall.RTN_UNICAST,
  165. IP: n.IP.ToIP(),
  166. HardwareAddr: n.MAC,
  167. })
  168. }
  169. func (dev *vxlanDevice) DelL3(n neigh) error {
  170. log.Infof("calling NeighDel: %v, %v", n.IP, n.MAC)
  171. return netlink.NeighDel(&netlink.Neigh{
  172. LinkIndex: dev.link.Index,
  173. State: netlink.NUD_REACHABLE,
  174. Type: syscall.RTN_UNICAST,
  175. IP: n.IP.ToIP(),
  176. HardwareAddr: n.MAC,
  177. })
  178. }
  179. func (dev *vxlanDevice) MonitorMisses(misses chan *netlink.Neigh) {
  180. nlsock, err := nl.Subscribe(syscall.NETLINK_ROUTE, syscall.RTNLGRP_NEIGH)
  181. if err != nil {
  182. log.Error("Failed to subscribe to netlink RTNLGRP_NEIGH messages")
  183. return
  184. }
  185. for {
  186. msgs, err := nlsock.Receive()
  187. if err != nil {
  188. log.Errorf("Failed to receive from netlink: %v ", err)
  189. time.Sleep(1 * time.Second)
  190. continue
  191. }
  192. for _, msg := range msgs {
  193. dev.processNeighMsg(msg, misses)
  194. }
  195. }
  196. }
  197. func isNeighResolving(state int) bool {
  198. return (state & (netlink.NUD_INCOMPLETE | netlink.NUD_STALE | netlink.NUD_DELAY | netlink.NUD_PROBE)) != 0
  199. }
  200. func (dev *vxlanDevice) processNeighMsg(msg syscall.NetlinkMessage, misses chan *netlink.Neigh) {
  201. neigh, err := netlink.NeighDeserialize(msg.Data)
  202. if err != nil {
  203. log.Error("Failed to deserialize netlink ndmsg: %v", err)
  204. return
  205. }
  206. if int(neigh.LinkIndex) != dev.link.Index {
  207. return
  208. }
  209. if msg.Header.Type != syscall.RTM_GETNEIGH && msg.Header.Type != syscall.RTM_NEWNEIGH {
  210. return
  211. }
  212. if !isNeighResolving(neigh.State) {
  213. // misses come with NUD_STALE bit set
  214. return
  215. }
  216. misses <- neigh
  217. }
  218. func vxlanLinksIncompat(l1, l2 netlink.Link) string {
  219. if l1.Type() != l2.Type() {
  220. return fmt.Sprintf("link type: %v vs %v", l1.Type(), l2.Type())
  221. }
  222. v1 := l1.(*netlink.Vxlan)
  223. v2 := l2.(*netlink.Vxlan)
  224. if v1.VxlanId != v2.VxlanId {
  225. return fmt.Sprintf("vni: %v vs %v", v1.VxlanId, v2.VxlanId)
  226. }
  227. if v1.VtepDevIndex > 0 && v2.VtepDevIndex > 0 && v1.VtepDevIndex != v2.VtepDevIndex {
  228. return fmt.Sprintf("vtep (external) interface: %v vs %v", v1.VtepDevIndex, v2.VtepDevIndex)
  229. }
  230. if len(v1.SrcAddr) > 0 && len(v2.SrcAddr) > 0 && !v1.SrcAddr.Equal(v2.SrcAddr) {
  231. return fmt.Sprintf("vtep (external) IP: %v vs %v", v1.SrcAddr, v2.SrcAddr)
  232. }
  233. if len(v1.Group) > 0 && len(v2.Group) > 0 && !v1.Group.Equal(v2.Group) {
  234. return fmt.Sprintf("group address: %v vs %v", v1.Group, v2.Group)
  235. }
  236. if v1.L2miss != v2.L2miss {
  237. return fmt.Sprintf("l2miss: %v vs %v", v1.L2miss, v2.L2miss)
  238. }
  239. if v1.Port > 0 && v2.Port > 0 && v1.Port != v2.Port {
  240. return fmt.Sprintf("port: %v vs %v", v1.Port, v2.Port)
  241. }
  242. return ""
  243. }
  244. // sets IP4 addr on link removing any existing ones first
  245. func setAddr4(link *netlink.Vxlan, ipn *net.IPNet) error {
  246. addrs, err := netlink.AddrList(link, syscall.AF_INET)
  247. if err != nil {
  248. return err
  249. }
  250. for _, addr := range addrs {
  251. if err = netlink.AddrDel(link, &addr); err != nil {
  252. return fmt.Errorf("failed to delete IPv4 addr %s from %s", addr.String(), link.Attrs().Name)
  253. }
  254. }
  255. addr := netlink.Addr{IPNet: ipn, Label: ""}
  256. if err = netlink.AddrAdd(link, &addr); err != nil {
  257. return fmt.Errorf("failed to add IP address %s to %s: %s", ipn.String(), link.Attrs().Name, err)
  258. }
  259. return nil
  260. }