interface.go 8.4 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313
  1. package nebula
  2. import (
  3. "errors"
  4. "io"
  5. "net"
  6. "os"
  7. "runtime"
  8. "time"
  9. "github.com/rcrowley/go-metrics"
  10. "github.com/sirupsen/logrus"
  11. "github.com/slackhq/nebula/cert"
  12. )
  13. const mtu = 9001
  14. type Inside interface {
  15. io.ReadWriteCloser
  16. Activate() error
  17. CidrNet() *net.IPNet
  18. DeviceName() string
  19. WriteRaw([]byte) error
  20. NewMultiQueueReader() (io.ReadWriteCloser, error)
  21. }
  22. type InterfaceConfig struct {
  23. HostMap *HostMap
  24. Outside *udpConn
  25. Inside Inside
  26. certState *CertState
  27. Cipher string
  28. Firewall *Firewall
  29. ServeDns bool
  30. HandshakeManager *HandshakeManager
  31. lightHouse *LightHouse
  32. checkInterval int
  33. pendingDeletionInterval int
  34. DropLocalBroadcast bool
  35. DropMulticast bool
  36. UDPBatchSize int
  37. routines int
  38. MessageMetrics *MessageMetrics
  39. version string
  40. caPool *cert.NebulaCAPool
  41. ConntrackCacheTimeout time.Duration
  42. l *logrus.Logger
  43. }
  44. type Interface struct {
  45. hostMap *HostMap
  46. outside *udpConn
  47. inside Inside
  48. certState *CertState
  49. cipher string
  50. firewall *Firewall
  51. connectionManager *connectionManager
  52. handshakeManager *HandshakeManager
  53. serveDns bool
  54. createTime time.Time
  55. lightHouse *LightHouse
  56. localBroadcast uint32
  57. myVpnIp uint32
  58. dropLocalBroadcast bool
  59. dropMulticast bool
  60. udpBatchSize int
  61. routines int
  62. caPool *cert.NebulaCAPool
  63. // rebindCount is used to decide if an active tunnel should trigger a punch notification through a lighthouse
  64. rebindCount int8
  65. version string
  66. conntrackCacheTimeout time.Duration
  67. writers []*udpConn
  68. readers []io.ReadWriteCloser
  69. metricHandshakes metrics.Histogram
  70. messageMetrics *MessageMetrics
  71. cachedPacketMetrics *cachedPacketMetrics
  72. l *logrus.Logger
  73. }
  74. func NewInterface(c *InterfaceConfig) (*Interface, error) {
  75. if c.Outside == nil {
  76. return nil, errors.New("no outside connection")
  77. }
  78. if c.Inside == nil {
  79. return nil, errors.New("no inside interface (tun)")
  80. }
  81. if c.certState == nil {
  82. return nil, errors.New("no certificate state")
  83. }
  84. if c.Firewall == nil {
  85. return nil, errors.New("no firewall rules")
  86. }
  87. ifce := &Interface{
  88. hostMap: c.HostMap,
  89. outside: c.Outside,
  90. inside: c.Inside,
  91. certState: c.certState,
  92. cipher: c.Cipher,
  93. firewall: c.Firewall,
  94. serveDns: c.ServeDns,
  95. handshakeManager: c.HandshakeManager,
  96. createTime: time.Now(),
  97. lightHouse: c.lightHouse,
  98. localBroadcast: ip2int(c.certState.certificate.Details.Ips[0].IP) | ^ip2int(c.certState.certificate.Details.Ips[0].Mask),
  99. dropLocalBroadcast: c.DropLocalBroadcast,
  100. dropMulticast: c.DropMulticast,
  101. udpBatchSize: c.UDPBatchSize,
  102. routines: c.routines,
  103. version: c.version,
  104. writers: make([]*udpConn, c.routines),
  105. readers: make([]io.ReadWriteCloser, c.routines),
  106. caPool: c.caPool,
  107. myVpnIp: ip2int(c.certState.certificate.Details.Ips[0].IP),
  108. conntrackCacheTimeout: c.ConntrackCacheTimeout,
  109. metricHandshakes: metrics.GetOrRegisterHistogram("handshakes", nil, metrics.NewExpDecaySample(1028, 0.015)),
  110. messageMetrics: c.MessageMetrics,
  111. cachedPacketMetrics: &cachedPacketMetrics{
  112. sent: metrics.GetOrRegisterCounter("hostinfo.cached_packets.sent", nil),
  113. dropped: metrics.GetOrRegisterCounter("hostinfo.cached_packets.dropped", nil),
  114. },
  115. l: c.l,
  116. }
  117. ifce.connectionManager = newConnectionManager(c.l, ifce, c.checkInterval, c.pendingDeletionInterval)
  118. return ifce, nil
  119. }
  120. // activate creates the interface on the host. After the interface is created, any
  121. // other services that want to bind listeners to its IP may do so successfully. However,
  122. // the interface isn't going to process anything until run() is called.
  123. func (f *Interface) activate() {
  124. // actually turn on tun dev
  125. addr, err := f.outside.LocalAddr()
  126. if err != nil {
  127. f.l.WithError(err).Error("Failed to get udp listen address")
  128. }
  129. f.l.WithField("interface", f.inside.DeviceName()).WithField("network", f.inside.CidrNet().String()).
  130. WithField("build", f.version).WithField("udpAddr", addr).
  131. Info("Nebula interface is active")
  132. metrics.GetOrRegisterGauge("routines", nil).Update(int64(f.routines))
  133. // Prepare n tun queues
  134. var reader io.ReadWriteCloser = f.inside
  135. for i := 0; i < f.routines; i++ {
  136. if i > 0 {
  137. reader, err = f.inside.NewMultiQueueReader()
  138. if err != nil {
  139. f.l.Fatal(err)
  140. }
  141. }
  142. f.readers[i] = reader
  143. }
  144. if err := f.inside.Activate(); err != nil {
  145. f.l.Fatal(err)
  146. }
  147. }
  148. func (f *Interface) run() {
  149. // Launch n queues to read packets from udp
  150. for i := 0; i < f.routines; i++ {
  151. go f.listenOut(i)
  152. }
  153. // Launch n queues to read packets from tun dev
  154. for i := 0; i < f.routines; i++ {
  155. go f.listenIn(f.readers[i], i)
  156. }
  157. }
  158. func (f *Interface) listenOut(i int) {
  159. runtime.LockOSThread()
  160. var li *udpConn
  161. // TODO clean this up with a coherent interface for each outside connection
  162. if i > 0 {
  163. li = f.writers[i]
  164. } else {
  165. li = f.outside
  166. }
  167. li.ListenOut(f, i)
  168. }
  169. func (f *Interface) listenIn(reader io.ReadWriteCloser, i int) {
  170. runtime.LockOSThread()
  171. packet := make([]byte, mtu)
  172. out := make([]byte, mtu)
  173. fwPacket := &FirewallPacket{}
  174. nb := make([]byte, 12, 12)
  175. conntrackCache := NewConntrackCacheTicker(f.conntrackCacheTimeout)
  176. for {
  177. n, err := reader.Read(packet)
  178. if err != nil {
  179. f.l.WithError(err).Error("Error while reading outbound packet")
  180. // This only seems to happen when something fatal happens to the fd, so exit.
  181. os.Exit(2)
  182. }
  183. f.consumeInsidePacket(packet[:n], fwPacket, nb, out, i, conntrackCache.Get(f.l))
  184. }
  185. }
  186. func (f *Interface) RegisterConfigChangeCallbacks(c *Config) {
  187. c.RegisterReloadCallback(f.reloadCA)
  188. c.RegisterReloadCallback(f.reloadCertKey)
  189. c.RegisterReloadCallback(f.reloadFirewall)
  190. for _, udpConn := range f.writers {
  191. c.RegisterReloadCallback(udpConn.reloadConfig)
  192. }
  193. }
  194. func (f *Interface) reloadCA(c *Config) {
  195. // reload and check regardless
  196. // todo: need mutex?
  197. newCAs, err := loadCAFromConfig(f.l, c)
  198. if err != nil {
  199. f.l.WithError(err).Error("Could not refresh trusted CA certificates")
  200. return
  201. }
  202. f.caPool = newCAs
  203. f.l.WithField("fingerprints", f.caPool.GetFingerprints()).Info("Trusted CA certificates refreshed")
  204. }
  205. func (f *Interface) reloadCertKey(c *Config) {
  206. // reload and check in all cases
  207. cs, err := NewCertStateFromConfig(c)
  208. if err != nil {
  209. f.l.WithError(err).Error("Could not refresh client cert")
  210. return
  211. }
  212. // did IP in cert change? if so, don't set
  213. oldIPs := f.certState.certificate.Details.Ips
  214. newIPs := cs.certificate.Details.Ips
  215. if len(oldIPs) > 0 && len(newIPs) > 0 && oldIPs[0].String() != newIPs[0].String() {
  216. f.l.WithField("new_ip", newIPs[0]).WithField("old_ip", oldIPs[0]).Error("IP in new cert was different from old")
  217. return
  218. }
  219. f.certState = cs
  220. f.l.WithField("cert", cs.certificate).Info("Client cert refreshed from disk")
  221. }
  222. func (f *Interface) reloadFirewall(c *Config) {
  223. //TODO: need to trigger/detect if the certificate changed too
  224. if c.HasChanged("firewall") == false {
  225. f.l.Debug("No firewall config change detected")
  226. return
  227. }
  228. fw, err := NewFirewallFromConfig(f.l, f.certState.certificate, c)
  229. if err != nil {
  230. f.l.WithError(err).Error("Error while creating firewall during reload")
  231. return
  232. }
  233. oldFw := f.firewall
  234. conntrack := oldFw.Conntrack
  235. conntrack.Lock()
  236. defer conntrack.Unlock()
  237. fw.rulesVersion = oldFw.rulesVersion + 1
  238. // If rulesVersion is back to zero, we have wrapped all the way around. Be
  239. // safe and just reset conntrack in this case.
  240. if fw.rulesVersion == 0 {
  241. f.l.WithField("firewallHash", fw.GetRuleHash()).
  242. WithField("oldFirewallHash", oldFw.GetRuleHash()).
  243. WithField("rulesVersion", fw.rulesVersion).
  244. Warn("firewall rulesVersion has overflowed, resetting conntrack")
  245. } else {
  246. fw.Conntrack = conntrack
  247. }
  248. f.firewall = fw
  249. oldFw.Destroy()
  250. f.l.WithField("firewallHash", fw.GetRuleHash()).
  251. WithField("oldFirewallHash", oldFw.GetRuleHash()).
  252. WithField("rulesVersion", fw.rulesVersion).
  253. Info("New firewall has been installed")
  254. }
  255. func (f *Interface) emitStats(i time.Duration) {
  256. ticker := time.NewTicker(i)
  257. udpStats := NewUDPStatsEmitter(f.writers)
  258. for range ticker.C {
  259. f.firewall.EmitStats()
  260. f.handshakeManager.EmitStats()
  261. udpStats()
  262. }
  263. }