interface.go 13 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479
  1. package nebula
  2. import (
  3. "context"
  4. "errors"
  5. "fmt"
  6. "io"
  7. "net/netip"
  8. "runtime"
  9. "sync"
  10. "sync/atomic"
  11. "time"
  12. "github.com/gaissmai/bart"
  13. "github.com/rcrowley/go-metrics"
  14. "github.com/sirupsen/logrus"
  15. "github.com/slackhq/nebula/config"
  16. "github.com/slackhq/nebula/firewall"
  17. "github.com/slackhq/nebula/header"
  18. "github.com/slackhq/nebula/overlay"
  19. "github.com/slackhq/nebula/udp"
  20. )
  21. const mtu = 9001
  22. type InterfaceConfig struct {
  23. HostMap *HostMap
  24. Outside udp.Conn
  25. Inside overlay.Device
  26. pki *PKI
  27. Cipher string
  28. Firewall *Firewall
  29. ServeDns bool
  30. HandshakeManager *HandshakeManager
  31. lightHouse *LightHouse
  32. connectionManager *connectionManager
  33. DropLocalBroadcast bool
  34. DropMulticast bool
  35. routines int
  36. MessageMetrics *MessageMetrics
  37. version string
  38. relayManager *relayManager
  39. punchy *Punchy
  40. tryPromoteEvery uint32
  41. reQueryEvery uint32
  42. reQueryWait time.Duration
  43. ConntrackCacheTimeout time.Duration
  44. l *logrus.Logger
  45. }
  46. type Interface struct {
  47. hostMap *HostMap
  48. outside udp.Conn
  49. inside overlay.Device
  50. pki *PKI
  51. firewall *Firewall
  52. connectionManager *connectionManager
  53. handshakeManager *HandshakeManager
  54. serveDns bool
  55. createTime time.Time
  56. lightHouse *LightHouse
  57. myBroadcastAddrsTable *bart.Lite
  58. myVpnAddrs []netip.Addr // A list of addresses assigned to us via our certificate
  59. myVpnAddrsTable *bart.Lite
  60. myVpnNetworks []netip.Prefix // A list of networks assigned to us via our certificate
  61. myVpnNetworksTable *bart.Lite
  62. dropLocalBroadcast bool
  63. dropMulticast bool
  64. routines int
  65. disconnectInvalid atomic.Bool
  66. closed atomic.Bool
  67. relayManager *relayManager
  68. tryPromoteEvery atomic.Uint32
  69. reQueryEvery atomic.Uint32
  70. reQueryWait atomic.Int64
  71. sendRecvErrorConfig sendRecvErrorConfig
  72. // rebindCount is used to decide if an active tunnel should trigger a punch notification through a lighthouse
  73. rebindCount int8
  74. version string
  75. conntrackCacheTimeout time.Duration
  76. writers []udp.Conn
  77. readers []io.ReadWriteCloser
  78. wg sync.WaitGroup
  79. metricHandshakes metrics.Histogram
  80. messageMetrics *MessageMetrics
  81. cachedPacketMetrics *cachedPacketMetrics
  82. l *logrus.Logger
  83. }
  84. type EncWriter interface {
  85. SendVia(via *HostInfo,
  86. relay *Relay,
  87. ad,
  88. nb,
  89. out []byte,
  90. nocopy bool,
  91. )
  92. SendMessageToVpnAddr(t header.MessageType, st header.MessageSubType, vpnAddr netip.Addr, p, nb, out []byte)
  93. SendMessageToHostInfo(t header.MessageType, st header.MessageSubType, hostinfo *HostInfo, p, nb, out []byte)
  94. Handshake(vpnAddr netip.Addr)
  95. GetHostInfo(vpnAddr netip.Addr) *HostInfo
  96. GetCertState() *CertState
  97. }
  98. type sendRecvErrorConfig uint8
  99. const (
  100. sendRecvErrorAlways sendRecvErrorConfig = iota
  101. sendRecvErrorNever
  102. sendRecvErrorPrivate
  103. )
  104. func (s sendRecvErrorConfig) ShouldSendRecvError(endpoint netip.AddrPort) bool {
  105. switch s {
  106. case sendRecvErrorPrivate:
  107. return endpoint.Addr().IsPrivate()
  108. case sendRecvErrorAlways:
  109. return true
  110. case sendRecvErrorNever:
  111. return false
  112. default:
  113. panic(fmt.Errorf("invalid sendRecvErrorConfig value: %d", s))
  114. }
  115. }
  116. func (s sendRecvErrorConfig) String() string {
  117. switch s {
  118. case sendRecvErrorAlways:
  119. return "always"
  120. case sendRecvErrorNever:
  121. return "never"
  122. case sendRecvErrorPrivate:
  123. return "private"
  124. default:
  125. return fmt.Sprintf("invalid(%d)", s)
  126. }
  127. }
  128. func NewInterface(ctx context.Context, c *InterfaceConfig) (*Interface, error) {
  129. if c.Outside == nil {
  130. return nil, errors.New("no outside connection")
  131. }
  132. if c.Inside == nil {
  133. return nil, errors.New("no inside interface (tun)")
  134. }
  135. if c.pki == nil {
  136. return nil, errors.New("no certificate state")
  137. }
  138. if c.Firewall == nil {
  139. return nil, errors.New("no firewall rules")
  140. }
  141. if c.connectionManager == nil {
  142. return nil, errors.New("no connection manager")
  143. }
  144. cs := c.pki.getCertState()
  145. ifce := &Interface{
  146. pki: c.pki,
  147. hostMap: c.HostMap,
  148. outside: c.Outside,
  149. inside: c.Inside,
  150. firewall: c.Firewall,
  151. serveDns: c.ServeDns,
  152. handshakeManager: c.HandshakeManager,
  153. createTime: time.Now(),
  154. lightHouse: c.lightHouse,
  155. dropLocalBroadcast: c.DropLocalBroadcast,
  156. dropMulticast: c.DropMulticast,
  157. routines: c.routines,
  158. version: c.version,
  159. writers: make([]udp.Conn, c.routines),
  160. readers: make([]io.ReadWriteCloser, c.routines),
  161. myVpnNetworks: cs.myVpnNetworks,
  162. myVpnNetworksTable: cs.myVpnNetworksTable,
  163. myVpnAddrs: cs.myVpnAddrs,
  164. myVpnAddrsTable: cs.myVpnAddrsTable,
  165. myBroadcastAddrsTable: cs.myVpnBroadcastAddrsTable,
  166. relayManager: c.relayManager,
  167. connectionManager: c.connectionManager,
  168. conntrackCacheTimeout: c.ConntrackCacheTimeout,
  169. metricHandshakes: metrics.GetOrRegisterHistogram("handshakes", nil, metrics.NewExpDecaySample(1028, 0.015)),
  170. messageMetrics: c.MessageMetrics,
  171. cachedPacketMetrics: &cachedPacketMetrics{
  172. sent: metrics.GetOrRegisterCounter("hostinfo.cached_packets.sent", nil),
  173. dropped: metrics.GetOrRegisterCounter("hostinfo.cached_packets.dropped", nil),
  174. },
  175. l: c.l,
  176. }
  177. ifce.tryPromoteEvery.Store(c.tryPromoteEvery)
  178. ifce.reQueryEvery.Store(c.reQueryEvery)
  179. ifce.reQueryWait.Store(int64(c.reQueryWait))
  180. ifce.connectionManager.intf = ifce
  181. return ifce, nil
  182. }
  183. // activate creates the interface on the host. After the interface is created, any
  184. // other services that want to bind listeners to its IP may do so successfully. However,
  185. // the interface isn't going to process anything until run() is called.
  186. func (f *Interface) activate() error {
  187. // actually turn on tun dev
  188. addr, err := f.outside.LocalAddr()
  189. if err != nil {
  190. f.l.WithError(err).Error("Failed to get udp listen address")
  191. }
  192. f.l.WithField("interface", f.inside.Name()).WithField("networks", f.myVpnNetworks).
  193. WithField("build", f.version).WithField("udpAddr", addr).
  194. WithField("boringcrypto", boringEnabled()).
  195. Info("Nebula interface is active")
  196. metrics.GetOrRegisterGauge("routines", nil).Update(int64(f.routines))
  197. // Prepare n tun queues
  198. var reader io.ReadWriteCloser = f.inside
  199. for i := 0; i < f.routines; i++ {
  200. if i > 0 {
  201. reader, err = f.inside.NewMultiQueueReader()
  202. if err != nil {
  203. return err
  204. }
  205. }
  206. f.readers[i] = reader
  207. }
  208. if err = f.inside.Activate(); err != nil {
  209. f.inside.Close()
  210. return err
  211. }
  212. return nil
  213. }
  214. func (f *Interface) run() (func(), error) {
  215. // Launch n queues to read packets from udp
  216. for i := 0; i < f.routines; i++ {
  217. go f.listenOut(i)
  218. f.wg.Add(1)
  219. }
  220. // Launch n queues to read packets from tun dev
  221. for i := 0; i < f.routines; i++ {
  222. go f.listenIn(f.readers[i], i)
  223. f.wg.Add(1)
  224. }
  225. return f.wg.Wait, nil
  226. }
  227. func (f *Interface) listenOut(i int) {
  228. runtime.LockOSThread()
  229. var li udp.Conn
  230. if i > 0 {
  231. li = f.writers[i]
  232. } else {
  233. li = f.outside
  234. }
  235. ctCache := firewall.NewConntrackCacheTicker(f.conntrackCacheTimeout)
  236. lhh := f.lightHouse.NewRequestHandler()
  237. plaintext := make([]byte, udp.MTU)
  238. h := &header.H{}
  239. fwPacket := &firewall.Packet{}
  240. nb := make([]byte, 12, 12)
  241. li.ListenOut(func(fromUdpAddr netip.AddrPort, payload []byte) {
  242. f.readOutsidePackets(fromUdpAddr, nil, plaintext[:0], payload, h, fwPacket, lhh, nb, i, ctCache.Get(f.l))
  243. })
  244. f.l.Errorf("udp reader %v is done", i)
  245. f.wg.Done()
  246. }
  247. func (f *Interface) listenIn(reader io.ReadWriteCloser, i int) {
  248. packet := make([]byte, mtu)
  249. out := make([]byte, mtu)
  250. fwPacket := &firewall.Packet{}
  251. nb := make([]byte, 12, 12)
  252. conntrackCache := firewall.NewConntrackCacheTicker(f.conntrackCacheTimeout)
  253. for {
  254. n, err := reader.Read(packet)
  255. if err != nil {
  256. if !f.closed.Load() {
  257. //TODO: should we close? yes
  258. f.l.WithError(err).Error("Error while reading outbound packet")
  259. }
  260. break
  261. }
  262. f.consumeInsidePacket(packet[:n], fwPacket, nb, out, i, conntrackCache.Get(f.l))
  263. }
  264. f.l.Errorf("tun reader %v is done", i)
  265. f.wg.Done()
  266. }
  267. func (f *Interface) RegisterConfigChangeCallbacks(c *config.C) {
  268. c.RegisterReloadCallback(f.reloadFirewall)
  269. c.RegisterReloadCallback(f.reloadSendRecvError)
  270. c.RegisterReloadCallback(f.reloadDisconnectInvalid)
  271. c.RegisterReloadCallback(f.reloadMisc)
  272. for _, udpConn := range f.writers {
  273. c.RegisterReloadCallback(udpConn.ReloadConfig)
  274. }
  275. }
  276. func (f *Interface) reloadDisconnectInvalid(c *config.C) {
  277. initial := c.InitialLoad()
  278. if initial || c.HasChanged("pki.disconnect_invalid") {
  279. f.disconnectInvalid.Store(c.GetBool("pki.disconnect_invalid", true))
  280. if !initial {
  281. f.l.Infof("pki.disconnect_invalid changed to %v", f.disconnectInvalid.Load())
  282. }
  283. }
  284. }
  285. func (f *Interface) reloadFirewall(c *config.C) {
  286. //TODO: need to trigger/detect if the certificate changed too
  287. if c.HasChanged("firewall") == false {
  288. f.l.Debug("No firewall config change detected")
  289. return
  290. }
  291. fw, err := NewFirewallFromConfig(f.l, f.pki.getCertState(), c)
  292. if err != nil {
  293. f.l.WithError(err).Error("Error while creating firewall during reload")
  294. return
  295. }
  296. oldFw := f.firewall
  297. conntrack := oldFw.Conntrack
  298. conntrack.Lock()
  299. defer conntrack.Unlock()
  300. fw.rulesVersion = oldFw.rulesVersion + 1
  301. // If rulesVersion is back to zero, we have wrapped all the way around. Be
  302. // safe and just reset conntrack in this case.
  303. if fw.rulesVersion == 0 {
  304. f.l.WithField("firewallHashes", fw.GetRuleHashes()).
  305. WithField("oldFirewallHashes", oldFw.GetRuleHashes()).
  306. WithField("rulesVersion", fw.rulesVersion).
  307. Warn("firewall rulesVersion has overflowed, resetting conntrack")
  308. } else {
  309. fw.Conntrack = conntrack
  310. }
  311. f.firewall = fw
  312. oldFw.Destroy()
  313. f.l.WithField("firewallHashes", fw.GetRuleHashes()).
  314. WithField("oldFirewallHashes", oldFw.GetRuleHashes()).
  315. WithField("rulesVersion", fw.rulesVersion).
  316. Info("New firewall has been installed")
  317. }
  318. func (f *Interface) reloadSendRecvError(c *config.C) {
  319. if c.InitialLoad() || c.HasChanged("listen.send_recv_error") {
  320. stringValue := c.GetString("listen.send_recv_error", "always")
  321. switch stringValue {
  322. case "always":
  323. f.sendRecvErrorConfig = sendRecvErrorAlways
  324. case "never":
  325. f.sendRecvErrorConfig = sendRecvErrorNever
  326. case "private":
  327. f.sendRecvErrorConfig = sendRecvErrorPrivate
  328. default:
  329. if c.GetBool("listen.send_recv_error", true) {
  330. f.sendRecvErrorConfig = sendRecvErrorAlways
  331. } else {
  332. f.sendRecvErrorConfig = sendRecvErrorNever
  333. }
  334. }
  335. f.l.WithField("sendRecvError", f.sendRecvErrorConfig.String()).
  336. Info("Loaded send_recv_error config")
  337. }
  338. }
  339. func (f *Interface) reloadMisc(c *config.C) {
  340. if c.HasChanged("counters.try_promote") {
  341. n := c.GetUint32("counters.try_promote", defaultPromoteEvery)
  342. f.tryPromoteEvery.Store(n)
  343. f.l.Info("counters.try_promote has changed")
  344. }
  345. if c.HasChanged("counters.requery_every_packets") {
  346. n := c.GetUint32("counters.requery_every_packets", defaultReQueryEvery)
  347. f.reQueryEvery.Store(n)
  348. f.l.Info("counters.requery_every_packets has changed")
  349. }
  350. if c.HasChanged("timers.requery_wait_duration") {
  351. n := c.GetDuration("timers.requery_wait_duration", defaultReQueryWait)
  352. f.reQueryWait.Store(int64(n))
  353. f.l.Info("timers.requery_wait_duration has changed")
  354. }
  355. }
  356. func (f *Interface) emitStats(ctx context.Context, i time.Duration) {
  357. ticker := time.NewTicker(i)
  358. defer ticker.Stop()
  359. udpStats := udp.NewUDPStatsEmitter(f.writers)
  360. certExpirationGauge := metrics.GetOrRegisterGauge("certificate.ttl_seconds", nil)
  361. certInitiatingVersion := metrics.GetOrRegisterGauge("certificate.initiating_version", nil)
  362. certMaxVersion := metrics.GetOrRegisterGauge("certificate.max_version", nil)
  363. for {
  364. select {
  365. case <-ctx.Done():
  366. return
  367. case <-ticker.C:
  368. f.firewall.EmitStats()
  369. f.handshakeManager.EmitStats()
  370. udpStats()
  371. certState := f.pki.getCertState()
  372. defaultCrt := certState.GetDefaultCertificate()
  373. certExpirationGauge.Update(int64(defaultCrt.NotAfter().Sub(time.Now()) / time.Second))
  374. certInitiatingVersion.Update(int64(defaultCrt.Version()))
  375. // Report the max certificate version we are capable of using
  376. if certState.v2Cert != nil {
  377. certMaxVersion.Update(int64(certState.v2Cert.Version()))
  378. } else {
  379. certMaxVersion.Update(int64(certState.v1Cert.Version()))
  380. }
  381. }
  382. }
  383. }
  384. func (f *Interface) GetHostInfo(vpnIp netip.Addr) *HostInfo {
  385. return f.hostMap.QueryVpnAddr(vpnIp)
  386. }
  387. func (f *Interface) GetCertState() *CertState {
  388. return f.pki.getCertState()
  389. }
  390. func (f *Interface) Close() error {
  391. f.closed.Store(true)
  392. // Release the udp readers
  393. for _, u := range f.writers {
  394. err := u.Close()
  395. if err != nil {
  396. f.l.WithError(err).Error("Error while closing udp socket")
  397. }
  398. }
  399. // Release the tun readers
  400. for _, u := range f.readers {
  401. err := u.Close()
  402. if err != nil {
  403. f.l.WithError(err).Error("Error while closing tun device")
  404. }
  405. }
  406. return nil
  407. }