interface.go 14 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518
  1. package nebula
  2. import (
  3. "context"
  4. "errors"
  5. "fmt"
  6. "io"
  7. "net/netip"
  8. "sync"
  9. "sync/atomic"
  10. "time"
  11. "github.com/gaissmai/bart"
  12. "github.com/rcrowley/go-metrics"
  13. "github.com/sirupsen/logrus"
  14. "github.com/slackhq/nebula/config"
  15. "github.com/slackhq/nebula/firewall"
  16. "github.com/slackhq/nebula/header"
  17. "github.com/slackhq/nebula/overlay"
  18. "github.com/slackhq/nebula/udp"
  19. )
  20. const mtu = 9001
  21. type InterfaceConfig struct {
  22. HostMap *HostMap
  23. Outside udp.Conn
  24. Inside overlay.Device
  25. pki *PKI
  26. Cipher string
  27. Firewall *Firewall
  28. ServeDns bool
  29. HandshakeManager *HandshakeManager
  30. lightHouse *LightHouse
  31. connectionManager *connectionManager
  32. DropLocalBroadcast bool
  33. DropMulticast bool
  34. routines int
  35. MessageMetrics *MessageMetrics
  36. version string
  37. relayManager *relayManager
  38. punchy *Punchy
  39. tryPromoteEvery uint32
  40. reQueryEvery uint32
  41. reQueryWait time.Duration
  42. ConntrackCacheTimeout time.Duration
  43. l *logrus.Logger
  44. }
  45. type Interface struct {
  46. hostMap *HostMap
  47. outside udp.Conn
  48. inside overlay.Device
  49. pki *PKI
  50. firewall *Firewall
  51. connectionManager *connectionManager
  52. handshakeManager *HandshakeManager
  53. serveDns bool
  54. createTime time.Time
  55. lightHouse *LightHouse
  56. myBroadcastAddrsTable *bart.Lite
  57. myVpnAddrs []netip.Addr // A list of addresses assigned to us via our certificate
  58. myVpnAddrsTable *bart.Lite
  59. myVpnNetworks []netip.Prefix // A list of networks assigned to us via our certificate
  60. myVpnNetworksTable *bart.Lite
  61. dropLocalBroadcast bool
  62. dropMulticast bool
  63. routines int
  64. disconnectInvalid atomic.Bool
  65. closed atomic.Bool
  66. relayManager *relayManager
  67. tryPromoteEvery atomic.Uint32
  68. reQueryEvery atomic.Uint32
  69. reQueryWait atomic.Int64
  70. sendRecvErrorConfig recvErrorConfig
  71. acceptRecvErrorConfig recvErrorConfig
  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 recvErrorConfig uint8
  99. const (
  100. recvErrorAlways recvErrorConfig = iota
  101. recvErrorNever
  102. recvErrorPrivate
  103. )
  104. func (s recvErrorConfig) ShouldRecvError(endpoint netip.AddrPort) bool {
  105. switch s {
  106. case recvErrorPrivate:
  107. return endpoint.Addr().IsPrivate()
  108. case recvErrorAlways:
  109. return true
  110. case recvErrorNever:
  111. return false
  112. default:
  113. panic(fmt.Errorf("invalid recvErrorConfig value: %d", s))
  114. }
  115. }
  116. func (s recvErrorConfig) String() string {
  117. switch s {
  118. case recvErrorAlways:
  119. return "always"
  120. case recvErrorNever:
  121. return "never"
  122. case recvErrorPrivate:
  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. if f.routines > 1 {
  197. if !f.inside.SupportsMultiqueue() || !f.outside.SupportsMultipleReaders() {
  198. f.routines = 1
  199. f.l.Warn("routines is not supported on this platform, falling back to a single routine")
  200. }
  201. }
  202. metrics.GetOrRegisterGauge("routines", nil).Update(int64(f.routines))
  203. // Prepare n tun queues
  204. var reader io.ReadWriteCloser = f.inside
  205. for i := 0; i < f.routines; i++ {
  206. if i > 0 {
  207. reader, err = f.inside.NewMultiQueueReader()
  208. if err != nil {
  209. return err
  210. }
  211. }
  212. f.readers[i] = reader
  213. }
  214. if err = f.inside.Activate(); err != nil {
  215. f.inside.Close()
  216. return err
  217. }
  218. return nil
  219. }
  220. func (f *Interface) run() (func(), error) {
  221. // Launch n queues to read packets from udp
  222. for i := 0; i < f.routines; i++ {
  223. f.wg.Go(func() {
  224. f.listenOut(i)
  225. })
  226. }
  227. // Launch n queues to read packets from tun dev
  228. for i := 0; i < f.routines; i++ {
  229. f.wg.Go(func() {
  230. f.listenIn(f.readers[i], i)
  231. })
  232. }
  233. return f.wg.Wait, nil
  234. }
  235. func (f *Interface) listenOut(i int) {
  236. var li udp.Conn
  237. if i > 0 {
  238. li = f.writers[i]
  239. } else {
  240. li = f.outside
  241. }
  242. ctCache := firewall.NewConntrackCacheTicker(f.conntrackCacheTimeout)
  243. lhh := f.lightHouse.NewRequestHandler()
  244. plaintext := make([]byte, udp.MTU)
  245. h := &header.H{}
  246. fwPacket := &firewall.Packet{}
  247. nb := make([]byte, 12, 12)
  248. err := li.ListenOut(func(fromUdpAddr netip.AddrPort, payload []byte) {
  249. f.readOutsidePackets(ViaSender{UdpAddr: fromUdpAddr}, plaintext[:0], payload, h, fwPacket, lhh, nb, i, ctCache.Get(f.l))
  250. })
  251. if err != nil && !f.closed.Load() {
  252. f.l.WithError(err).Error("Error while reading inbound packet, closing")
  253. //TODO: Trigger Control to close
  254. }
  255. f.l.Debugf("underlay reader %v is done", i)
  256. }
  257. func (f *Interface) listenIn(reader io.ReadWriteCloser, i int) {
  258. packet := make([]byte, mtu)
  259. out := make([]byte, mtu)
  260. fwPacket := &firewall.Packet{}
  261. nb := make([]byte, 12, 12)
  262. conntrackCache := firewall.NewConntrackCacheTicker(f.conntrackCacheTimeout)
  263. for {
  264. n, err := reader.Read(packet)
  265. if err != nil {
  266. if !f.closed.Load() {
  267. f.l.WithError(err).WithField("reader", i).Error("Error while reading outbound packet, closing")
  268. //TODO: Trigger Control to close
  269. }
  270. break
  271. }
  272. f.consumeInsidePacket(packet[:n], fwPacket, nb, out, i, conntrackCache.Get(f.l))
  273. }
  274. f.l.Debugf("overlay reader %v is done", i)
  275. }
  276. func (f *Interface) RegisterConfigChangeCallbacks(c *config.C) {
  277. c.RegisterReloadCallback(f.reloadFirewall)
  278. c.RegisterReloadCallback(f.reloadSendRecvError)
  279. c.RegisterReloadCallback(f.reloadAcceptRecvError)
  280. c.RegisterReloadCallback(f.reloadDisconnectInvalid)
  281. c.RegisterReloadCallback(f.reloadMisc)
  282. for _, udpConn := range f.writers {
  283. c.RegisterReloadCallback(udpConn.ReloadConfig)
  284. }
  285. }
  286. func (f *Interface) reloadDisconnectInvalid(c *config.C) {
  287. initial := c.InitialLoad()
  288. if initial || c.HasChanged("pki.disconnect_invalid") {
  289. f.disconnectInvalid.Store(c.GetBool("pki.disconnect_invalid", true))
  290. if !initial {
  291. f.l.Infof("pki.disconnect_invalid changed to %v", f.disconnectInvalid.Load())
  292. }
  293. }
  294. }
  295. func (f *Interface) reloadFirewall(c *config.C) {
  296. //TODO: need to trigger/detect if the certificate changed too
  297. if c.HasChanged("firewall") == false {
  298. f.l.Debug("No firewall config change detected")
  299. return
  300. }
  301. fw, err := NewFirewallFromConfig(f.l, f.pki.getCertState(), c)
  302. if err != nil {
  303. f.l.WithError(err).Error("Error while creating firewall during reload")
  304. return
  305. }
  306. oldFw := f.firewall
  307. conntrack := oldFw.Conntrack
  308. conntrack.Lock()
  309. defer conntrack.Unlock()
  310. fw.rulesVersion = oldFw.rulesVersion + 1
  311. // If rulesVersion is back to zero, we have wrapped all the way around. Be
  312. // safe and just reset conntrack in this case.
  313. if fw.rulesVersion == 0 {
  314. f.l.WithField("firewallHashes", fw.GetRuleHashes()).
  315. WithField("oldFirewallHashes", oldFw.GetRuleHashes()).
  316. WithField("rulesVersion", fw.rulesVersion).
  317. Warn("firewall rulesVersion has overflowed, resetting conntrack")
  318. } else {
  319. fw.Conntrack = conntrack
  320. }
  321. f.firewall = fw
  322. oldFw.Destroy()
  323. f.l.WithField("firewallHashes", fw.GetRuleHashes()).
  324. WithField("oldFirewallHashes", oldFw.GetRuleHashes()).
  325. WithField("rulesVersion", fw.rulesVersion).
  326. Info("New firewall has been installed")
  327. }
  328. func (f *Interface) reloadSendRecvError(c *config.C) {
  329. if c.InitialLoad() || c.HasChanged("listen.send_recv_error") {
  330. stringValue := c.GetString("listen.send_recv_error", "always")
  331. switch stringValue {
  332. case "always":
  333. f.sendRecvErrorConfig = recvErrorAlways
  334. case "never":
  335. f.sendRecvErrorConfig = recvErrorNever
  336. case "private":
  337. f.sendRecvErrorConfig = recvErrorPrivate
  338. default:
  339. if c.GetBool("listen.send_recv_error", true) {
  340. f.sendRecvErrorConfig = recvErrorAlways
  341. } else {
  342. f.sendRecvErrorConfig = recvErrorNever
  343. }
  344. }
  345. f.l.WithField("sendRecvError", f.sendRecvErrorConfig.String()).
  346. Info("Loaded send_recv_error config")
  347. }
  348. }
  349. func (f *Interface) reloadAcceptRecvError(c *config.C) {
  350. if c.InitialLoad() || c.HasChanged("listen.accept_recv_error") {
  351. stringValue := c.GetString("listen.accept_recv_error", "always")
  352. switch stringValue {
  353. case "always":
  354. f.acceptRecvErrorConfig = recvErrorAlways
  355. case "never":
  356. f.acceptRecvErrorConfig = recvErrorNever
  357. case "private":
  358. f.acceptRecvErrorConfig = recvErrorPrivate
  359. default:
  360. if c.GetBool("listen.accept_recv_error", true) {
  361. f.acceptRecvErrorConfig = recvErrorAlways
  362. } else {
  363. f.acceptRecvErrorConfig = recvErrorNever
  364. }
  365. }
  366. f.l.WithField("acceptRecvError", f.acceptRecvErrorConfig.String()).
  367. Info("Loaded accept_recv_error config")
  368. }
  369. }
  370. func (f *Interface) reloadMisc(c *config.C) {
  371. if c.HasChanged("counters.try_promote") {
  372. n := c.GetUint32("counters.try_promote", defaultPromoteEvery)
  373. f.tryPromoteEvery.Store(n)
  374. f.l.Info("counters.try_promote has changed")
  375. }
  376. if c.HasChanged("counters.requery_every_packets") {
  377. n := c.GetUint32("counters.requery_every_packets", defaultReQueryEvery)
  378. f.reQueryEvery.Store(n)
  379. f.l.Info("counters.requery_every_packets has changed")
  380. }
  381. if c.HasChanged("timers.requery_wait_duration") {
  382. n := c.GetDuration("timers.requery_wait_duration", defaultReQueryWait)
  383. f.reQueryWait.Store(int64(n))
  384. f.l.Info("timers.requery_wait_duration has changed")
  385. }
  386. }
  387. func (f *Interface) emitStats(ctx context.Context, i time.Duration) {
  388. ticker := time.NewTicker(i)
  389. defer ticker.Stop()
  390. udpStats := udp.NewUDPStatsEmitter(f.writers)
  391. certExpirationGauge := metrics.GetOrRegisterGauge("certificate.ttl_seconds", nil)
  392. certInitiatingVersion := metrics.GetOrRegisterGauge("certificate.initiating_version", nil)
  393. certMaxVersion := metrics.GetOrRegisterGauge("certificate.max_version", nil)
  394. for {
  395. select {
  396. case <-ctx.Done():
  397. return
  398. case <-ticker.C:
  399. f.firewall.EmitStats()
  400. f.handshakeManager.EmitStats()
  401. udpStats()
  402. certState := f.pki.getCertState()
  403. defaultCrt := certState.GetDefaultCertificate()
  404. certExpirationGauge.Update(int64(defaultCrt.NotAfter().Sub(time.Now()) / time.Second))
  405. certInitiatingVersion.Update(int64(defaultCrt.Version()))
  406. // Report the max certificate version we are capable of using
  407. if certState.v2Cert != nil {
  408. certMaxVersion.Update(int64(certState.v2Cert.Version()))
  409. } else {
  410. certMaxVersion.Update(int64(certState.v1Cert.Version()))
  411. }
  412. }
  413. }
  414. }
  415. func (f *Interface) GetHostInfo(vpnIp netip.Addr) *HostInfo {
  416. return f.hostMap.QueryVpnAddr(vpnIp)
  417. }
  418. func (f *Interface) GetCertState() *CertState {
  419. return f.pki.getCertState()
  420. }
  421. func (f *Interface) Close() error {
  422. f.closed.Store(true)
  423. // Release the udp readers
  424. for _, u := range f.writers {
  425. err := u.Close()
  426. if err != nil {
  427. f.l.WithError(err).Error("Error while closing udp socket")
  428. }
  429. }
  430. // Release the tun readers
  431. for i, r := range f.readers {
  432. if i == 0 {
  433. continue // f.readers[0] is f.inside, which we want to save for last, since it closes other stuff too
  434. }
  435. if err := r.Close(); err != nil {
  436. f.l.WithError(err).Error("Error while closing tun reader")
  437. }
  438. }
  439. // Release the tun device
  440. return f.inside.Close()
  441. }