interface.go 15 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542
  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. DnsServer *dnsServer
  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. dnsServer *dnsServer
  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. ctx context.Context
  77. writers []udp.Conn
  78. readers []io.ReadWriteCloser
  79. wg sync.WaitGroup
  80. // fatalErr holds the first unexpected reader error that caused shutdown.
  81. // nil means "no fatal error" (yet)
  82. fatalErr atomic.Pointer[error]
  83. // triggerShutdown is a function that will be run exactly once, when onFatal swaps something non-nil into fatalErr
  84. triggerShutdown func()
  85. metricHandshakes metrics.Histogram
  86. messageMetrics *MessageMetrics
  87. cachedPacketMetrics *cachedPacketMetrics
  88. l *logrus.Logger
  89. }
  90. type EncWriter interface {
  91. SendVia(via *HostInfo,
  92. relay *Relay,
  93. ad,
  94. nb,
  95. out []byte,
  96. nocopy bool,
  97. )
  98. SendMessageToVpnAddr(t header.MessageType, st header.MessageSubType, vpnAddr netip.Addr, p, nb, out []byte)
  99. SendMessageToHostInfo(t header.MessageType, st header.MessageSubType, hostinfo *HostInfo, p, nb, out []byte)
  100. Handshake(vpnAddr netip.Addr)
  101. GetHostInfo(vpnAddr netip.Addr) *HostInfo
  102. GetCertState() *CertState
  103. }
  104. type recvErrorConfig uint8
  105. const (
  106. recvErrorAlways recvErrorConfig = iota
  107. recvErrorNever
  108. recvErrorPrivate
  109. )
  110. func (s recvErrorConfig) ShouldRecvError(endpoint netip.AddrPort) bool {
  111. switch s {
  112. case recvErrorPrivate:
  113. return endpoint.Addr().IsPrivate()
  114. case recvErrorAlways:
  115. return true
  116. case recvErrorNever:
  117. return false
  118. default:
  119. panic(fmt.Errorf("invalid recvErrorConfig value: %d", s))
  120. }
  121. }
  122. func (s recvErrorConfig) String() string {
  123. switch s {
  124. case recvErrorAlways:
  125. return "always"
  126. case recvErrorNever:
  127. return "never"
  128. case recvErrorPrivate:
  129. return "private"
  130. default:
  131. return fmt.Sprintf("invalid(%d)", s)
  132. }
  133. }
  134. func NewInterface(ctx context.Context, c *InterfaceConfig) (*Interface, error) {
  135. if c.Outside == nil {
  136. return nil, errors.New("no outside connection")
  137. }
  138. if c.Inside == nil {
  139. return nil, errors.New("no inside interface (tun)")
  140. }
  141. if c.pki == nil {
  142. return nil, errors.New("no certificate state")
  143. }
  144. if c.Firewall == nil {
  145. return nil, errors.New("no firewall rules")
  146. }
  147. if c.connectionManager == nil {
  148. return nil, errors.New("no connection manager")
  149. }
  150. cs := c.pki.getCertState()
  151. ifce := &Interface{
  152. ctx: ctx,
  153. pki: c.pki,
  154. hostMap: c.HostMap,
  155. outside: c.Outside,
  156. inside: c.Inside,
  157. firewall: c.Firewall,
  158. dnsServer: c.DnsServer,
  159. handshakeManager: c.HandshakeManager,
  160. createTime: time.Now(),
  161. lightHouse: c.lightHouse,
  162. dropLocalBroadcast: c.DropLocalBroadcast,
  163. dropMulticast: c.DropMulticast,
  164. routines: c.routines,
  165. version: c.version,
  166. writers: make([]udp.Conn, c.routines),
  167. readers: make([]io.ReadWriteCloser, c.routines),
  168. myVpnNetworks: cs.myVpnNetworks,
  169. myVpnNetworksTable: cs.myVpnNetworksTable,
  170. myVpnAddrs: cs.myVpnAddrs,
  171. myVpnAddrsTable: cs.myVpnAddrsTable,
  172. myBroadcastAddrsTable: cs.myVpnBroadcastAddrsTable,
  173. relayManager: c.relayManager,
  174. connectionManager: c.connectionManager,
  175. conntrackCacheTimeout: c.ConntrackCacheTimeout,
  176. metricHandshakes: metrics.GetOrRegisterHistogram("handshakes", nil, metrics.NewExpDecaySample(1028, 0.015)),
  177. messageMetrics: c.MessageMetrics,
  178. cachedPacketMetrics: &cachedPacketMetrics{
  179. sent: metrics.GetOrRegisterCounter("hostinfo.cached_packets.sent", nil),
  180. dropped: metrics.GetOrRegisterCounter("hostinfo.cached_packets.dropped", nil),
  181. },
  182. l: c.l,
  183. }
  184. ifce.tryPromoteEvery.Store(c.tryPromoteEvery)
  185. ifce.reQueryEvery.Store(c.reQueryEvery)
  186. ifce.reQueryWait.Store(int64(c.reQueryWait))
  187. ifce.connectionManager.intf = ifce
  188. return ifce, nil
  189. }
  190. // activate creates the interface on the host. After the interface is created, any
  191. // other services that want to bind listeners to its IP may do so successfully. However,
  192. // the interface isn't going to process anything until run() is called.
  193. func (f *Interface) activate() error {
  194. // actually turn on tun dev
  195. addr, err := f.outside.LocalAddr()
  196. if err != nil {
  197. f.l.WithError(err).Error("Failed to get udp listen address")
  198. }
  199. f.l.WithField("interface", f.inside.Name()).WithField("networks", f.myVpnNetworks).
  200. WithField("build", f.version).WithField("udpAddr", addr).
  201. WithField("boringcrypto", boringEnabled()).
  202. Info("Nebula interface is active")
  203. if f.routines > 1 {
  204. if !f.inside.SupportsMultiqueue() || !f.outside.SupportsMultipleReaders() {
  205. f.routines = 1
  206. f.l.Warn("routines is not supported on this platform, falling back to a single routine")
  207. }
  208. }
  209. metrics.GetOrRegisterGauge("routines", nil).Update(int64(f.routines))
  210. // Prepare n tun queues
  211. var reader io.ReadWriteCloser = f.inside
  212. for i := 0; i < f.routines; i++ {
  213. if i > 0 {
  214. reader, err = f.inside.NewMultiQueueReader()
  215. if err != nil {
  216. return err
  217. }
  218. }
  219. f.readers[i] = reader
  220. }
  221. f.wg.Add(1) // for us to wait on Close() to return
  222. if err = f.inside.Activate(); err != nil {
  223. f.wg.Done()
  224. f.inside.Close()
  225. return err
  226. }
  227. return nil
  228. }
  229. func (f *Interface) run() (func() error, error) {
  230. // Launch n queues to read packets from udp
  231. for i := 0; i < f.routines; i++ {
  232. f.wg.Go(func() {
  233. f.listenOut(i)
  234. })
  235. }
  236. // Launch n queues to read packets from tun dev
  237. for i := 0; i < f.routines; i++ {
  238. f.wg.Go(func() {
  239. f.listenIn(f.readers[i], i)
  240. })
  241. }
  242. return func() error {
  243. f.wg.Wait()
  244. if e := f.fatalErr.Load(); e != nil {
  245. return *e
  246. }
  247. return nil
  248. }, nil
  249. }
  250. // onFatal stores the first fatal reader error, and calls triggerShutdown if it was the first one
  251. func (f *Interface) onFatal(err error) {
  252. swapped := f.fatalErr.CompareAndSwap(nil, &err)
  253. if !swapped {
  254. return
  255. }
  256. if f.triggerShutdown != nil {
  257. f.triggerShutdown()
  258. }
  259. }
  260. func (f *Interface) listenOut(i int) {
  261. var li udp.Conn
  262. if i > 0 {
  263. li = f.writers[i]
  264. } else {
  265. li = f.outside
  266. }
  267. ctCache := firewall.NewConntrackCacheTicker(f.ctx, f.conntrackCacheTimeout)
  268. lhh := f.lightHouse.NewRequestHandler()
  269. plaintext := make([]byte, udp.MTU)
  270. h := &header.H{}
  271. fwPacket := &firewall.Packet{}
  272. nb := make([]byte, 12, 12)
  273. err := li.ListenOut(func(fromUdpAddr netip.AddrPort, payload []byte) {
  274. f.readOutsidePackets(ViaSender{UdpAddr: fromUdpAddr}, plaintext[:0], payload, h, fwPacket, lhh, nb, i, ctCache.Get(f.l))
  275. })
  276. if err != nil && !f.closed.Load() {
  277. f.l.WithError(err).Error("Error while reading inbound packet, closing")
  278. f.onFatal(err)
  279. }
  280. f.l.Debugf("underlay reader %v is done", i)
  281. }
  282. func (f *Interface) listenIn(reader io.ReadWriteCloser, i int) {
  283. packet := make([]byte, mtu)
  284. out := make([]byte, mtu)
  285. fwPacket := &firewall.Packet{}
  286. nb := make([]byte, 12, 12)
  287. conntrackCache := firewall.NewConntrackCacheTicker(f.ctx, f.conntrackCacheTimeout)
  288. for {
  289. n, err := reader.Read(packet)
  290. if err != nil {
  291. if !f.closed.Load() {
  292. f.l.WithError(err).WithField("reader", i).Error("Error while reading outbound packet, closing")
  293. f.onFatal(err)
  294. }
  295. break
  296. }
  297. f.consumeInsidePacket(packet[:n], fwPacket, nb, out, i, conntrackCache.Get(f.l))
  298. }
  299. f.l.Debugf("overlay reader %v is done", i)
  300. }
  301. func (f *Interface) RegisterConfigChangeCallbacks(c *config.C) {
  302. c.RegisterReloadCallback(f.reloadFirewall)
  303. c.RegisterReloadCallback(f.reloadSendRecvError)
  304. c.RegisterReloadCallback(f.reloadAcceptRecvError)
  305. c.RegisterReloadCallback(f.reloadDisconnectInvalid)
  306. c.RegisterReloadCallback(f.reloadMisc)
  307. for _, udpConn := range f.writers {
  308. c.RegisterReloadCallback(udpConn.ReloadConfig)
  309. }
  310. }
  311. func (f *Interface) reloadDisconnectInvalid(c *config.C) {
  312. initial := c.InitialLoad()
  313. if initial || c.HasChanged("pki.disconnect_invalid") {
  314. f.disconnectInvalid.Store(c.GetBool("pki.disconnect_invalid", true))
  315. if !initial {
  316. f.l.Infof("pki.disconnect_invalid changed to %v", f.disconnectInvalid.Load())
  317. }
  318. }
  319. }
  320. func (f *Interface) reloadFirewall(c *config.C) {
  321. //TODO: need to trigger/detect if the certificate changed too
  322. if c.HasChanged("firewall") == false {
  323. f.l.Debug("No firewall config change detected")
  324. return
  325. }
  326. fw, err := NewFirewallFromConfig(f.l, f.pki.getCertState(), c)
  327. if err != nil {
  328. f.l.WithError(err).Error("Error while creating firewall during reload")
  329. return
  330. }
  331. oldFw := f.firewall
  332. conntrack := oldFw.Conntrack
  333. conntrack.Lock()
  334. defer conntrack.Unlock()
  335. fw.rulesVersion = oldFw.rulesVersion + 1
  336. // If rulesVersion is back to zero, we have wrapped all the way around. Be
  337. // safe and just reset conntrack in this case.
  338. if fw.rulesVersion == 0 {
  339. f.l.WithField("firewallHashes", fw.GetRuleHashes()).
  340. WithField("oldFirewallHashes", oldFw.GetRuleHashes()).
  341. WithField("rulesVersion", fw.rulesVersion).
  342. Warn("firewall rulesVersion has overflowed, resetting conntrack")
  343. } else {
  344. fw.Conntrack = conntrack
  345. }
  346. f.firewall = fw
  347. oldFw.Destroy()
  348. f.l.WithField("firewallHashes", fw.GetRuleHashes()).
  349. WithField("oldFirewallHashes", oldFw.GetRuleHashes()).
  350. WithField("rulesVersion", fw.rulesVersion).
  351. Info("New firewall has been installed")
  352. }
  353. func (f *Interface) reloadSendRecvError(c *config.C) {
  354. if c.InitialLoad() || c.HasChanged("listen.send_recv_error") {
  355. stringValue := c.GetString("listen.send_recv_error", "always")
  356. switch stringValue {
  357. case "always":
  358. f.sendRecvErrorConfig = recvErrorAlways
  359. case "never":
  360. f.sendRecvErrorConfig = recvErrorNever
  361. case "private":
  362. f.sendRecvErrorConfig = recvErrorPrivate
  363. default:
  364. if c.GetBool("listen.send_recv_error", true) {
  365. f.sendRecvErrorConfig = recvErrorAlways
  366. } else {
  367. f.sendRecvErrorConfig = recvErrorNever
  368. }
  369. }
  370. f.l.WithField("sendRecvError", f.sendRecvErrorConfig.String()).
  371. Info("Loaded send_recv_error config")
  372. }
  373. }
  374. func (f *Interface) reloadAcceptRecvError(c *config.C) {
  375. if c.InitialLoad() || c.HasChanged("listen.accept_recv_error") {
  376. stringValue := c.GetString("listen.accept_recv_error", "always")
  377. switch stringValue {
  378. case "always":
  379. f.acceptRecvErrorConfig = recvErrorAlways
  380. case "never":
  381. f.acceptRecvErrorConfig = recvErrorNever
  382. case "private":
  383. f.acceptRecvErrorConfig = recvErrorPrivate
  384. default:
  385. if c.GetBool("listen.accept_recv_error", true) {
  386. f.acceptRecvErrorConfig = recvErrorAlways
  387. } else {
  388. f.acceptRecvErrorConfig = recvErrorNever
  389. }
  390. }
  391. f.l.WithField("acceptRecvError", f.acceptRecvErrorConfig.String()).
  392. Info("Loaded accept_recv_error config")
  393. }
  394. }
  395. func (f *Interface) reloadMisc(c *config.C) {
  396. if c.HasChanged("counters.try_promote") {
  397. n := c.GetUint32("counters.try_promote", defaultPromoteEvery)
  398. f.tryPromoteEvery.Store(n)
  399. f.l.Info("counters.try_promote has changed")
  400. }
  401. if c.HasChanged("counters.requery_every_packets") {
  402. n := c.GetUint32("counters.requery_every_packets", defaultReQueryEvery)
  403. f.reQueryEvery.Store(n)
  404. f.l.Info("counters.requery_every_packets has changed")
  405. }
  406. if c.HasChanged("timers.requery_wait_duration") {
  407. n := c.GetDuration("timers.requery_wait_duration", defaultReQueryWait)
  408. f.reQueryWait.Store(int64(n))
  409. f.l.Info("timers.requery_wait_duration has changed")
  410. }
  411. }
  412. func (f *Interface) emitStats(ctx context.Context, i time.Duration) {
  413. ticker := time.NewTicker(i)
  414. defer ticker.Stop()
  415. udpStats := udp.NewUDPStatsEmitter(f.writers)
  416. certExpirationGauge := metrics.GetOrRegisterGauge("certificate.ttl_seconds", nil)
  417. certInitiatingVersion := metrics.GetOrRegisterGauge("certificate.initiating_version", nil)
  418. certMaxVersion := metrics.GetOrRegisterGauge("certificate.max_version", nil)
  419. for {
  420. select {
  421. case <-ctx.Done():
  422. return
  423. case <-ticker.C:
  424. f.firewall.EmitStats()
  425. f.handshakeManager.EmitStats()
  426. udpStats()
  427. certState := f.pki.getCertState()
  428. defaultCrt := certState.GetDefaultCertificate()
  429. certExpirationGauge.Update(int64(defaultCrt.NotAfter().Sub(time.Now()) / time.Second))
  430. certInitiatingVersion.Update(int64(defaultCrt.Version()))
  431. // Report the max certificate version we are capable of using
  432. if certState.v2Cert != nil {
  433. certMaxVersion.Update(int64(certState.v2Cert.Version()))
  434. } else {
  435. certMaxVersion.Update(int64(certState.v1Cert.Version()))
  436. }
  437. }
  438. }
  439. }
  440. func (f *Interface) GetHostInfo(vpnIp netip.Addr) *HostInfo {
  441. return f.hostMap.QueryVpnAddr(vpnIp)
  442. }
  443. func (f *Interface) GetCertState() *CertState {
  444. return f.pki.getCertState()
  445. }
  446. func (f *Interface) Close() error {
  447. var errs []error
  448. f.closed.Store(true)
  449. // Release the udp readers
  450. for i, u := range f.writers {
  451. err := u.Close()
  452. if err != nil {
  453. f.l.WithError(err).WithField("writer", i).Error("Error while closing udp socket")
  454. errs = append(errs, err)
  455. }
  456. }
  457. // Release the tun device (closing the tun also closes all readers)
  458. closeErr := f.inside.Close()
  459. if closeErr != nil {
  460. errs = append(errs, closeErr)
  461. }
  462. f.wg.Done()
  463. return errors.Join(errs...)
  464. }