interface.go 15 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565
  1. package nebula
  2. import (
  3. "context"
  4. "errors"
  5. "fmt"
  6. "io"
  7. "log/slog"
  8. "net/netip"
  9. "slices"
  10. "sync"
  11. "sync/atomic"
  12. "time"
  13. "github.com/gaissmai/bart"
  14. "github.com/rcrowley/go-metrics"
  15. "github.com/slackhq/nebula/cert"
  16. "github.com/slackhq/nebula/config"
  17. "github.com/slackhq/nebula/firewall"
  18. "github.com/slackhq/nebula/header"
  19. "github.com/slackhq/nebula/overlay"
  20. "github.com/slackhq/nebula/udp"
  21. )
  22. const mtu = 9001
  23. type InterfaceConfig struct {
  24. HostMap *HostMap
  25. Outside udp.Conn
  26. Inside overlay.Device
  27. pki *PKI
  28. Cipher string
  29. Firewall *Firewall
  30. DnsServer *dnsServer
  31. HandshakeManager *HandshakeManager
  32. lightHouse *LightHouse
  33. connectionManager *connectionManager
  34. DropLocalBroadcast bool
  35. DropMulticast bool
  36. routines int
  37. MessageMetrics *MessageMetrics
  38. version string
  39. relayManager *relayManager
  40. punchy *Punchy
  41. tryPromoteEvery uint32
  42. reQueryEvery uint32
  43. reQueryWait time.Duration
  44. ConntrackCacheTimeout time.Duration
  45. l *slog.Logger
  46. }
  47. type Interface struct {
  48. hostMap *HostMap
  49. outside udp.Conn
  50. inside overlay.Device
  51. pki *PKI
  52. firewall *Firewall
  53. connectionManager *connectionManager
  54. handshakeManager *HandshakeManager
  55. dnsServer *dnsServer
  56. createTime time.Time
  57. lightHouse *LightHouse
  58. myBroadcastAddrsTable *bart.Lite
  59. myVpnAddrs []netip.Addr // A list of addresses assigned to us via our certificate
  60. myVpnAddrsTable *bart.Lite
  61. myVpnNetworks []netip.Prefix // A list of networks assigned to us via our certificate
  62. myVpnNetworksTable *bart.Lite
  63. dropLocalBroadcast bool
  64. dropMulticast bool
  65. routines int
  66. disconnectInvalid atomic.Bool
  67. closed atomic.Bool
  68. relayManager *relayManager
  69. tryPromoteEvery atomic.Uint32
  70. reQueryEvery atomic.Uint32
  71. reQueryWait atomic.Int64
  72. sendRecvErrorConfig recvErrorConfig
  73. acceptRecvErrorConfig recvErrorConfig
  74. // rebindCount is used to decide if an active tunnel should trigger a punch notification through a lighthouse
  75. rebindCount int8
  76. version string
  77. conntrackCacheTimeout time.Duration
  78. ctx context.Context
  79. writers []udp.Conn
  80. readers []io.ReadWriteCloser
  81. wg sync.WaitGroup
  82. // fatalErr holds the first unexpected reader error that caused shutdown.
  83. // nil means "no fatal error" (yet)
  84. fatalErr atomic.Pointer[error]
  85. // triggerShutdown is a function that will be run exactly once, when onFatal swaps something non-nil into fatalErr
  86. triggerShutdown func()
  87. metricHandshakes metrics.Histogram
  88. messageMetrics *MessageMetrics
  89. cachedPacketMetrics *cachedPacketMetrics
  90. l *slog.Logger
  91. }
  92. type EncWriter interface {
  93. SendVia(via *HostInfo,
  94. relay *Relay,
  95. ad,
  96. nb,
  97. out []byte,
  98. nocopy bool,
  99. )
  100. SendMessageToVpnAddr(t header.MessageType, st header.MessageSubType, vpnAddr netip.Addr, p, nb, out []byte)
  101. SendMessageToHostInfo(t header.MessageType, st header.MessageSubType, hostinfo *HostInfo, p, nb, out []byte)
  102. Handshake(vpnAddr netip.Addr)
  103. GetHostInfo(vpnAddr netip.Addr) *HostInfo
  104. GetCertState() *CertState
  105. }
  106. type recvErrorConfig uint8
  107. const (
  108. recvErrorAlways recvErrorConfig = iota
  109. recvErrorNever
  110. recvErrorPrivate
  111. )
  112. func (s recvErrorConfig) ShouldRecvError(endpoint netip.AddrPort) bool {
  113. switch s {
  114. case recvErrorPrivate:
  115. return endpoint.Addr().IsPrivate()
  116. case recvErrorAlways:
  117. return true
  118. case recvErrorNever:
  119. return false
  120. default:
  121. panic(fmt.Errorf("invalid recvErrorConfig value: %d", s))
  122. }
  123. }
  124. func (s recvErrorConfig) String() string {
  125. switch s {
  126. case recvErrorAlways:
  127. return "always"
  128. case recvErrorNever:
  129. return "never"
  130. case recvErrorPrivate:
  131. return "private"
  132. default:
  133. return fmt.Sprintf("invalid(%d)", s)
  134. }
  135. }
  136. func NewInterface(ctx context.Context, c *InterfaceConfig) (*Interface, error) {
  137. if c.Outside == nil {
  138. return nil, errors.New("no outside connection")
  139. }
  140. if c.Inside == nil {
  141. return nil, errors.New("no inside interface (tun)")
  142. }
  143. if c.pki == nil {
  144. return nil, errors.New("no certificate state")
  145. }
  146. if c.Firewall == nil {
  147. return nil, errors.New("no firewall rules")
  148. }
  149. if c.connectionManager == nil {
  150. return nil, errors.New("no connection manager")
  151. }
  152. cs := c.pki.getCertState()
  153. ifce := &Interface{
  154. ctx: ctx,
  155. pki: c.pki,
  156. hostMap: c.HostMap,
  157. outside: c.Outside,
  158. inside: c.Inside,
  159. firewall: c.Firewall,
  160. dnsServer: c.DnsServer,
  161. handshakeManager: c.HandshakeManager,
  162. createTime: time.Now(),
  163. lightHouse: c.lightHouse,
  164. dropLocalBroadcast: c.DropLocalBroadcast,
  165. dropMulticast: c.DropMulticast,
  166. routines: c.routines,
  167. version: c.version,
  168. writers: make([]udp.Conn, c.routines),
  169. readers: make([]io.ReadWriteCloser, c.routines),
  170. myVpnNetworks: cs.myVpnNetworks,
  171. myVpnNetworksTable: cs.myVpnNetworksTable,
  172. myVpnAddrs: cs.myVpnAddrs,
  173. myVpnAddrsTable: cs.myVpnAddrsTable,
  174. myBroadcastAddrsTable: cs.myVpnBroadcastAddrsTable,
  175. relayManager: c.relayManager,
  176. connectionManager: c.connectionManager,
  177. conntrackCacheTimeout: c.ConntrackCacheTimeout,
  178. metricHandshakes: metrics.GetOrRegisterHistogram("handshakes", nil, metrics.NewExpDecaySample(1028, 0.015)),
  179. messageMetrics: c.MessageMetrics,
  180. cachedPacketMetrics: &cachedPacketMetrics{
  181. sent: metrics.GetOrRegisterCounter("hostinfo.cached_packets.sent", nil),
  182. dropped: metrics.GetOrRegisterCounter("hostinfo.cached_packets.dropped", nil),
  183. },
  184. l: c.l,
  185. }
  186. ifce.tryPromoteEvery.Store(c.tryPromoteEvery)
  187. ifce.reQueryEvery.Store(c.reQueryEvery)
  188. ifce.reQueryWait.Store(int64(c.reQueryWait))
  189. ifce.connectionManager.intf = ifce
  190. return ifce, nil
  191. }
  192. // activate creates the interface on the host. After the interface is created, any
  193. // other services that want to bind listeners to its IP may do so successfully. However,
  194. // the interface isn't going to process anything until run() is called.
  195. func (f *Interface) activate() error {
  196. // actually turn on tun dev
  197. addr, err := f.outside.LocalAddr()
  198. if err != nil {
  199. f.l.Error("Failed to get udp listen address", "error", err)
  200. }
  201. f.l.Info("Nebula interface is active",
  202. "interface", f.inside.Name(),
  203. "networks", f.myVpnNetworks,
  204. "build", f.version,
  205. "udpAddr", addr,
  206. "boringcrypto", boringEnabled(),
  207. )
  208. if f.routines > 1 {
  209. if !f.inside.SupportsMultiqueue() || !f.outside.SupportsMultipleReaders() {
  210. f.routines = 1
  211. f.l.Warn("routines is not supported on this platform, falling back to a single routine")
  212. }
  213. }
  214. metrics.GetOrRegisterGauge("routines", nil).Update(int64(f.routines))
  215. // Prepare n tun queues
  216. var reader io.ReadWriteCloser = f.inside
  217. for i := 0; i < f.routines; i++ {
  218. if i > 0 {
  219. reader, err = f.inside.NewMultiQueueReader()
  220. if err != nil {
  221. return err
  222. }
  223. }
  224. f.readers[i] = reader
  225. }
  226. f.wg.Add(1) // for us to wait on Close() to return
  227. if err = f.inside.Activate(); err != nil {
  228. f.wg.Done()
  229. f.inside.Close()
  230. return err
  231. }
  232. return nil
  233. }
  234. func (f *Interface) run() (func() error, error) {
  235. // Launch n queues to read packets from udp
  236. for i := 0; i < f.routines; i++ {
  237. f.wg.Go(func() {
  238. f.listenOut(i)
  239. })
  240. }
  241. // Launch n queues to read packets from tun dev
  242. for i := 0; i < f.routines; i++ {
  243. f.wg.Go(func() {
  244. f.listenIn(f.readers[i], i)
  245. })
  246. }
  247. return func() error {
  248. f.wg.Wait()
  249. if e := f.fatalErr.Load(); e != nil {
  250. return *e
  251. }
  252. return nil
  253. }, nil
  254. }
  255. // onFatal stores the first fatal reader error, and calls triggerShutdown if it was the first one
  256. func (f *Interface) onFatal(err error) {
  257. swapped := f.fatalErr.CompareAndSwap(nil, &err)
  258. if !swapped {
  259. return
  260. }
  261. if f.triggerShutdown != nil {
  262. f.triggerShutdown()
  263. }
  264. }
  265. func (f *Interface) listenOut(i int) {
  266. var li udp.Conn
  267. if i > 0 {
  268. li = f.writers[i]
  269. } else {
  270. li = f.outside
  271. }
  272. ctCache := firewall.NewConntrackCacheTicker(f.ctx, f.l, f.conntrackCacheTimeout)
  273. lhh := f.lightHouse.NewRequestHandler()
  274. plaintext := make([]byte, udp.MTU)
  275. h := &header.H{}
  276. fwPacket := &firewall.Packet{}
  277. nb := make([]byte, 12, 12)
  278. err := li.ListenOut(func(fromUdpAddr netip.AddrPort, payload []byte) {
  279. f.readOutsidePackets(ViaSender{UdpAddr: fromUdpAddr}, plaintext[:0], payload, h, fwPacket, lhh, nb, i, ctCache.Get())
  280. })
  281. if err != nil && !f.closed.Load() {
  282. f.l.Error("Error while reading inbound packet, closing", "error", err)
  283. f.onFatal(err)
  284. }
  285. f.l.Debug("underlay reader is done", "reader", i)
  286. }
  287. func (f *Interface) listenIn(reader io.ReadWriteCloser, i int) {
  288. packet := make([]byte, mtu)
  289. out := make([]byte, mtu)
  290. fwPacket := &firewall.Packet{}
  291. nb := make([]byte, 12, 12)
  292. conntrackCache := firewall.NewConntrackCacheTicker(f.ctx, f.l, f.conntrackCacheTimeout)
  293. for {
  294. n, err := reader.Read(packet)
  295. if err != nil {
  296. if !f.closed.Load() {
  297. f.l.Error("Error while reading outbound packet, closing", "error", err, "reader", i)
  298. f.onFatal(err)
  299. }
  300. break
  301. }
  302. f.consumeInsidePacket(packet[:n], fwPacket, nb, out, i, conntrackCache.Get())
  303. }
  304. f.l.Debug("overlay reader is done", "reader", i)
  305. }
  306. func (f *Interface) RegisterConfigChangeCallbacks(c *config.C) {
  307. c.RegisterReloadCallback(f.reloadFirewall)
  308. c.RegisterReloadCallback(f.reloadSendRecvError)
  309. c.RegisterReloadCallback(f.reloadAcceptRecvError)
  310. c.RegisterReloadCallback(f.reloadDisconnectInvalid)
  311. c.RegisterReloadCallback(f.reloadMisc)
  312. for _, udpConn := range f.writers {
  313. c.RegisterReloadCallback(udpConn.ReloadConfig)
  314. }
  315. }
  316. func (f *Interface) reloadDisconnectInvalid(c *config.C) {
  317. initial := c.InitialLoad()
  318. if initial || c.HasChanged("pki.disconnect_invalid") {
  319. f.disconnectInvalid.Store(c.GetBool("pki.disconnect_invalid", true))
  320. if !initial {
  321. f.l.Info("pki.disconnect_invalid changed", "value", f.disconnectInvalid.Load())
  322. }
  323. }
  324. }
  325. func (f *Interface) reloadFirewall(c *config.C) {
  326. cs := f.pki.getCertState()
  327. curCert := cs.getCertificate(cert.Version2)
  328. if curCert == nil {
  329. curCert = cs.getCertificate(cert.Version1)
  330. }
  331. // The firewall builds its routableNetworks set from the certificate's UnsafeNetworks at construction.
  332. // Check to see if that set has changed, and if so, rebuild the firewall.
  333. certUnsafeChanged := curCert != nil && !slices.Equal(curCert.UnsafeNetworks(), f.firewall.unsafeNetworks)
  334. if !c.HasChanged("firewall") && !certUnsafeChanged {
  335. f.l.Debug("No firewall config change detected")
  336. return
  337. }
  338. fw, err := NewFirewallFromConfig(f.l, cs, c)
  339. if err != nil {
  340. f.l.Error("Error while creating firewall during reload", "error", err)
  341. return
  342. }
  343. oldFw := f.firewall
  344. conntrack := oldFw.Conntrack
  345. conntrack.Lock()
  346. defer conntrack.Unlock()
  347. fw.rulesVersion = oldFw.rulesVersion + 1
  348. // If rulesVersion is back to zero, we have wrapped all the way around. Be
  349. // safe and just reset conntrack in this case.
  350. if fw.rulesVersion == 0 {
  351. f.l.Warn("firewall rulesVersion has overflowed, resetting conntrack",
  352. "firewallHashes", fw.GetRuleHashes(),
  353. "oldFirewallHashes", oldFw.GetRuleHashes(),
  354. "rulesVersion", fw.rulesVersion,
  355. )
  356. } else {
  357. fw.Conntrack = conntrack
  358. }
  359. f.firewall = fw
  360. oldFw.Destroy()
  361. f.l.Info("New firewall has been installed",
  362. "firewallHashes", fw.GetRuleHashes(),
  363. "oldFirewallHashes", oldFw.GetRuleHashes(),
  364. "rulesVersion", fw.rulesVersion,
  365. )
  366. }
  367. func (f *Interface) reloadSendRecvError(c *config.C) {
  368. if c.InitialLoad() || c.HasChanged("listen.send_recv_error") {
  369. stringValue := c.GetString("listen.send_recv_error", "always")
  370. switch stringValue {
  371. case "always":
  372. f.sendRecvErrorConfig = recvErrorAlways
  373. case "never":
  374. f.sendRecvErrorConfig = recvErrorNever
  375. case "private":
  376. f.sendRecvErrorConfig = recvErrorPrivate
  377. default:
  378. if c.GetBool("listen.send_recv_error", true) {
  379. f.sendRecvErrorConfig = recvErrorAlways
  380. } else {
  381. f.sendRecvErrorConfig = recvErrorNever
  382. }
  383. }
  384. f.l.Info("Loaded send_recv_error config", "sendRecvError", f.sendRecvErrorConfig.String())
  385. }
  386. }
  387. func (f *Interface) reloadAcceptRecvError(c *config.C) {
  388. if c.InitialLoad() || c.HasChanged("listen.accept_recv_error") {
  389. stringValue := c.GetString("listen.accept_recv_error", "always")
  390. switch stringValue {
  391. case "always":
  392. f.acceptRecvErrorConfig = recvErrorAlways
  393. case "never":
  394. f.acceptRecvErrorConfig = recvErrorNever
  395. case "private":
  396. f.acceptRecvErrorConfig = recvErrorPrivate
  397. default:
  398. if c.GetBool("listen.accept_recv_error", true) {
  399. f.acceptRecvErrorConfig = recvErrorAlways
  400. } else {
  401. f.acceptRecvErrorConfig = recvErrorNever
  402. }
  403. }
  404. f.l.Info("Loaded accept_recv_error config", "acceptRecvError", f.acceptRecvErrorConfig.String())
  405. }
  406. }
  407. func (f *Interface) reloadMisc(c *config.C) {
  408. if c.HasChanged("counters.try_promote") {
  409. n := c.GetUint32("counters.try_promote", defaultPromoteEvery)
  410. f.tryPromoteEvery.Store(n)
  411. f.l.Info("counters.try_promote has changed")
  412. }
  413. if c.HasChanged("counters.requery_every_packets") {
  414. n := c.GetUint32("counters.requery_every_packets", defaultReQueryEvery)
  415. f.reQueryEvery.Store(n)
  416. f.l.Info("counters.requery_every_packets has changed")
  417. }
  418. if c.HasChanged("timers.requery_wait_duration") {
  419. n := c.GetDuration("timers.requery_wait_duration", defaultReQueryWait)
  420. f.reQueryWait.Store(int64(n))
  421. f.l.Info("timers.requery_wait_duration has changed")
  422. }
  423. }
  424. func (f *Interface) emitStats(ctx context.Context, i time.Duration) {
  425. ticker := time.NewTicker(i)
  426. defer ticker.Stop()
  427. udpStats := udp.NewUDPStatsEmitter(f.writers)
  428. certExpirationGauge := metrics.GetOrRegisterGauge("certificate.ttl_seconds", nil)
  429. certInitiatingVersion := metrics.GetOrRegisterGauge("certificate.initiating_version", nil)
  430. certMaxVersion := metrics.GetOrRegisterGauge("certificate.max_version", nil)
  431. emit := func() {
  432. f.firewall.EmitStats()
  433. f.handshakeManager.EmitStats()
  434. udpStats()
  435. certState := f.pki.getCertState()
  436. defaultCrt := certState.GetDefaultCertificate()
  437. certExpirationGauge.Update(int64(defaultCrt.NotAfter().Sub(time.Now()) / time.Second))
  438. certInitiatingVersion.Update(int64(defaultCrt.Version()))
  439. // Report the max certificate version we are capable of using
  440. if certState.v2Cert != nil {
  441. certMaxVersion.Update(int64(certState.v2Cert.Version()))
  442. } else {
  443. certMaxVersion.Update(int64(certState.v1Cert.Version()))
  444. }
  445. }
  446. // Prime gauges so a Prometheus scrape that lands before the first tick
  447. // sees real values instead of the zero defaults (issue #907).
  448. emit()
  449. for {
  450. select {
  451. case <-ctx.Done():
  452. return
  453. case <-ticker.C:
  454. emit()
  455. }
  456. }
  457. }
  458. func (f *Interface) GetHostInfo(vpnIp netip.Addr) *HostInfo {
  459. return f.hostMap.QueryVpnAddr(vpnIp)
  460. }
  461. func (f *Interface) GetCertState() *CertState {
  462. return f.pki.getCertState()
  463. }
  464. func (f *Interface) Close() error {
  465. var errs []error
  466. f.closed.Store(true)
  467. // Release the udp readers
  468. for i, u := range f.writers {
  469. err := u.Close()
  470. if err != nil {
  471. f.l.Error("Error while closing udp socket", "error", err, "writer", i)
  472. errs = append(errs, err)
  473. }
  474. }
  475. // Release the tun device (closing the tun also closes all readers)
  476. closeErr := f.inside.Close()
  477. if closeErr != nil {
  478. errs = append(errs, closeErr)
  479. }
  480. f.wg.Done()
  481. return errors.Join(errs...)
  482. }