interface.go 15 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550
  1. package nebula
  2. import (
  3. "context"
  4. "errors"
  5. "fmt"
  6. "io"
  7. "log/slog"
  8. "net/netip"
  9. "sync"
  10. "sync/atomic"
  11. "time"
  12. "github.com/gaissmai/bart"
  13. "github.com/rcrowley/go-metrics"
  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. pmtudManager *pmtudManager
  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 *slog.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. pmtudManager *pmtudManager
  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. pmtudManager: c.pmtudManager,
  178. conntrackCacheTimeout: c.ConntrackCacheTimeout,
  179. metricHandshakes: metrics.GetOrRegisterHistogram("handshakes", nil, metrics.NewExpDecaySample(1028, 0.015)),
  180. messageMetrics: c.MessageMetrics,
  181. cachedPacketMetrics: &cachedPacketMetrics{
  182. sent: metrics.GetOrRegisterCounter("hostinfo.cached_packets.sent", nil),
  183. dropped: metrics.GetOrRegisterCounter("hostinfo.cached_packets.dropped", nil),
  184. },
  185. l: c.l,
  186. }
  187. ifce.tryPromoteEvery.Store(c.tryPromoteEvery)
  188. ifce.reQueryEvery.Store(c.reQueryEvery)
  189. ifce.reQueryWait.Store(int64(c.reQueryWait))
  190. ifce.connectionManager.intf = ifce
  191. ifce.pmtudManager.intf = ifce
  192. return ifce, nil
  193. }
  194. // activate creates the interface on the host. After the interface is created, any
  195. // other services that want to bind listeners to its IP may do so successfully. However,
  196. // the interface isn't going to process anything until run() is called.
  197. func (f *Interface) activate() error {
  198. // actually turn on tun dev
  199. addr, err := f.outside.LocalAddr()
  200. if err != nil {
  201. f.l.Error("Failed to get udp listen address", "error", err)
  202. }
  203. f.l.Info("Nebula interface is active",
  204. "interface", f.inside.Name(),
  205. "networks", f.myVpnNetworks,
  206. "build", f.version,
  207. "udpAddr", addr,
  208. "boringcrypto", boringEnabled(),
  209. )
  210. if f.routines > 1 {
  211. if !f.inside.SupportsMultiqueue() || !f.outside.SupportsMultipleReaders() {
  212. f.routines = 1
  213. f.l.Warn("routines is not supported on this platform, falling back to a single routine")
  214. }
  215. }
  216. metrics.GetOrRegisterGauge("routines", nil).Update(int64(f.routines))
  217. // Prepare n tun queues
  218. var reader io.ReadWriteCloser = f.inside
  219. for i := 0; i < f.routines; i++ {
  220. if i > 0 {
  221. reader, err = f.inside.NewMultiQueueReader()
  222. if err != nil {
  223. return err
  224. }
  225. }
  226. f.readers[i] = reader
  227. }
  228. f.wg.Add(1) // for us to wait on Close() to return
  229. if err = f.inside.Activate(); err != nil {
  230. f.wg.Done()
  231. f.inside.Close()
  232. return err
  233. }
  234. return nil
  235. }
  236. func (f *Interface) run() (func() error, error) {
  237. // Launch n queues to read packets from udp
  238. for i := 0; i < f.routines; i++ {
  239. f.wg.Go(func() {
  240. f.listenOut(i)
  241. })
  242. }
  243. // Launch n queues to read packets from tun dev
  244. for i := 0; i < f.routines; i++ {
  245. f.wg.Go(func() {
  246. f.listenIn(f.readers[i], i)
  247. })
  248. }
  249. return func() error {
  250. f.wg.Wait()
  251. if e := f.fatalErr.Load(); e != nil {
  252. return *e
  253. }
  254. return nil
  255. }, nil
  256. }
  257. // onFatal stores the first fatal reader error, and calls triggerShutdown if it was the first one
  258. func (f *Interface) onFatal(err error) {
  259. swapped := f.fatalErr.CompareAndSwap(nil, &err)
  260. if !swapped {
  261. return
  262. }
  263. if f.triggerShutdown != nil {
  264. f.triggerShutdown()
  265. }
  266. }
  267. func (f *Interface) listenOut(i int) {
  268. var li udp.Conn
  269. if i > 0 {
  270. li = f.writers[i]
  271. } else {
  272. li = f.outside
  273. }
  274. ctCache := firewall.NewConntrackCacheTicker(f.ctx, f.l, f.conntrackCacheTimeout)
  275. lhh := f.lightHouse.NewRequestHandler()
  276. plaintext := make([]byte, udp.MTU)
  277. h := &header.H{}
  278. fwPacket := &firewall.Packet{}
  279. nb := make([]byte, 12, 12)
  280. err := li.ListenOut(func(fromUdpAddr netip.AddrPort, payload []byte) {
  281. f.readOutsidePackets(ViaSender{UdpAddr: fromUdpAddr}, plaintext[:0], payload, h, fwPacket, lhh, nb, i, ctCache.Get())
  282. })
  283. if err != nil && !f.closed.Load() {
  284. f.l.Error("Error while reading inbound packet, closing", "error", err)
  285. f.onFatal(err)
  286. }
  287. f.l.Debug("underlay reader is done", "reader", i)
  288. }
  289. func (f *Interface) listenIn(reader io.ReadWriteCloser, i int) {
  290. packet := make([]byte, mtu)
  291. out := make([]byte, mtu)
  292. fwPacket := &firewall.Packet{}
  293. nb := make([]byte, 12, 12)
  294. conntrackCache := firewall.NewConntrackCacheTicker(f.ctx, f.l, f.conntrackCacheTimeout)
  295. for {
  296. n, err := reader.Read(packet)
  297. if err != nil {
  298. if !f.closed.Load() {
  299. f.l.Error("Error while reading outbound packet, closing", "error", err, "reader", i)
  300. f.onFatal(err)
  301. }
  302. break
  303. }
  304. f.consumeInsidePacket(packet[:n], fwPacket, nb, out, i, conntrackCache.Get())
  305. }
  306. f.l.Debug("overlay reader is done", "reader", i)
  307. }
  308. func (f *Interface) RegisterConfigChangeCallbacks(c *config.C) {
  309. c.RegisterReloadCallback(f.reloadFirewall)
  310. c.RegisterReloadCallback(f.reloadSendRecvError)
  311. c.RegisterReloadCallback(f.reloadAcceptRecvError)
  312. c.RegisterReloadCallback(f.reloadDisconnectInvalid)
  313. c.RegisterReloadCallback(f.reloadMisc)
  314. for _, udpConn := range f.writers {
  315. c.RegisterReloadCallback(udpConn.ReloadConfig)
  316. }
  317. }
  318. func (f *Interface) reloadDisconnectInvalid(c *config.C) {
  319. initial := c.InitialLoad()
  320. if initial || c.HasChanged("pki.disconnect_invalid") {
  321. f.disconnectInvalid.Store(c.GetBool("pki.disconnect_invalid", true))
  322. if !initial {
  323. f.l.Info("pki.disconnect_invalid changed", "value", f.disconnectInvalid.Load())
  324. }
  325. }
  326. }
  327. func (f *Interface) reloadFirewall(c *config.C) {
  328. //TODO: need to trigger/detect if the certificate changed too
  329. if c.HasChanged("firewall") == false {
  330. f.l.Debug("No firewall config change detected")
  331. return
  332. }
  333. fw, err := NewFirewallFromConfig(f.l, f.pki.getCertState(), c)
  334. if err != nil {
  335. f.l.Error("Error while creating firewall during reload", "error", err)
  336. return
  337. }
  338. oldFw := f.firewall
  339. conntrack := oldFw.Conntrack
  340. conntrack.Lock()
  341. defer conntrack.Unlock()
  342. fw.rulesVersion = oldFw.rulesVersion + 1
  343. // If rulesVersion is back to zero, we have wrapped all the way around. Be
  344. // safe and just reset conntrack in this case.
  345. if fw.rulesVersion == 0 {
  346. f.l.Warn("firewall rulesVersion has overflowed, resetting conntrack",
  347. "firewallHashes", fw.GetRuleHashes(),
  348. "oldFirewallHashes", oldFw.GetRuleHashes(),
  349. "rulesVersion", fw.rulesVersion,
  350. )
  351. } else {
  352. fw.Conntrack = conntrack
  353. }
  354. f.firewall = fw
  355. oldFw.Destroy()
  356. f.l.Info("New firewall has been installed",
  357. "firewallHashes", fw.GetRuleHashes(),
  358. "oldFirewallHashes", oldFw.GetRuleHashes(),
  359. "rulesVersion", fw.rulesVersion,
  360. )
  361. }
  362. func (f *Interface) reloadSendRecvError(c *config.C) {
  363. if c.InitialLoad() || c.HasChanged("listen.send_recv_error") {
  364. stringValue := c.GetString("listen.send_recv_error", "always")
  365. switch stringValue {
  366. case "always":
  367. f.sendRecvErrorConfig = recvErrorAlways
  368. case "never":
  369. f.sendRecvErrorConfig = recvErrorNever
  370. case "private":
  371. f.sendRecvErrorConfig = recvErrorPrivate
  372. default:
  373. if c.GetBool("listen.send_recv_error", true) {
  374. f.sendRecvErrorConfig = recvErrorAlways
  375. } else {
  376. f.sendRecvErrorConfig = recvErrorNever
  377. }
  378. }
  379. f.l.Info("Loaded send_recv_error config", "sendRecvError", f.sendRecvErrorConfig.String())
  380. }
  381. }
  382. func (f *Interface) reloadAcceptRecvError(c *config.C) {
  383. if c.InitialLoad() || c.HasChanged("listen.accept_recv_error") {
  384. stringValue := c.GetString("listen.accept_recv_error", "always")
  385. switch stringValue {
  386. case "always":
  387. f.acceptRecvErrorConfig = recvErrorAlways
  388. case "never":
  389. f.acceptRecvErrorConfig = recvErrorNever
  390. case "private":
  391. f.acceptRecvErrorConfig = recvErrorPrivate
  392. default:
  393. if c.GetBool("listen.accept_recv_error", true) {
  394. f.acceptRecvErrorConfig = recvErrorAlways
  395. } else {
  396. f.acceptRecvErrorConfig = recvErrorNever
  397. }
  398. }
  399. f.l.Info("Loaded accept_recv_error config", "acceptRecvError", f.acceptRecvErrorConfig.String())
  400. }
  401. }
  402. func (f *Interface) reloadMisc(c *config.C) {
  403. if c.HasChanged("counters.try_promote") {
  404. n := c.GetUint32("counters.try_promote", defaultPromoteEvery)
  405. f.tryPromoteEvery.Store(n)
  406. f.l.Info("counters.try_promote has changed")
  407. }
  408. if c.HasChanged("counters.requery_every_packets") {
  409. n := c.GetUint32("counters.requery_every_packets", defaultReQueryEvery)
  410. f.reQueryEvery.Store(n)
  411. f.l.Info("counters.requery_every_packets has changed")
  412. }
  413. if c.HasChanged("timers.requery_wait_duration") {
  414. n := c.GetDuration("timers.requery_wait_duration", defaultReQueryWait)
  415. f.reQueryWait.Store(int64(n))
  416. f.l.Info("timers.requery_wait_duration has changed")
  417. }
  418. }
  419. func (f *Interface) emitStats(ctx context.Context, i time.Duration) {
  420. ticker := time.NewTicker(i)
  421. defer ticker.Stop()
  422. udpStats := udp.NewUDPStatsEmitter(f.writers)
  423. certExpirationGauge := metrics.GetOrRegisterGauge("certificate.ttl_seconds", nil)
  424. certInitiatingVersion := metrics.GetOrRegisterGauge("certificate.initiating_version", nil)
  425. certMaxVersion := metrics.GetOrRegisterGauge("certificate.max_version", nil)
  426. for {
  427. select {
  428. case <-ctx.Done():
  429. return
  430. case <-ticker.C:
  431. f.firewall.EmitStats()
  432. f.handshakeManager.EmitStats()
  433. udpStats()
  434. certState := f.pki.getCertState()
  435. defaultCrt := certState.GetDefaultCertificate()
  436. certExpirationGauge.Update(int64(defaultCrt.NotAfter().Sub(time.Now()) / time.Second))
  437. certInitiatingVersion.Update(int64(defaultCrt.Version()))
  438. // Report the max certificate version we are capable of using
  439. if certState.v2Cert != nil {
  440. certMaxVersion.Update(int64(certState.v2Cert.Version()))
  441. } else {
  442. certMaxVersion.Update(int64(certState.v1Cert.Version()))
  443. }
  444. }
  445. }
  446. }
  447. func (f *Interface) GetHostInfo(vpnIp netip.Addr) *HostInfo {
  448. return f.hostMap.QueryVpnAddr(vpnIp)
  449. }
  450. func (f *Interface) GetCertState() *CertState {
  451. return f.pki.getCertState()
  452. }
  453. func (f *Interface) Close() error {
  454. var errs []error
  455. f.closed.Store(true)
  456. // Release the udp readers
  457. for i, u := range f.writers {
  458. err := u.Close()
  459. if err != nil {
  460. f.l.Error("Error while closing udp socket", "error", err, "writer", i)
  461. errs = append(errs, err)
  462. }
  463. }
  464. // Release the tun device (closing the tun also closes all readers)
  465. closeErr := f.inside.Close()
  466. if closeErr != nil {
  467. errs = append(errs, closeErr)
  468. }
  469. f.wg.Done()
  470. return errors.Join(errs...)
  471. }