interface.go 15 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554
  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. 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 *slog.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 *slog.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.Error("Failed to get udp listen address", "error", err)
  198. }
  199. f.l.Info("Nebula interface is active",
  200. "interface", f.inside.Name(),
  201. "networks", f.myVpnNetworks,
  202. "build", f.version,
  203. "udpAddr", addr,
  204. "boringcrypto", boringEnabled(),
  205. )
  206. if f.routines > 1 {
  207. if !f.inside.SupportsMultiqueue() || !f.outside.SupportsMultipleReaders() {
  208. f.routines = 1
  209. f.l.Warn("routines is not supported on this platform, falling back to a single routine")
  210. }
  211. }
  212. metrics.GetOrRegisterGauge("routines", nil).Update(int64(f.routines))
  213. // Prepare n tun queues
  214. var reader io.ReadWriteCloser = f.inside
  215. for i := 0; i < f.routines; i++ {
  216. if i > 0 {
  217. reader, err = f.inside.NewMultiQueueReader()
  218. if err != nil {
  219. return err
  220. }
  221. }
  222. f.readers[i] = reader
  223. }
  224. f.wg.Add(1) // for us to wait on Close() to return
  225. if err = f.inside.Activate(); err != nil {
  226. f.wg.Done()
  227. f.inside.Close()
  228. return err
  229. }
  230. return nil
  231. }
  232. func (f *Interface) run() (func() error, error) {
  233. // Launch n queues to read packets from udp
  234. for i := 0; i < f.routines; i++ {
  235. f.wg.Go(func() {
  236. f.listenOut(i)
  237. })
  238. }
  239. // Launch n queues to read packets from tun dev
  240. for i := 0; i < f.routines; i++ {
  241. f.wg.Go(func() {
  242. f.listenIn(f.readers[i], i)
  243. })
  244. }
  245. return func() error {
  246. f.wg.Wait()
  247. if e := f.fatalErr.Load(); e != nil {
  248. return *e
  249. }
  250. return nil
  251. }, nil
  252. }
  253. // onFatal stores the first fatal reader error, and calls triggerShutdown if it was the first one
  254. func (f *Interface) onFatal(err error) {
  255. swapped := f.fatalErr.CompareAndSwap(nil, &err)
  256. if !swapped {
  257. return
  258. }
  259. if f.triggerShutdown != nil {
  260. f.triggerShutdown()
  261. }
  262. }
  263. func (f *Interface) listenOut(i int) {
  264. var li udp.Conn
  265. if i > 0 {
  266. li = f.writers[i]
  267. } else {
  268. li = f.outside
  269. }
  270. ctCache := firewall.NewConntrackCacheTicker(f.ctx, f.l, f.conntrackCacheTimeout)
  271. lhh := f.lightHouse.NewRequestHandler()
  272. plaintext := make([]byte, udp.MTU)
  273. h := &header.H{}
  274. fwPacket := &firewall.Packet{}
  275. nb := make([]byte, 12, 12)
  276. err := li.ListenOut(func(fromUdpAddr netip.AddrPort, payload []byte) {
  277. f.readOutsidePackets(ViaSender{UdpAddr: fromUdpAddr}, plaintext[:0], payload, h, fwPacket, lhh, nb, i, ctCache.Get())
  278. })
  279. if err != nil && !f.closed.Load() {
  280. f.l.Error("Error while reading inbound packet, closing", "error", err)
  281. f.onFatal(err)
  282. }
  283. f.l.Debug("underlay reader is done", "reader", i)
  284. }
  285. func (f *Interface) listenIn(reader io.ReadWriteCloser, i int) {
  286. packet := make([]byte, mtu)
  287. out := make([]byte, mtu)
  288. fwPacket := &firewall.Packet{}
  289. nb := make([]byte, 12, 12)
  290. conntrackCache := firewall.NewConntrackCacheTicker(f.ctx, f.l, f.conntrackCacheTimeout)
  291. for {
  292. n, err := reader.Read(packet)
  293. if err != nil {
  294. if !f.closed.Load() {
  295. f.l.Error("Error while reading outbound packet, closing", "error", err, "reader", i)
  296. f.onFatal(err)
  297. }
  298. break
  299. }
  300. f.consumeInsidePacket(packet[:n], fwPacket, nb, out, i, conntrackCache.Get())
  301. }
  302. f.l.Debug("overlay reader is done", "reader", i)
  303. }
  304. func (f *Interface) RegisterConfigChangeCallbacks(c *config.C) {
  305. c.RegisterReloadCallback(f.reloadFirewall)
  306. c.RegisterReloadCallback(f.reloadSendRecvError)
  307. c.RegisterReloadCallback(f.reloadAcceptRecvError)
  308. c.RegisterReloadCallback(f.reloadDisconnectInvalid)
  309. c.RegisterReloadCallback(f.reloadMisc)
  310. for _, udpConn := range f.writers {
  311. c.RegisterReloadCallback(udpConn.ReloadConfig)
  312. }
  313. }
  314. func (f *Interface) reloadDisconnectInvalid(c *config.C) {
  315. initial := c.InitialLoad()
  316. if initial || c.HasChanged("pki.disconnect_invalid") {
  317. f.disconnectInvalid.Store(c.GetBool("pki.disconnect_invalid", true))
  318. if !initial {
  319. f.l.Info("pki.disconnect_invalid changed", "value", f.disconnectInvalid.Load())
  320. }
  321. }
  322. }
  323. func (f *Interface) reloadFirewall(c *config.C) {
  324. //TODO: need to trigger/detect if the certificate changed too
  325. if c.HasChanged("firewall") == false {
  326. f.l.Debug("No firewall config change detected")
  327. return
  328. }
  329. fw, err := NewFirewallFromConfig(f.l, f.pki.getCertState(), c)
  330. if err != nil {
  331. f.l.Error("Error while creating firewall during reload", "error", err)
  332. return
  333. }
  334. oldFw := f.firewall
  335. conntrack := oldFw.Conntrack
  336. conntrack.Lock()
  337. defer conntrack.Unlock()
  338. fw.rulesVersion = oldFw.rulesVersion + 1
  339. // If rulesVersion is back to zero, we have wrapped all the way around. Be
  340. // safe and just reset conntrack in this case.
  341. if fw.rulesVersion == 0 {
  342. f.l.Warn("firewall rulesVersion has overflowed, resetting conntrack",
  343. "firewallHashes", fw.GetRuleHashes(),
  344. "oldFirewallHashes", oldFw.GetRuleHashes(),
  345. "rulesVersion", fw.rulesVersion,
  346. )
  347. } else {
  348. fw.Conntrack = conntrack
  349. }
  350. f.firewall = fw
  351. oldFw.Destroy()
  352. f.l.Info("New firewall has been installed",
  353. "firewallHashes", fw.GetRuleHashes(),
  354. "oldFirewallHashes", oldFw.GetRuleHashes(),
  355. "rulesVersion", fw.rulesVersion,
  356. )
  357. }
  358. func (f *Interface) reloadSendRecvError(c *config.C) {
  359. if c.InitialLoad() || c.HasChanged("listen.send_recv_error") {
  360. stringValue := c.GetString("listen.send_recv_error", "always")
  361. switch stringValue {
  362. case "always":
  363. f.sendRecvErrorConfig = recvErrorAlways
  364. case "never":
  365. f.sendRecvErrorConfig = recvErrorNever
  366. case "private":
  367. f.sendRecvErrorConfig = recvErrorPrivate
  368. default:
  369. if c.GetBool("listen.send_recv_error", true) {
  370. f.sendRecvErrorConfig = recvErrorAlways
  371. } else {
  372. f.sendRecvErrorConfig = recvErrorNever
  373. }
  374. }
  375. f.l.Info("Loaded send_recv_error config", "sendRecvError", f.sendRecvErrorConfig.String())
  376. }
  377. }
  378. func (f *Interface) reloadAcceptRecvError(c *config.C) {
  379. if c.InitialLoad() || c.HasChanged("listen.accept_recv_error") {
  380. stringValue := c.GetString("listen.accept_recv_error", "always")
  381. switch stringValue {
  382. case "always":
  383. f.acceptRecvErrorConfig = recvErrorAlways
  384. case "never":
  385. f.acceptRecvErrorConfig = recvErrorNever
  386. case "private":
  387. f.acceptRecvErrorConfig = recvErrorPrivate
  388. default:
  389. if c.GetBool("listen.accept_recv_error", true) {
  390. f.acceptRecvErrorConfig = recvErrorAlways
  391. } else {
  392. f.acceptRecvErrorConfig = recvErrorNever
  393. }
  394. }
  395. f.l.Info("Loaded accept_recv_error config", "acceptRecvError", f.acceptRecvErrorConfig.String())
  396. }
  397. }
  398. func (f *Interface) reloadMisc(c *config.C) {
  399. if c.HasChanged("counters.try_promote") {
  400. n := c.GetUint32("counters.try_promote", defaultPromoteEvery)
  401. f.tryPromoteEvery.Store(n)
  402. f.l.Info("counters.try_promote has changed")
  403. }
  404. if c.HasChanged("counters.requery_every_packets") {
  405. n := c.GetUint32("counters.requery_every_packets", defaultReQueryEvery)
  406. f.reQueryEvery.Store(n)
  407. f.l.Info("counters.requery_every_packets has changed")
  408. }
  409. if c.HasChanged("timers.requery_wait_duration") {
  410. n := c.GetDuration("timers.requery_wait_duration", defaultReQueryWait)
  411. f.reQueryWait.Store(int64(n))
  412. f.l.Info("timers.requery_wait_duration has changed")
  413. }
  414. }
  415. func (f *Interface) emitStats(ctx context.Context, i time.Duration) {
  416. ticker := time.NewTicker(i)
  417. defer ticker.Stop()
  418. udpStats := udp.NewUDPStatsEmitter(f.writers)
  419. certExpirationGauge := metrics.GetOrRegisterGauge("certificate.ttl_seconds", nil)
  420. certInitiatingVersion := metrics.GetOrRegisterGauge("certificate.initiating_version", nil)
  421. certMaxVersion := metrics.GetOrRegisterGauge("certificate.max_version", nil)
  422. emit := func() {
  423. f.firewall.EmitStats()
  424. f.handshakeManager.EmitStats()
  425. udpStats()
  426. certState := f.pki.getCertState()
  427. defaultCrt := certState.GetDefaultCertificate()
  428. certExpirationGauge.Update(int64(defaultCrt.NotAfter().Sub(time.Now()) / time.Second))
  429. certInitiatingVersion.Update(int64(defaultCrt.Version()))
  430. // Report the max certificate version we are capable of using
  431. if certState.v2Cert != nil {
  432. certMaxVersion.Update(int64(certState.v2Cert.Version()))
  433. } else {
  434. certMaxVersion.Update(int64(certState.v1Cert.Version()))
  435. }
  436. }
  437. // Prime gauges so a Prometheus scrape that lands before the first tick
  438. // sees real values instead of the zero defaults (issue #907).
  439. emit()
  440. for {
  441. select {
  442. case <-ctx.Done():
  443. return
  444. case <-ticker.C:
  445. emit()
  446. }
  447. }
  448. }
  449. func (f *Interface) GetHostInfo(vpnIp netip.Addr) *HostInfo {
  450. return f.hostMap.QueryVpnAddr(vpnIp)
  451. }
  452. func (f *Interface) GetCertState() *CertState {
  453. return f.pki.getCertState()
  454. }
  455. func (f *Interface) Close() error {
  456. var errs []error
  457. f.closed.Store(true)
  458. // Release the udp readers
  459. for i, u := range f.writers {
  460. err := u.Close()
  461. if err != nil {
  462. f.l.Error("Error while closing udp socket", "error", err, "writer", i)
  463. errs = append(errs, err)
  464. }
  465. }
  466. // Release the tun device (closing the tun also closes all readers)
  467. closeErr := f.inside.Close()
  468. if closeErr != nil {
  469. errs = append(errs, closeErr)
  470. }
  471. f.wg.Done()
  472. return errors.Join(errs...)
  473. }