interface.go 14 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545
  1. package nebula
  2. import (
  3. "context"
  4. "errors"
  5. "fmt"
  6. "io"
  7. "net/netip"
  8. "runtime"
  9. "sync"
  10. "sync/atomic"
  11. "time"
  12. "github.com/gaissmai/bart"
  13. "github.com/rcrowley/go-metrics"
  14. "github.com/sirupsen/logrus"
  15. "github.com/slackhq/nebula/config"
  16. "github.com/slackhq/nebula/firewall"
  17. "github.com/slackhq/nebula/header"
  18. "github.com/slackhq/nebula/overlay"
  19. "github.com/slackhq/nebula/packet"
  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. ServeDns bool
  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 *logrus.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. serveDns bool
  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 sendRecvErrorConfig
  73. // rebindCount is used to decide if an active tunnel should trigger a punch notification through a lighthouse
  74. rebindCount int8
  75. version string
  76. conntrackCacheTimeout time.Duration
  77. writers []udp.Conn
  78. readers []io.ReadWriteCloser
  79. wg sync.WaitGroup
  80. metricHandshakes metrics.Histogram
  81. messageMetrics *MessageMetrics
  82. cachedPacketMetrics *cachedPacketMetrics
  83. l *logrus.Logger
  84. pktPool *packet.Pool
  85. inbound chan *packet.Packet
  86. outbound chan *packet.Packet
  87. }
  88. type EncWriter interface {
  89. SendVia(via *HostInfo,
  90. relay *Relay,
  91. ad,
  92. nb,
  93. out []byte,
  94. nocopy bool,
  95. )
  96. SendMessageToVpnAddr(t header.MessageType, st header.MessageSubType, vpnAddr netip.Addr, p, nb, out []byte)
  97. SendMessageToHostInfo(t header.MessageType, st header.MessageSubType, hostinfo *HostInfo, p, nb, out []byte)
  98. Handshake(vpnAddr netip.Addr)
  99. GetHostInfo(vpnAddr netip.Addr) *HostInfo
  100. GetCertState() *CertState
  101. }
  102. type sendRecvErrorConfig uint8
  103. const (
  104. sendRecvErrorAlways sendRecvErrorConfig = iota
  105. sendRecvErrorNever
  106. sendRecvErrorPrivate
  107. )
  108. func (s sendRecvErrorConfig) ShouldSendRecvError(endpoint netip.AddrPort) bool {
  109. switch s {
  110. case sendRecvErrorPrivate:
  111. return endpoint.Addr().IsPrivate()
  112. case sendRecvErrorAlways:
  113. return true
  114. case sendRecvErrorNever:
  115. return false
  116. default:
  117. panic(fmt.Errorf("invalid sendRecvErrorConfig value: %d", s))
  118. }
  119. }
  120. func (s sendRecvErrorConfig) String() string {
  121. switch s {
  122. case sendRecvErrorAlways:
  123. return "always"
  124. case sendRecvErrorNever:
  125. return "never"
  126. case sendRecvErrorPrivate:
  127. return "private"
  128. default:
  129. return fmt.Sprintf("invalid(%d)", s)
  130. }
  131. }
  132. func NewInterface(ctx context.Context, c *InterfaceConfig) (*Interface, error) {
  133. if c.Outside == nil {
  134. return nil, errors.New("no outside connection")
  135. }
  136. if c.Inside == nil {
  137. return nil, errors.New("no inside interface (tun)")
  138. }
  139. if c.pki == nil {
  140. return nil, errors.New("no certificate state")
  141. }
  142. if c.Firewall == nil {
  143. return nil, errors.New("no firewall rules")
  144. }
  145. if c.connectionManager == nil {
  146. return nil, errors.New("no connection manager")
  147. }
  148. cs := c.pki.getCertState()
  149. ifce := &Interface{
  150. pki: c.pki,
  151. hostMap: c.HostMap,
  152. outside: c.Outside,
  153. inside: c.Inside,
  154. firewall: c.Firewall,
  155. serveDns: c.ServeDns,
  156. handshakeManager: c.HandshakeManager,
  157. createTime: time.Now(),
  158. lightHouse: c.lightHouse,
  159. dropLocalBroadcast: c.DropLocalBroadcast,
  160. dropMulticast: c.DropMulticast,
  161. routines: c.routines,
  162. version: c.version,
  163. writers: make([]udp.Conn, c.routines),
  164. readers: make([]io.ReadWriteCloser, c.routines),
  165. myVpnNetworks: cs.myVpnNetworks,
  166. myVpnNetworksTable: cs.myVpnNetworksTable,
  167. myVpnAddrs: cs.myVpnAddrs,
  168. myVpnAddrsTable: cs.myVpnAddrsTable,
  169. myBroadcastAddrsTable: cs.myVpnBroadcastAddrsTable,
  170. relayManager: c.relayManager,
  171. connectionManager: c.connectionManager,
  172. conntrackCacheTimeout: c.ConntrackCacheTimeout,
  173. metricHandshakes: metrics.GetOrRegisterHistogram("handshakes", nil, metrics.NewExpDecaySample(1028, 0.015)),
  174. messageMetrics: c.MessageMetrics,
  175. cachedPacketMetrics: &cachedPacketMetrics{
  176. sent: metrics.GetOrRegisterCounter("hostinfo.cached_packets.sent", nil),
  177. dropped: metrics.GetOrRegisterCounter("hostinfo.cached_packets.dropped", nil),
  178. },
  179. //TODO: configurable size
  180. inbound: make(chan *packet.Packet, 2048),
  181. outbound: make(chan *packet.Packet, 2048),
  182. l: c.l,
  183. }
  184. ifce.pktPool = packet.GetPool()
  185. ifce.tryPromoteEvery.Store(c.tryPromoteEvery)
  186. ifce.reQueryEvery.Store(c.reQueryEvery)
  187. ifce.reQueryWait.Store(int64(c.reQueryWait))
  188. ifce.connectionManager.intf = ifce
  189. return ifce, nil
  190. }
  191. // activate creates the interface on the host. After the interface is created, any
  192. // other services that want to bind listeners to its IP may do so successfully. However,
  193. // the interface isn't going to process anything until run() is called.
  194. func (f *Interface) activate() error {
  195. // actually turn on tun dev
  196. addr, err := f.outside.LocalAddr()
  197. if err != nil {
  198. f.l.WithError(err).Error("Failed to get udp listen address")
  199. }
  200. f.l.WithField("interface", f.inside.Name()).WithField("networks", f.myVpnNetworks).
  201. WithField("build", f.version).WithField("udpAddr", addr).
  202. WithField("boringcrypto", boringEnabled()).
  203. Info("Nebula interface is active")
  204. metrics.GetOrRegisterGauge("routines", nil).Update(int64(f.routines))
  205. // Prepare n tun queues
  206. var reader io.ReadWriteCloser = f.inside
  207. for i := 0; i < f.routines; i++ {
  208. if i > 0 {
  209. reader, err = f.inside.NewMultiQueueReader()
  210. if err != nil {
  211. return err
  212. }
  213. }
  214. f.readers[i] = reader
  215. }
  216. if err = f.inside.Activate(); err != nil {
  217. f.inside.Close()
  218. return err
  219. }
  220. return nil
  221. }
  222. func (f *Interface) run(c context.Context) (func(), error) {
  223. for i := 0; i < f.routines; i++ {
  224. // read packets from udp and queue to f.inbound
  225. f.wg.Add(1)
  226. go f.listenOut(i)
  227. // Launch n queues to read packets from inside tun dev and queue to f.outbound
  228. //todo this never stops f.wg.Add(1)
  229. go f.listenIn(f.readers[i], i)
  230. // Launch n workers to process traffic from f.inbound and smash it onto the inside of the tun
  231. f.wg.Add(1)
  232. go f.workerIn(i, c)
  233. f.wg.Add(1)
  234. go f.workerIn(i, c)
  235. // read from f.outbound and write to UDP (outside the tun)
  236. f.wg.Add(1)
  237. go f.workerOut(i, c)
  238. }
  239. return f.wg.Wait, nil
  240. }
  241. func (f *Interface) listenOut(i int) {
  242. runtime.LockOSThread()
  243. var li udp.Conn
  244. if i > 0 {
  245. li = f.writers[i]
  246. } else {
  247. li = f.outside
  248. }
  249. err := li.ListenOut(f.pktPool.Get, f.inbound)
  250. if err != nil && !f.closed.Load() {
  251. f.l.WithError(err).Error("Error while reading packet inbound packet, closing")
  252. //TODO: Trigger Control to close
  253. }
  254. f.l.Debugf("underlay reader %v is done", i)
  255. f.wg.Done()
  256. }
  257. func (f *Interface) listenIn(reader io.ReadWriteCloser, i int) {
  258. runtime.LockOSThread()
  259. for {
  260. p := f.pktPool.Get()
  261. n, err := reader.Read(p.Payload)
  262. if err != nil {
  263. if !f.closed.Load() {
  264. f.l.WithError(err).Error("Error while reading outbound packet, closing")
  265. //TODO: Trigger Control to close
  266. }
  267. break
  268. }
  269. p.Payload = (p.Payload)[:n]
  270. //TODO: nonblocking channel write
  271. f.outbound <- p
  272. //select {
  273. //case f.outbound <- p:
  274. //default:
  275. // f.l.Error("Dropped packet from outbound channel")
  276. //}
  277. }
  278. f.l.Debugf("overlay reader %v is done", i)
  279. f.wg.Done()
  280. }
  281. func (f *Interface) workerIn(i int, ctx context.Context) {
  282. lhh := f.lightHouse.NewRequestHandler()
  283. conntrackCache := firewall.NewConntrackCacheTicker(f.conntrackCacheTimeout)
  284. fwPacket2 := &firewall.Packet{}
  285. nb2 := make([]byte, 12, 12)
  286. result2 := make([]byte, mtu)
  287. h := &header.H{}
  288. for {
  289. select {
  290. case p := <-f.inbound:
  291. if p.SegSize > 0 && p.SegSize < len(p.Payload) {
  292. for offset := 0; offset < len(p.Payload); offset += p.SegSize {
  293. end := offset + p.SegSize
  294. if end > len(p.Payload) {
  295. end = len(p.Payload)
  296. }
  297. f.readOutsidePackets(p.Addr, nil, result2[:0], p.Payload[offset:end], h, fwPacket2, lhh, nb2, i, conntrackCache.Get(f.l))
  298. }
  299. } else {
  300. f.readOutsidePackets(p.Addr, nil, result2[:0], p.Payload, h, fwPacket2, lhh, nb2, i, conntrackCache.Get(f.l))
  301. }
  302. f.pktPool.Put(p)
  303. case <-ctx.Done():
  304. f.wg.Done()
  305. return
  306. }
  307. }
  308. }
  309. func (f *Interface) workerOut(i int, ctx context.Context) {
  310. conntrackCache := firewall.NewConntrackCacheTicker(f.conntrackCacheTimeout)
  311. fwPacket1 := &firewall.Packet{}
  312. nb1 := make([]byte, 12, 12)
  313. result1 := make([]byte, mtu)
  314. for {
  315. select {
  316. case data := <-f.outbound:
  317. f.consumeInsidePacket(data.Payload, fwPacket1, nb1, result1, i, conntrackCache.Get(f.l))
  318. f.pktPool.Put(data)
  319. case <-ctx.Done():
  320. f.wg.Done()
  321. return
  322. }
  323. }
  324. }
  325. func (f *Interface) RegisterConfigChangeCallbacks(c *config.C) {
  326. c.RegisterReloadCallback(f.reloadFirewall)
  327. c.RegisterReloadCallback(f.reloadSendRecvError)
  328. c.RegisterReloadCallback(f.reloadDisconnectInvalid)
  329. c.RegisterReloadCallback(f.reloadMisc)
  330. for _, udpConn := range f.writers {
  331. c.RegisterReloadCallback(udpConn.ReloadConfig)
  332. }
  333. }
  334. func (f *Interface) reloadDisconnectInvalid(c *config.C) {
  335. initial := c.InitialLoad()
  336. if initial || c.HasChanged("pki.disconnect_invalid") {
  337. f.disconnectInvalid.Store(c.GetBool("pki.disconnect_invalid", true))
  338. if !initial {
  339. f.l.Infof("pki.disconnect_invalid changed to %v", f.disconnectInvalid.Load())
  340. }
  341. }
  342. }
  343. func (f *Interface) reloadFirewall(c *config.C) {
  344. //TODO: need to trigger/detect if the certificate changed too
  345. if c.HasChanged("firewall") == false {
  346. f.l.Debug("No firewall config change detected")
  347. return
  348. }
  349. fw, err := NewFirewallFromConfig(f.l, f.pki.getCertState(), c)
  350. if err != nil {
  351. f.l.WithError(err).Error("Error while creating firewall during reload")
  352. return
  353. }
  354. oldFw := f.firewall
  355. conntrack := oldFw.Conntrack
  356. conntrack.Lock()
  357. defer conntrack.Unlock()
  358. fw.rulesVersion = oldFw.rulesVersion + 1
  359. // If rulesVersion is back to zero, we have wrapped all the way around. Be
  360. // safe and just reset conntrack in this case.
  361. if fw.rulesVersion == 0 {
  362. f.l.WithField("firewallHashes", fw.GetRuleHashes()).
  363. WithField("oldFirewallHashes", oldFw.GetRuleHashes()).
  364. WithField("rulesVersion", fw.rulesVersion).
  365. Warn("firewall rulesVersion has overflowed, resetting conntrack")
  366. } else {
  367. fw.Conntrack = conntrack
  368. }
  369. f.firewall = fw
  370. oldFw.Destroy()
  371. f.l.WithField("firewallHashes", fw.GetRuleHashes()).
  372. WithField("oldFirewallHashes", oldFw.GetRuleHashes()).
  373. WithField("rulesVersion", fw.rulesVersion).
  374. Info("New firewall has been installed")
  375. }
  376. func (f *Interface) reloadSendRecvError(c *config.C) {
  377. if c.InitialLoad() || c.HasChanged("listen.send_recv_error") {
  378. stringValue := c.GetString("listen.send_recv_error", "always")
  379. switch stringValue {
  380. case "always":
  381. f.sendRecvErrorConfig = sendRecvErrorAlways
  382. case "never":
  383. f.sendRecvErrorConfig = sendRecvErrorNever
  384. case "private":
  385. f.sendRecvErrorConfig = sendRecvErrorPrivate
  386. default:
  387. if c.GetBool("listen.send_recv_error", true) {
  388. f.sendRecvErrorConfig = sendRecvErrorAlways
  389. } else {
  390. f.sendRecvErrorConfig = sendRecvErrorNever
  391. }
  392. }
  393. f.l.WithField("sendRecvError", f.sendRecvErrorConfig.String()).
  394. Info("Loaded send_recv_error config")
  395. }
  396. }
  397. func (f *Interface) reloadMisc(c *config.C) {
  398. if c.HasChanged("counters.try_promote") {
  399. n := c.GetUint32("counters.try_promote", defaultPromoteEvery)
  400. f.tryPromoteEvery.Store(n)
  401. f.l.Info("counters.try_promote has changed")
  402. }
  403. if c.HasChanged("counters.requery_every_packets") {
  404. n := c.GetUint32("counters.requery_every_packets", defaultReQueryEvery)
  405. f.reQueryEvery.Store(n)
  406. f.l.Info("counters.requery_every_packets has changed")
  407. }
  408. if c.HasChanged("timers.requery_wait_duration") {
  409. n := c.GetDuration("timers.requery_wait_duration", defaultReQueryWait)
  410. f.reQueryWait.Store(int64(n))
  411. f.l.Info("timers.requery_wait_duration has changed")
  412. }
  413. }
  414. func (f *Interface) emitStats(ctx context.Context, i time.Duration) {
  415. ticker := time.NewTicker(i)
  416. defer ticker.Stop()
  417. udpStats := udp.NewUDPStatsEmitter(f.writers)
  418. certExpirationGauge := metrics.GetOrRegisterGauge("certificate.ttl_seconds", nil)
  419. certInitiatingVersion := metrics.GetOrRegisterGauge("certificate.initiating_version", nil)
  420. certMaxVersion := metrics.GetOrRegisterGauge("certificate.max_version", nil)
  421. for {
  422. select {
  423. case <-ctx.Done():
  424. return
  425. case <-ticker.C:
  426. f.firewall.EmitStats()
  427. f.handshakeManager.EmitStats()
  428. udpStats()
  429. certState := f.pki.getCertState()
  430. defaultCrt := certState.GetDefaultCertificate()
  431. certExpirationGauge.Update(int64(defaultCrt.NotAfter().Sub(time.Now()) / time.Second))
  432. certInitiatingVersion.Update(int64(defaultCrt.Version()))
  433. // Report the max certificate version we are capable of using
  434. if certState.v2Cert != nil {
  435. certMaxVersion.Update(int64(certState.v2Cert.Version()))
  436. } else {
  437. certMaxVersion.Update(int64(certState.v1Cert.Version()))
  438. }
  439. }
  440. }
  441. }
  442. func (f *Interface) GetHostInfo(vpnIp netip.Addr) *HostInfo {
  443. return f.hostMap.QueryVpnAddr(vpnIp)
  444. }
  445. func (f *Interface) GetCertState() *CertState {
  446. return f.pki.getCertState()
  447. }
  448. func (f *Interface) Close() error {
  449. f.closed.Store(true)
  450. // Release the udp readers
  451. for _, u := range f.writers {
  452. err := u.Close()
  453. if err != nil {
  454. f.l.WithError(err).Error("Error while closing udp socket")
  455. }
  456. }
  457. // Release the tun readers
  458. for _, u := range f.readers {
  459. err := u.Close()
  460. if err != nil {
  461. f.l.WithError(err).Error("Error while closing tun device")
  462. }
  463. }
  464. return nil
  465. }