interface.go 13 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470
  1. package nebula
  2. import (
  3. "context"
  4. "errors"
  5. "fmt"
  6. "io"
  7. "net/netip"
  8. "os"
  9. "runtime"
  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/udp"
  20. )
  21. const mtu = 9001
  22. type InterfaceConfig struct {
  23. HostMap *HostMap
  24. Outside udp.Conn
  25. Inside overlay.Device
  26. pki *PKI
  27. Cipher string
  28. Firewall *Firewall
  29. ServeDns bool
  30. HandshakeManager *HandshakeManager
  31. lightHouse *LightHouse
  32. connectionManager *connectionManager
  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 *logrus.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. handshakeManager *HandshakeManager
  54. serveDns bool
  55. createTime time.Time
  56. lightHouse *LightHouse
  57. myBroadcastAddrsTable *bart.Lite
  58. myVpnAddrs []netip.Addr // A list of addresses assigned to us via our certificate
  59. myVpnAddrsTable *bart.Lite
  60. myVpnNetworks []netip.Prefix // A list of networks assigned to us via our certificate
  61. myVpnNetworksTable *bart.Lite
  62. dropLocalBroadcast bool
  63. dropMulticast bool
  64. routines int
  65. disconnectInvalid atomic.Bool
  66. closed atomic.Bool
  67. relayManager *relayManager
  68. tryPromoteEvery atomic.Uint32
  69. reQueryEvery atomic.Uint32
  70. reQueryWait atomic.Int64
  71. sendRecvErrorConfig sendRecvErrorConfig
  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. writers []udp.Conn
  77. readers []io.ReadWriteCloser
  78. metricHandshakes metrics.Histogram
  79. messageMetrics *MessageMetrics
  80. cachedPacketMetrics *cachedPacketMetrics
  81. l *logrus.Logger
  82. }
  83. type EncWriter interface {
  84. SendVia(via *HostInfo,
  85. relay *Relay,
  86. ad,
  87. nb,
  88. out []byte,
  89. nocopy bool,
  90. )
  91. SendMessageToVpnAddr(t header.MessageType, st header.MessageSubType, vpnAddr netip.Addr, p, nb, out []byte)
  92. SendMessageToHostInfo(t header.MessageType, st header.MessageSubType, hostinfo *HostInfo, p, nb, out []byte)
  93. Handshake(vpnAddr netip.Addr)
  94. GetHostInfo(vpnAddr netip.Addr) *HostInfo
  95. GetCertState() *CertState
  96. }
  97. type sendRecvErrorConfig uint8
  98. const (
  99. sendRecvErrorAlways sendRecvErrorConfig = iota
  100. sendRecvErrorNever
  101. sendRecvErrorPrivate
  102. )
  103. func (s sendRecvErrorConfig) ShouldSendRecvError(endpoint netip.AddrPort) bool {
  104. switch s {
  105. case sendRecvErrorPrivate:
  106. return endpoint.Addr().IsPrivate()
  107. case sendRecvErrorAlways:
  108. return true
  109. case sendRecvErrorNever:
  110. return false
  111. default:
  112. panic(fmt.Errorf("invalid sendRecvErrorConfig value: %d", s))
  113. }
  114. }
  115. func (s sendRecvErrorConfig) String() string {
  116. switch s {
  117. case sendRecvErrorAlways:
  118. return "always"
  119. case sendRecvErrorNever:
  120. return "never"
  121. case sendRecvErrorPrivate:
  122. return "private"
  123. default:
  124. return fmt.Sprintf("invalid(%d)", s)
  125. }
  126. }
  127. func NewInterface(ctx context.Context, c *InterfaceConfig) (*Interface, error) {
  128. if c.Outside == nil {
  129. return nil, errors.New("no outside connection")
  130. }
  131. if c.Inside == nil {
  132. return nil, errors.New("no inside interface (tun)")
  133. }
  134. if c.pki == nil {
  135. return nil, errors.New("no certificate state")
  136. }
  137. if c.Firewall == nil {
  138. return nil, errors.New("no firewall rules")
  139. }
  140. if c.connectionManager == nil {
  141. return nil, errors.New("no connection manager")
  142. }
  143. cs := c.pki.getCertState()
  144. ifce := &Interface{
  145. pki: c.pki,
  146. hostMap: c.HostMap,
  147. outside: c.Outside,
  148. inside: c.Inside,
  149. firewall: c.Firewall,
  150. serveDns: c.ServeDns,
  151. handshakeManager: c.HandshakeManager,
  152. createTime: time.Now(),
  153. lightHouse: c.lightHouse,
  154. dropLocalBroadcast: c.DropLocalBroadcast,
  155. dropMulticast: c.DropMulticast,
  156. routines: c.routines,
  157. version: c.version,
  158. writers: make([]udp.Conn, c.routines),
  159. readers: make([]io.ReadWriteCloser, c.routines),
  160. myVpnNetworks: cs.myVpnNetworks,
  161. myVpnNetworksTable: cs.myVpnNetworksTable,
  162. myVpnAddrs: cs.myVpnAddrs,
  163. myVpnAddrsTable: cs.myVpnAddrsTable,
  164. myBroadcastAddrsTable: cs.myVpnBroadcastAddrsTable,
  165. relayManager: c.relayManager,
  166. connectionManager: c.connectionManager,
  167. conntrackCacheTimeout: c.ConntrackCacheTimeout,
  168. metricHandshakes: metrics.GetOrRegisterHistogram("handshakes", nil, metrics.NewExpDecaySample(1028, 0.015)),
  169. messageMetrics: c.MessageMetrics,
  170. cachedPacketMetrics: &cachedPacketMetrics{
  171. sent: metrics.GetOrRegisterCounter("hostinfo.cached_packets.sent", nil),
  172. dropped: metrics.GetOrRegisterCounter("hostinfo.cached_packets.dropped", nil),
  173. },
  174. l: c.l,
  175. }
  176. ifce.tryPromoteEvery.Store(c.tryPromoteEvery)
  177. ifce.reQueryEvery.Store(c.reQueryEvery)
  178. ifce.reQueryWait.Store(int64(c.reQueryWait))
  179. ifce.connectionManager.intf = ifce
  180. return ifce, nil
  181. }
  182. // activate creates the interface on the host. After the interface is created, any
  183. // other services that want to bind listeners to its IP may do so successfully. However,
  184. // the interface isn't going to process anything until run() is called.
  185. func (f *Interface) activate() {
  186. // actually turn on tun dev
  187. addr, err := f.outside.LocalAddr()
  188. if err != nil {
  189. f.l.WithError(err).Error("Failed to get udp listen address")
  190. }
  191. f.l.WithField("interface", f.inside.Name()).WithField("networks", f.myVpnNetworks).
  192. WithField("build", f.version).WithField("udpAddr", addr).
  193. WithField("boringcrypto", boringEnabled()).
  194. Info("Nebula interface is active")
  195. if f.routines > 1 {
  196. if !f.inside.SupportsMultiqueue() || !f.outside.SupportsMultipleReaders() {
  197. f.routines = 1
  198. f.l.Warn("routines is not supported on this platform, falling back to a single routine")
  199. }
  200. }
  201. metrics.GetOrRegisterGauge("routines", nil).Update(int64(f.routines))
  202. // Prepare n tun queues
  203. var reader io.ReadWriteCloser = f.inside
  204. for i := 0; i < f.routines; i++ {
  205. if i > 0 {
  206. reader, err = f.inside.NewMultiQueueReader()
  207. if err != nil {
  208. f.l.Fatal(err)
  209. }
  210. }
  211. f.readers[i] = reader
  212. }
  213. if err := f.inside.Activate(); err != nil {
  214. f.inside.Close()
  215. f.l.Fatal(err)
  216. }
  217. }
  218. func (f *Interface) run() {
  219. // Launch n queues to read packets from udp
  220. for i := 0; i < f.routines; i++ {
  221. go f.listenOut(i)
  222. }
  223. // Launch n queues to read packets from tun dev
  224. for i := 0; i < f.routines; i++ {
  225. go f.listenIn(f.readers[i], i)
  226. }
  227. }
  228. func (f *Interface) listenOut(i int) {
  229. runtime.LockOSThread()
  230. var li udp.Conn
  231. if i > 0 {
  232. li = f.writers[i]
  233. } else {
  234. li = f.outside
  235. }
  236. ctCache := firewall.NewConntrackCacheTicker(f.conntrackCacheTimeout)
  237. lhh := f.lightHouse.NewRequestHandler()
  238. plaintext := make([]byte, udp.MTU)
  239. h := &header.H{}
  240. fwPacket := &firewall.Packet{}
  241. nb := make([]byte, 12, 12)
  242. li.ListenOut(func(fromUdpAddr netip.AddrPort, payload []byte) {
  243. f.readOutsidePackets(ViaSender{UdpAddr: fromUdpAddr}, plaintext[:0], payload, h, fwPacket, lhh, nb, i, ctCache.Get(f.l))
  244. })
  245. }
  246. func (f *Interface) listenIn(reader io.ReadWriteCloser, i int) {
  247. runtime.LockOSThread()
  248. packet := make([]byte, mtu)
  249. out := make([]byte, mtu)
  250. fwPacket := &firewall.Packet{}
  251. nb := make([]byte, 12, 12)
  252. conntrackCache := firewall.NewConntrackCacheTicker(f.conntrackCacheTimeout)
  253. for {
  254. n, err := reader.Read(packet)
  255. if err != nil {
  256. if errors.Is(err, os.ErrClosed) && f.closed.Load() {
  257. return
  258. }
  259. f.l.WithError(err).Error("Error while reading outbound packet")
  260. // This only seems to happen when something fatal happens to the fd, so exit.
  261. os.Exit(2)
  262. }
  263. f.consumeInsidePacket(packet[:n], fwPacket, nb, out, i, conntrackCache.Get(f.l))
  264. }
  265. }
  266. func (f *Interface) RegisterConfigChangeCallbacks(c *config.C) {
  267. c.RegisterReloadCallback(f.reloadFirewall)
  268. c.RegisterReloadCallback(f.reloadSendRecvError)
  269. c.RegisterReloadCallback(f.reloadDisconnectInvalid)
  270. c.RegisterReloadCallback(f.reloadMisc)
  271. for _, udpConn := range f.writers {
  272. c.RegisterReloadCallback(udpConn.ReloadConfig)
  273. }
  274. }
  275. func (f *Interface) reloadDisconnectInvalid(c *config.C) {
  276. initial := c.InitialLoad()
  277. if initial || c.HasChanged("pki.disconnect_invalid") {
  278. f.disconnectInvalid.Store(c.GetBool("pki.disconnect_invalid", true))
  279. if !initial {
  280. f.l.Infof("pki.disconnect_invalid changed to %v", f.disconnectInvalid.Load())
  281. }
  282. }
  283. }
  284. func (f *Interface) reloadFirewall(c *config.C) {
  285. //TODO: need to trigger/detect if the certificate changed too
  286. if c.HasChanged("firewall") == false {
  287. f.l.Debug("No firewall config change detected")
  288. return
  289. }
  290. fw, err := NewFirewallFromConfig(f.l, f.pki.getCertState(), c)
  291. if err != nil {
  292. f.l.WithError(err).Error("Error while creating firewall during reload")
  293. return
  294. }
  295. oldFw := f.firewall
  296. conntrack := oldFw.Conntrack
  297. conntrack.Lock()
  298. defer conntrack.Unlock()
  299. fw.rulesVersion = oldFw.rulesVersion + 1
  300. // If rulesVersion is back to zero, we have wrapped all the way around. Be
  301. // safe and just reset conntrack in this case.
  302. if fw.rulesVersion == 0 {
  303. f.l.WithField("firewallHashes", fw.GetRuleHashes()).
  304. WithField("oldFirewallHashes", oldFw.GetRuleHashes()).
  305. WithField("rulesVersion", fw.rulesVersion).
  306. Warn("firewall rulesVersion has overflowed, resetting conntrack")
  307. } else {
  308. fw.Conntrack = conntrack
  309. }
  310. f.firewall = fw
  311. oldFw.Destroy()
  312. f.l.WithField("firewallHashes", fw.GetRuleHashes()).
  313. WithField("oldFirewallHashes", oldFw.GetRuleHashes()).
  314. WithField("rulesVersion", fw.rulesVersion).
  315. Info("New firewall has been installed")
  316. }
  317. func (f *Interface) reloadSendRecvError(c *config.C) {
  318. if c.InitialLoad() || c.HasChanged("listen.send_recv_error") {
  319. stringValue := c.GetString("listen.send_recv_error", "always")
  320. switch stringValue {
  321. case "always":
  322. f.sendRecvErrorConfig = sendRecvErrorAlways
  323. case "never":
  324. f.sendRecvErrorConfig = sendRecvErrorNever
  325. case "private":
  326. f.sendRecvErrorConfig = sendRecvErrorPrivate
  327. default:
  328. if c.GetBool("listen.send_recv_error", true) {
  329. f.sendRecvErrorConfig = sendRecvErrorAlways
  330. } else {
  331. f.sendRecvErrorConfig = sendRecvErrorNever
  332. }
  333. }
  334. f.l.WithField("sendRecvError", f.sendRecvErrorConfig.String()).
  335. Info("Loaded send_recv_error config")
  336. }
  337. }
  338. func (f *Interface) reloadMisc(c *config.C) {
  339. if c.HasChanged("counters.try_promote") {
  340. n := c.GetUint32("counters.try_promote", defaultPromoteEvery)
  341. f.tryPromoteEvery.Store(n)
  342. f.l.Info("counters.try_promote has changed")
  343. }
  344. if c.HasChanged("counters.requery_every_packets") {
  345. n := c.GetUint32("counters.requery_every_packets", defaultReQueryEvery)
  346. f.reQueryEvery.Store(n)
  347. f.l.Info("counters.requery_every_packets has changed")
  348. }
  349. if c.HasChanged("timers.requery_wait_duration") {
  350. n := c.GetDuration("timers.requery_wait_duration", defaultReQueryWait)
  351. f.reQueryWait.Store(int64(n))
  352. f.l.Info("timers.requery_wait_duration has changed")
  353. }
  354. }
  355. func (f *Interface) emitStats(ctx context.Context, i time.Duration) {
  356. ticker := time.NewTicker(i)
  357. defer ticker.Stop()
  358. udpStats := udp.NewUDPStatsEmitter(f.writers)
  359. certExpirationGauge := metrics.GetOrRegisterGauge("certificate.ttl_seconds", nil)
  360. certInitiatingVersion := metrics.GetOrRegisterGauge("certificate.initiating_version", nil)
  361. certMaxVersion := metrics.GetOrRegisterGauge("certificate.max_version", nil)
  362. for {
  363. select {
  364. case <-ctx.Done():
  365. return
  366. case <-ticker.C:
  367. f.firewall.EmitStats()
  368. f.handshakeManager.EmitStats()
  369. udpStats()
  370. certState := f.pki.getCertState()
  371. defaultCrt := certState.GetDefaultCertificate()
  372. certExpirationGauge.Update(int64(defaultCrt.NotAfter().Sub(time.Now()) / time.Second))
  373. certInitiatingVersion.Update(int64(defaultCrt.Version()))
  374. // Report the max certificate version we are capable of using
  375. if certState.v2Cert != nil {
  376. certMaxVersion.Update(int64(certState.v2Cert.Version()))
  377. } else {
  378. certMaxVersion.Update(int64(certState.v1Cert.Version()))
  379. }
  380. }
  381. }
  382. }
  383. func (f *Interface) GetHostInfo(vpnIp netip.Addr) *HostInfo {
  384. return f.hostMap.QueryVpnAddr(vpnIp)
  385. }
  386. func (f *Interface) GetCertState() *CertState {
  387. return f.pki.getCertState()
  388. }
  389. func (f *Interface) Close() error {
  390. f.closed.Store(true)
  391. for _, u := range f.writers {
  392. err := u.Close()
  393. if err != nil {
  394. f.l.WithError(err).Error("Error while closing udp socket")
  395. }
  396. }
  397. // Release the tun device
  398. return f.inside.Close()
  399. }