main.go 11 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350
  1. package nebula
  2. import (
  3. "context"
  4. "encoding/binary"
  5. "fmt"
  6. "net"
  7. "time"
  8. "github.com/sirupsen/logrus"
  9. "github.com/slackhq/nebula/config"
  10. "github.com/slackhq/nebula/overlay"
  11. "github.com/slackhq/nebula/sshd"
  12. "github.com/slackhq/nebula/udp"
  13. "github.com/slackhq/nebula/util"
  14. "gopkg.in/yaml.v2"
  15. )
  16. type m map[string]interface{}
  17. func Main(c *config.C, configTest bool, buildVersion string, logger *logrus.Logger, deviceFactory overlay.DeviceFactory) (retcon *Control, reterr error) {
  18. ctx, cancel := context.WithCancel(context.Background())
  19. // Automatically cancel the context if Main returns an error, to signal all created goroutines to quit.
  20. defer func() {
  21. if reterr != nil {
  22. cancel()
  23. }
  24. }()
  25. l := logger
  26. l.Formatter = &logrus.TextFormatter{
  27. FullTimestamp: true,
  28. }
  29. // Print the config if in test, the exit comes later
  30. if configTest {
  31. b, err := yaml.Marshal(c.Settings)
  32. if err != nil {
  33. return nil, err
  34. }
  35. // Print the final config
  36. l.Println(string(b))
  37. }
  38. err := configLogger(l, c)
  39. if err != nil {
  40. return nil, util.ContextualizeIfNeeded("Failed to configure the logger", err)
  41. }
  42. c.RegisterReloadCallback(func(c *config.C) {
  43. err := configLogger(l, c)
  44. if err != nil {
  45. l.WithError(err).Error("Failed to configure the logger")
  46. }
  47. })
  48. pki, err := NewPKIFromConfig(l, c)
  49. if err != nil {
  50. return nil, util.ContextualizeIfNeeded("Failed to load PKI from config", err)
  51. }
  52. certificate := pki.GetCertState().Certificate
  53. fw, err := NewFirewallFromConfig(l, certificate, c)
  54. if err != nil {
  55. return nil, util.ContextualizeIfNeeded("Error while loading firewall rules", err)
  56. }
  57. l.WithField("firewallHashes", fw.GetRuleHashes()).Info("Firewall started")
  58. // TODO: make sure mask is 4 bytes
  59. tunCidr := certificate.Details.Ips[0]
  60. ssh, err := sshd.NewSSHServer(l.WithField("subsystem", "sshd"))
  61. if err != nil {
  62. return nil, util.ContextualizeIfNeeded("Error while creating SSH server", err)
  63. }
  64. wireSSHReload(l, ssh, c)
  65. var sshStart func()
  66. if c.GetBool("sshd.enabled", false) {
  67. sshStart, err = configSSH(l, ssh, c)
  68. if err != nil {
  69. return nil, util.ContextualizeIfNeeded("Error while configuring the sshd", err)
  70. }
  71. }
  72. ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
  73. // All non system modifying configuration consumption should live above this line
  74. // tun config, listeners, anything modifying the computer should be below
  75. ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
  76. var routines int
  77. // If `routines` is set, use that and ignore the specific values
  78. if routines = c.GetInt("routines", 0); routines != 0 {
  79. if routines < 1 {
  80. routines = 1
  81. }
  82. if routines > 1 {
  83. l.WithField("routines", routines).Info("Using multiple routines")
  84. }
  85. } else {
  86. // deprecated and undocumented
  87. tunQueues := c.GetInt("tun.routines", 1)
  88. udpQueues := c.GetInt("listen.routines", 1)
  89. if tunQueues > udpQueues {
  90. routines = tunQueues
  91. } else {
  92. routines = udpQueues
  93. }
  94. if routines != 1 {
  95. l.WithField("routines", routines).Warn("Setting tun.routines and listen.routines is deprecated. Use `routines` instead")
  96. }
  97. }
  98. // EXPERIMENTAL
  99. // Intentionally not documented yet while we do more testing and determine
  100. // a good default value.
  101. conntrackCacheTimeout := c.GetDuration("firewall.conntrack.routine_cache_timeout", 0)
  102. if routines > 1 && !c.IsSet("firewall.conntrack.routine_cache_timeout") {
  103. // Use a different default if we are running with multiple routines
  104. conntrackCacheTimeout = 1 * time.Second
  105. }
  106. if conntrackCacheTimeout > 0 {
  107. l.WithField("duration", conntrackCacheTimeout).Info("Using routine-local conntrack cache")
  108. }
  109. var tun overlay.Device
  110. if !configTest {
  111. c.CatchHUP(ctx)
  112. if deviceFactory == nil {
  113. deviceFactory = overlay.NewDeviceFromConfig
  114. }
  115. tun, err = deviceFactory(c, l, tunCidr, routines)
  116. if err != nil {
  117. return nil, util.ContextualizeIfNeeded("Failed to get a tun/tap device", err)
  118. }
  119. defer func() {
  120. if reterr != nil {
  121. tun.Close()
  122. }
  123. }()
  124. }
  125. // set up our UDP listener
  126. udpConns := make([]udp.Conn, routines)
  127. port := c.GetInt("listen.port", 0)
  128. if !configTest {
  129. rawListenHost := c.GetString("listen.host", "0.0.0.0")
  130. var listenHost *net.IPAddr
  131. if rawListenHost == "[::]" {
  132. // Old guidance was to provide the literal `[::]` in `listen.host` but that won't resolve.
  133. listenHost = &net.IPAddr{IP: net.IPv6zero}
  134. } else {
  135. listenHost, err = net.ResolveIPAddr("ip", rawListenHost)
  136. if err != nil {
  137. return nil, util.ContextualizeIfNeeded("Failed to resolve listen.host", err)
  138. }
  139. }
  140. for i := 0; i < routines; i++ {
  141. l.Infof("listening %q %d", listenHost.IP, port)
  142. udpServer, err := udp.NewListener(l, listenHost.IP, port, routines > 1, c.GetInt("listen.batch", 64))
  143. if err != nil {
  144. return nil, util.NewContextualError("Failed to open udp listener", m{"queue": i}, err)
  145. }
  146. udpServer.ReloadConfig(c)
  147. udpConns[i] = udpServer
  148. }
  149. }
  150. // Set up my internal host map
  151. var preferredRanges []*net.IPNet
  152. rawPreferredRanges := c.GetStringSlice("preferred_ranges", []string{})
  153. // First, check if 'preferred_ranges' is set and fallback to 'local_range'
  154. if len(rawPreferredRanges) > 0 {
  155. for _, rawPreferredRange := range rawPreferredRanges {
  156. _, preferredRange, err := net.ParseCIDR(rawPreferredRange)
  157. if err != nil {
  158. return nil, util.ContextualizeIfNeeded("Failed to parse preferred ranges", err)
  159. }
  160. preferredRanges = append(preferredRanges, preferredRange)
  161. }
  162. }
  163. // local_range was superseded by preferred_ranges. If it is still present,
  164. // merge the local_range setting into preferred_ranges. We will probably
  165. // deprecate local_range and remove in the future.
  166. rawLocalRange := c.GetString("local_range", "")
  167. if rawLocalRange != "" {
  168. _, localRange, err := net.ParseCIDR(rawLocalRange)
  169. if err != nil {
  170. return nil, util.ContextualizeIfNeeded("Failed to parse local_range", err)
  171. }
  172. // Check if the entry for local_range was already specified in
  173. // preferred_ranges. Don't put it into the slice twice if so.
  174. var found bool
  175. for _, r := range preferredRanges {
  176. if r.String() == localRange.String() {
  177. found = true
  178. break
  179. }
  180. }
  181. if !found {
  182. preferredRanges = append(preferredRanges, localRange)
  183. }
  184. }
  185. hostMap := NewHostMap(l, tunCidr, preferredRanges)
  186. hostMap.metricsEnabled = c.GetBool("stats.message_metrics", false)
  187. l.
  188. WithField("network", hostMap.vpnCIDR.String()).
  189. WithField("preferredRanges", hostMap.preferredRanges).
  190. Info("Main HostMap created")
  191. punchy := NewPunchyFromConfig(l, c)
  192. lightHouse, err := NewLightHouseFromConfig(ctx, l, c, tunCidr, udpConns[0], punchy)
  193. if err != nil {
  194. return nil, util.ContextualizeIfNeeded("Failed to initialize lighthouse handler", err)
  195. }
  196. var messageMetrics *MessageMetrics
  197. if c.GetBool("stats.message_metrics", false) {
  198. messageMetrics = newMessageMetrics()
  199. } else {
  200. messageMetrics = newMessageMetricsOnlyRecvError()
  201. }
  202. useRelays := c.GetBool("relay.use_relays", DefaultUseRelays) && !c.GetBool("relay.am_relay", false)
  203. handshakeConfig := HandshakeConfig{
  204. tryInterval: c.GetDuration("handshakes.try_interval", DefaultHandshakeTryInterval),
  205. retries: c.GetInt("handshakes.retries", DefaultHandshakeRetries),
  206. triggerBuffer: c.GetInt("handshakes.trigger_buffer", DefaultHandshakeTriggerBuffer),
  207. useRelays: useRelays,
  208. messageMetrics: messageMetrics,
  209. }
  210. handshakeManager := NewHandshakeManager(l, hostMap, lightHouse, udpConns[0], handshakeConfig)
  211. lightHouse.handshakeTrigger = handshakeManager.trigger
  212. serveDns := false
  213. if c.GetBool("lighthouse.serve_dns", false) {
  214. if c.GetBool("lighthouse.am_lighthouse", false) {
  215. serveDns = true
  216. } else {
  217. l.Warn("DNS server refusing to run because this host is not a lighthouse.")
  218. }
  219. }
  220. checkInterval := c.GetInt("timers.connection_alive_interval", 5)
  221. pendingDeletionInterval := c.GetInt("timers.pending_deletion_interval", 10)
  222. ifConfig := &InterfaceConfig{
  223. HostMap: hostMap,
  224. Inside: tun,
  225. Outside: udpConns[0],
  226. pki: pki,
  227. Cipher: c.GetString("cipher", "aes"),
  228. Firewall: fw,
  229. ServeDns: serveDns,
  230. HandshakeManager: handshakeManager,
  231. lightHouse: lightHouse,
  232. checkInterval: time.Second * time.Duration(checkInterval),
  233. pendingDeletionInterval: time.Second * time.Duration(pendingDeletionInterval),
  234. tryPromoteEvery: c.GetUint32("counters.try_promote", defaultPromoteEvery),
  235. reQueryEvery: c.GetUint32("counters.requery_every_packets", defaultReQueryEvery),
  236. reQueryWait: c.GetDuration("timers.requery_wait_duration", defaultReQueryWait),
  237. DropLocalBroadcast: c.GetBool("tun.drop_local_broadcast", false),
  238. DropMulticast: c.GetBool("tun.drop_multicast", false),
  239. routines: routines,
  240. MessageMetrics: messageMetrics,
  241. version: buildVersion,
  242. relayManager: NewRelayManager(ctx, l, hostMap, c),
  243. punchy: punchy,
  244. ConntrackCacheTimeout: conntrackCacheTimeout,
  245. l: l,
  246. }
  247. switch ifConfig.Cipher {
  248. case "aes":
  249. noiseEndianness = binary.BigEndian
  250. case "chachapoly":
  251. noiseEndianness = binary.LittleEndian
  252. default:
  253. return nil, fmt.Errorf("unknown cipher: %v", ifConfig.Cipher)
  254. }
  255. var ifce *Interface
  256. if !configTest {
  257. ifce, err = NewInterface(ctx, ifConfig)
  258. if err != nil {
  259. return nil, fmt.Errorf("failed to initialize interface: %s", err)
  260. }
  261. // TODO: Better way to attach these, probably want a new interface in InterfaceConfig
  262. // I don't want to make this initial commit too far-reaching though
  263. ifce.writers = udpConns
  264. lightHouse.ifce = ifce
  265. ifce.RegisterConfigChangeCallbacks(c)
  266. ifce.reloadDisconnectInvalid(c)
  267. ifce.reloadSendRecvError(c)
  268. handshakeManager.f = ifce
  269. go handshakeManager.Run(ctx)
  270. }
  271. // TODO - stats third-party modules start uncancellable goroutines. Update those libs to accept
  272. // a context so that they can exit when the context is Done.
  273. statsStart, err := startStats(l, c, buildVersion, configTest)
  274. if err != nil {
  275. return nil, util.ContextualizeIfNeeded("Failed to start stats emitter", err)
  276. }
  277. if configTest {
  278. return nil, nil
  279. }
  280. //TODO: check if we _should_ be emitting stats
  281. go ifce.emitStats(ctx, c.GetDuration("stats.interval", time.Second*10))
  282. attachCommands(l, c, ssh, ifce)
  283. // Start DNS server last to allow using the nebula IP as lighthouse.dns.host
  284. var dnsStart func()
  285. if lightHouse.amLighthouse && serveDns {
  286. l.Debugln("Starting dns server")
  287. dnsStart = dnsMain(l, hostMap, c)
  288. }
  289. return &Control{
  290. ifce,
  291. l,
  292. ctx,
  293. cancel,
  294. sshStart,
  295. statsStart,
  296. dnsStart,
  297. lightHouse.StartUpdateWorker,
  298. }, nil
  299. }