daemon.go 14 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437
  1. package functions
  2. import (
  3. "context"
  4. "encoding/json"
  5. "fmt"
  6. "log"
  7. "os"
  8. "os/signal"
  9. "runtime"
  10. "sync"
  11. "syscall"
  12. "time"
  13. mqtt "github.com/eclipse/paho.mqtt.golang"
  14. "github.com/gravitl/netmaker/logger"
  15. "github.com/gravitl/netmaker/models"
  16. "github.com/gravitl/netmaker/netclient/config"
  17. "github.com/gravitl/netmaker/netclient/local"
  18. "github.com/gravitl/netmaker/netclient/ncutils"
  19. "github.com/gravitl/netmaker/netclient/wireguard"
  20. "golang.zx2c4.com/wireguard/wgctrl/wgtypes"
  21. )
  22. // ServerKeepalive - stores time of last server keepalive message
  23. var keepalive chan string = make(chan string)
  24. var messageCache = make(map[string]string, 20)
  25. const lastNodeUpdate = "lnu"
  26. const lastPeerUpdate = "lpu"
  27. func insert(network, which, cache string) {
  28. var mu sync.Mutex
  29. mu.Lock()
  30. defer mu.Unlock()
  31. messageCache[fmt.Sprintf("%s%s", network, which)] = cache
  32. }
  33. func read(network, which string) string {
  34. return messageCache[fmt.Sprintf("%s%s", network, which)]
  35. }
  36. // Daemon runs netclient daemon from command line
  37. func Daemon() error {
  38. ctx, cancel := context.WithCancel(context.Background())
  39. networks, err := ncutils.GetSystemNetworks()
  40. if err != nil {
  41. cancel()
  42. return err
  43. }
  44. for _, network := range networks {
  45. go MessageQueue(ctx, network)
  46. }
  47. quit := make(chan os.Signal, 1)
  48. signal.Notify(quit, syscall.SIGTERM, os.Interrupt)
  49. <-quit
  50. cancel()
  51. ncutils.Log("all done")
  52. return nil
  53. }
  54. // SetupMQTT creates a connection to broker and return client
  55. func SetupMQTT(cfg *config.ClientConfig) mqtt.Client {
  56. opts := mqtt.NewClientOptions()
  57. for _, server := range cfg.Node.NetworkSettings.DefaultServerAddrs {
  58. if server.Address != "" && server.IsLeader {
  59. ncutils.Log(fmt.Sprintf("adding server (%s) to listen on network %s \n", server.Address, cfg.Node.Network))
  60. opts.AddBroker(server.Address + ":1883")
  61. break
  62. }
  63. }
  64. opts.SetDefaultPublishHandler(All)
  65. client := mqtt.NewClient(opts)
  66. tperiod := time.Now().Add(10 * time.Second)
  67. for {
  68. if token := client.Connect(); token.Wait() && token.Error() != nil {
  69. logger.Log(2, "unable to connect to broker, retrying ...")
  70. if time.Now().Afer(tperiod) {
  71. log.Fatal(0, "could not connect to broker, exiting ...", token.Error())
  72. }
  73. } else {
  74. break
  75. }
  76. time.Sleep(2 * time.Second)
  77. }
  78. return client
  79. }
  80. // MessageQueue sets up Message Queue and subsribes/publishes updates to/from server
  81. func MessageQueue(ctx context.Context, network string) {
  82. ncutils.Log("netclient go routine started for " + network)
  83. var cfg config.ClientConfig
  84. cfg.Network = network
  85. cfg.ReadConfig()
  86. ncutils.Log("daemon started for network:" + network)
  87. client := SetupMQTT(&cfg)
  88. if cfg.DebugOn {
  89. if token := client.Subscribe("#", 0, nil); token.Wait() && token.Error() != nil {
  90. log.Fatal(token.Error())
  91. }
  92. ncutils.Log("subscribed to all topics for debugging purposes")
  93. }
  94. if token := client.Subscribe("update/"+cfg.Node.ID, 0, mqtt.MessageHandler(NodeUpdate)); token.Wait() && token.Error() != nil {
  95. log.Fatal(token.Error())
  96. }
  97. if cfg.DebugOn {
  98. ncutils.Log("subscribed to node updates for node " + cfg.Node.Name + " update/" + cfg.Node.ID)
  99. }
  100. if token := client.Subscribe("update/peers/"+cfg.Node.ID, 0, mqtt.MessageHandler(UpdatePeers)); token.Wait() && token.Error() != nil {
  101. log.Fatal(token.Error())
  102. }
  103. if cfg.DebugOn {
  104. ncutils.Log("subscribed to node updates for node " + cfg.Node.Name + " update/peers/" + cfg.Node.ID)
  105. }
  106. var id string
  107. for _, server := range cfg.NetworkSettings.DefaultServerAddrs {
  108. if server.IsLeader {
  109. id = server.ID
  110. }
  111. if server.Address != "" {
  112. if token := client.Subscribe("serverkeepalive/"+id, 0, mqtt.MessageHandler(ServerKeepAlive)); token.Wait() && token.Error() != nil {
  113. log.Fatal(token.Error())
  114. }
  115. if cfg.DebugOn {
  116. ncutils.Log("subscribed to server keepalives")
  117. }
  118. } else {
  119. ncutils.Log("leader not defined for network" + cfg.Network)
  120. }
  121. }
  122. defer client.Disconnect(250)
  123. go MonitorKeepalive(ctx, client, &cfg)
  124. go Checkin(ctx, &cfg, network)
  125. <-ctx.Done()
  126. ncutils.Log("shutting down daemon")
  127. }
  128. // All -- mqtt message hander for all ('#') topics
  129. var All mqtt.MessageHandler = func(client mqtt.Client, msg mqtt.Message) {
  130. ncutils.Log("default message handler -- received message but not handling")
  131. ncutils.Log("Topic: " + string(msg.Topic()))
  132. //ncutils.Log("Message: " + string(msg.Payload()))
  133. }
  134. // NodeUpdate -- mqtt message handler for /update/<NodeID> topic
  135. func NodeUpdate(client mqtt.Client, msg mqtt.Message) {
  136. //potentiall blocking i/o so do this in a go routine
  137. go func() {
  138. var newNode models.Node
  139. var cfg config.ClientConfig
  140. err := json.Unmarshal(msg.Payload(), &newNode)
  141. if err != nil {
  142. ncutils.Log("error unmarshalling node update data" + err.Error())
  143. return
  144. }
  145. ncutils.Log("received message to update node " + newNode.Name)
  146. // see if cache hit, if so skip
  147. var currentMessage = read(newNode.Network, lastNodeUpdate)
  148. if currentMessage == string(msg.Payload()) {
  149. return
  150. }
  151. insert(newNode.Network, lastNodeUpdate, string(msg.Payload()))
  152. cfg.Network = newNode.Network
  153. cfg.ReadConfig()
  154. //check if interface name has changed if so delete.
  155. if cfg.Node.Interface != newNode.Interface {
  156. if err = wireguard.RemoveConf(cfg.Node.Interface, true); err != nil {
  157. ncutils.PrintLog("could not delete old interface "+cfg.Node.Interface+": "+err.Error(), 1)
  158. }
  159. }
  160. newNode.PullChanges = "no"
  161. //ensure that OS never changes
  162. newNode.OS = runtime.GOOS
  163. cfg.Node = newNode
  164. switch newNode.Action {
  165. case models.NODE_DELETE:
  166. if err := RemoveLocalInstance(&cfg, cfg.Network); err != nil {
  167. ncutils.PrintLog("error deleting local instance: "+err.Error(), 1)
  168. return
  169. }
  170. if token := client.Unsubscribe("update/"+newNode.ID, "update/peers/"+newNode.ID); token.Wait() && token.Error() != nil {
  171. ncutils.PrintLog("error unsubscribing during node deletion", 1)
  172. }
  173. return
  174. case models.NODE_UPDATE_KEY:
  175. if err := UpdateKeys(&cfg, client); err != nil {
  176. ncutils.PrintLog("err updating wireguard keys: "+err.Error(), 1)
  177. }
  178. case models.NODE_NOOP:
  179. default:
  180. }
  181. //Save new config
  182. if err := config.Write(&cfg, cfg.Network); err != nil {
  183. ncutils.PrintLog("error updating node configuration: "+err.Error(), 1)
  184. }
  185. nameserver := cfg.Server.CoreDNSAddr
  186. privateKey, err := wireguard.RetrievePrivKey(newNode.Network)
  187. if err != nil {
  188. ncutils.Log("error reading PrivateKey " + err.Error())
  189. return
  190. }
  191. file := ncutils.GetNetclientPathSpecific() + cfg.Node.Interface + ".conf"
  192. if err := wireguard.UpdateWgInterface(file, privateKey, nameserver, newNode); err != nil {
  193. ncutils.Log("error updating wireguard config " + err.Error())
  194. return
  195. }
  196. ncutils.Log("applyWGQuickConf to " + file)
  197. err = wireguard.ApplyWGQuickConf(file)
  198. if err != nil {
  199. ncutils.Log("error restarting wg after node update " + err.Error())
  200. return
  201. }
  202. //deal with DNS
  203. if newNode.DNSOn == "yes" {
  204. ncutils.Log("setting up DNS")
  205. if err = local.UpdateDNS(cfg.Node.Interface, cfg.Network, cfg.Server.CoreDNSAddr); err != nil {
  206. ncutils.Log("error applying dns" + err.Error())
  207. }
  208. } else {
  209. ncutils.Log("settng DNS off")
  210. _, err := ncutils.RunCmd("/usr/bin/resolvectl revert "+cfg.Node.Interface, true)
  211. if err != nil {
  212. ncutils.Log("error applying dns" + err.Error())
  213. }
  214. }
  215. }()
  216. }
  217. // UpdatePeers -- mqtt message handler for /update/peers/<NodeID> topic
  218. func UpdatePeers(client mqtt.Client, msg mqtt.Message) {
  219. go func() {
  220. var peerUpdate models.PeerUpdate
  221. err := json.Unmarshal(msg.Payload(), &peerUpdate)
  222. if err != nil {
  223. ncutils.Log("error unmarshalling peer data")
  224. return
  225. }
  226. // see if cache hit, if so skip
  227. var currentMessage = read(peerUpdate.Network, lastPeerUpdate)
  228. if currentMessage == string(msg.Payload()) {
  229. return
  230. }
  231. insert(peerUpdate.Network, lastPeerUpdate, string(msg.Payload()))
  232. ncutils.Log("update peer handler")
  233. var cfg config.ClientConfig
  234. cfg.Network = peerUpdate.Network
  235. cfg.ReadConfig()
  236. var shouldReSub = shouldResub(cfg.Node.NetworkSettings.DefaultServerAddrs, peerUpdate.ServerAddrs)
  237. if shouldReSub {
  238. if err := config.ModConfig(&cfg.Node); err == nil {
  239. Resubscribe(client, &cfg)
  240. cfg.Node.NetworkSettings.DefaultServerAddrs = peerUpdate.ServerAddrs
  241. } else {
  242. ncutils.Log("resub required but mod config failed")
  243. }
  244. file := ncutils.GetNetclientPathSpecific() + cfg.Node.Interface + ".conf"
  245. err = wireguard.UpdateWgPeers(file, peerUpdate.Peers)
  246. if err != nil {
  247. ncutils.Log("error updating wireguard peers" + err.Error())
  248. return
  249. }
  250. ncutils.Log("applyWGQuickConf to " + file)
  251. err = wireguard.ApplyWGQuickConf(file)
  252. if err != nil {
  253. ncutils.Log("error restarting wg after peer update " + err.Error())
  254. return
  255. }
  256. }
  257. }()
  258. }
  259. // MonitorKeepalive - checks time last server keepalive received. If more than 3+ minutes, notify and resubscribe
  260. func MonitorKeepalive(ctx context.Context, client mqtt.Client, cfg *config.ClientConfig) {
  261. var id string
  262. lastUpdate := time.Now()
  263. for _, servAddr := range cfg.NetworkSettings.DefaultServerAddrs {
  264. if servAddr.IsLeader {
  265. id = servAddr.ID
  266. }
  267. }
  268. for {
  269. select {
  270. case <-ctx.Done():
  271. return
  272. case <-time.After(time.Second * 150):
  273. if time.Since(lastUpdate) > time.Second*200 { // more than 3+ minutes
  274. Resubscribe(client, cfg)
  275. }
  276. case serverID := <-keepalive:
  277. if serverID != id {
  278. // not my server, put back
  279. if cfg.DebugOn {
  280. ncutils.Log("Monitor keepalive received wrong message on channel, putting back")
  281. }
  282. keepalive <- serverID
  283. } else {
  284. lastUpdate = time.Now()
  285. if cfg.DebugOn {
  286. ncutils.Log("updating checking time for" + cfg.Network)
  287. }
  288. }
  289. }
  290. }
  291. }
  292. // ServerKeepAlive -- handler to react to keepalive messages published by server
  293. func ServerKeepAlive(client mqtt.Client, msg mqtt.Message) {
  294. serverid := string(msg.Payload())
  295. keepalive <- serverid
  296. ncutils.Log("keepalive from server")
  297. }
  298. // Resubscribe --- handles resubscribing if needed
  299. func Resubscribe(client mqtt.Client, cfg *config.ClientConfig) error {
  300. ncutils.Log("resubbing on network " + cfg.Node.Network)
  301. client.Disconnect(250)
  302. client = SetupMQTT(cfg)
  303. if token := client.Subscribe("update/"+cfg.Node.ID, 0, NodeUpdate); token.Wait() && token.Error() != nil {
  304. log.Fatal(token.Error())
  305. }
  306. if cfg.DebugOn {
  307. ncutils.Log("subscribed to node updates for node " + cfg.Node.Name + " update/" + cfg.Node.ID)
  308. }
  309. if token := client.Subscribe("update/peers/"+cfg.Node.ID, 0, UpdatePeers); token.Wait() && token.Error() != nil {
  310. log.Fatal(token.Error())
  311. }
  312. ncutils.Log("finished re subbing")
  313. return nil
  314. }
  315. // UpdateKeys -- updates private key and returns new publickey
  316. func UpdateKeys(cfg *config.ClientConfig, client mqtt.Client) error {
  317. ncutils.Log("received message to update keys")
  318. //potentiall blocking i/o so do this in a go routine
  319. key, err := wgtypes.GeneratePrivateKey()
  320. if err != nil {
  321. ncutils.Log("error generating privatekey " + err.Error())
  322. return err
  323. }
  324. file := ncutils.GetNetclientPathSpecific() + cfg.Node.Interface + ".conf"
  325. if err := wireguard.UpdatePrivateKey(file, key.String()); err != nil {
  326. ncutils.Log("error updating wireguard key " + err.Error())
  327. return err
  328. }
  329. cfg.Node.PublicKey = key.PublicKey().String()
  330. PublishNodeUpdate(cfg)
  331. if err := config.ModConfig(&cfg.Node); err != nil {
  332. ncutils.Log("error updating local config " + err.Error())
  333. }
  334. return nil
  335. }
  336. // Checkin -- go routine that checks for public or local ip changes, publishes changes
  337. // if there are no updates, simply "pings" the server as a checkin
  338. func Checkin(ctx context.Context, cfg *config.ClientConfig, network string) {
  339. for {
  340. select {
  341. case <-ctx.Done():
  342. ncutils.Log("Checkin cancelled")
  343. return
  344. //delay should be configuraable -> use cfg.Node.NetworkSettings.DefaultCheckInInterval ??
  345. case <-time.After(time.Second * 60):
  346. ncutils.Log("Checkin running")
  347. //read latest config
  348. cfg.ReadConfig()
  349. if cfg.Node.Roaming == "yes" && cfg.Node.IsStatic != "yes" {
  350. extIP, err := ncutils.GetPublicIP()
  351. if err != nil {
  352. ncutils.PrintLog("error encountered checking ip addresses: "+err.Error(), 1)
  353. }
  354. if cfg.Node.Endpoint != extIP && extIP != "" {
  355. ncutils.PrintLog("endpoint has changed from "+cfg.Node.Endpoint+" to "+extIP, 1)
  356. cfg.Node.Endpoint = extIP
  357. PublishNodeUpdate(cfg)
  358. }
  359. intIP, err := getPrivateAddr()
  360. if err != nil {
  361. ncutils.PrintLog("error encountered checking ip addresses: "+err.Error(), 1)
  362. }
  363. if cfg.Node.LocalAddress != intIP && intIP != "" {
  364. ncutils.PrintLog("local Address has changed from "+cfg.Node.LocalAddress+" to "+intIP, 1)
  365. cfg.Node.LocalAddress = intIP
  366. PublishNodeUpdate(cfg)
  367. }
  368. } else {
  369. localIP, err := ncutils.GetLocalIP(cfg.Node.LocalRange)
  370. if err != nil {
  371. ncutils.PrintLog("error encountered checking ip addresses: "+err.Error(), 1)
  372. }
  373. if cfg.Node.Endpoint != localIP && localIP != "" {
  374. ncutils.PrintLog("endpoint has changed from "+cfg.Node.Endpoint+" to "+localIP, 1)
  375. cfg.Node.Endpoint = localIP
  376. PublishNodeUpdate(cfg)
  377. }
  378. }
  379. Hello(cfg, network)
  380. ncutils.Log("Checkin complete")
  381. }
  382. }
  383. }
  384. // PublishNodeUpdates -- saves node and pushes changes to broker
  385. func PublishNodeUpdate(cfg *config.ClientConfig) {
  386. if err := config.Write(cfg, cfg.Network); err != nil {
  387. ncutils.Log("error saving configuration" + err.Error())
  388. }
  389. client := SetupMQTT(cfg)
  390. data, err := json.Marshal(cfg.Node)
  391. if err != nil {
  392. ncutils.Log("error marshling node update " + err.Error())
  393. }
  394. if token := client.Publish("update/"+cfg.Node.ID, 0, false, data); token.Wait() && token.Error() != nil {
  395. ncutils.Log("error publishing endpoint update " + token.Error().Error())
  396. }
  397. client.Disconnect(250)
  398. }
  399. // Hello -- ping the broker to let server know node is alive and doing fine
  400. func Hello(cfg *config.ClientConfig, network string) {
  401. client := SetupMQTT(cfg)
  402. if token := client.Publish("ping/"+cfg.Node.ID, 2, false, "hello world!"); token.Wait() && token.Error() != nil {
  403. ncutils.Log("error publishing ping " + token.Error().Error())
  404. }
  405. client.Disconnect(250)
  406. }
  407. func shouldResub(currentServers, newServers []models.ServerAddr) bool {
  408. if len(currentServers) != len(newServers) {
  409. return true
  410. }
  411. for _, srv := range currentServers {
  412. if !ncutils.ServerAddrSliceContains(newServers, srv) {
  413. return true
  414. }
  415. }
  416. return false
  417. }