daemon.go 15 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480
  1. package functions
  2. import (
  3. "context"
  4. "encoding/json"
  5. "fmt"
  6. "log"
  7. "os"
  8. "os/signal"
  9. "runtime"
  10. "strings"
  11. "sync"
  12. "syscall"
  13. "time"
  14. mqtt "github.com/eclipse/paho.mqtt.golang"
  15. "github.com/gravitl/netmaker/logger"
  16. "github.com/gravitl/netmaker/models"
  17. "github.com/gravitl/netmaker/netclient/auth"
  18. "github.com/gravitl/netmaker/netclient/config"
  19. "github.com/gravitl/netmaker/netclient/local"
  20. "github.com/gravitl/netmaker/netclient/ncutils"
  21. "github.com/gravitl/netmaker/netclient/wireguard"
  22. "golang.zx2c4.com/wireguard/wgctrl/wgtypes"
  23. )
  24. // ServerKeepalive - stores time of last server keepalive message
  25. var keepalive = make(map[string]time.Time, 3)
  26. var messageCache = make(map[string]string, 20)
  27. const lastNodeUpdate = "lnu"
  28. const lastPeerUpdate = "lpu"
  29. func insert(network, which, cache string) {
  30. var mu sync.Mutex
  31. mu.Lock()
  32. defer mu.Unlock()
  33. messageCache[fmt.Sprintf("%s%s", network, which)] = cache
  34. }
  35. func read(network, which string) string {
  36. return messageCache[fmt.Sprintf("%s%s", network, which)]
  37. }
  38. // Daemon runs netclient daemon from command line
  39. func Daemon() error {
  40. ctx, cancel := context.WithCancel(context.Background())
  41. networks, err := ncutils.GetSystemNetworks()
  42. if err != nil {
  43. cancel()
  44. return err
  45. }
  46. for _, network := range networks {
  47. go MessageQueue(ctx, network)
  48. }
  49. quit := make(chan os.Signal, 1)
  50. signal.Notify(quit, syscall.SIGTERM, os.Interrupt)
  51. <-quit
  52. cancel()
  53. ncutils.Log("all done")
  54. return nil
  55. }
  56. // SetupMQTT creates a connection to broker and return client
  57. func SetupMQTT(cfg *config.ClientConfig) mqtt.Client {
  58. opts := mqtt.NewClientOptions()
  59. for _, server := range cfg.Node.NetworkSettings.DefaultServerAddrs {
  60. if server.Address != "" && server.IsLeader {
  61. ncutils.Log(fmt.Sprintf("adding server (%s) to listen on network %s \n", server.Address, cfg.Node.Network))
  62. opts.AddBroker(server.Address + ":1883")
  63. break
  64. }
  65. }
  66. opts.SetDefaultPublishHandler(All)
  67. client := mqtt.NewClient(opts)
  68. tperiod := time.Now().Add(10 * time.Second)
  69. for {
  70. if token := client.Connect(); token.Wait() && token.Error() != nil {
  71. logger.Log(2, "unable to connect to broker, retrying ...")
  72. if time.Now().After(tperiod) {
  73. log.Fatal(0, "could not connect to broker, exiting ...", token.Error())
  74. }
  75. } else {
  76. break
  77. }
  78. time.Sleep(2 * time.Second)
  79. }
  80. return client
  81. }
  82. // MessageQueue sets up Message Queue and subsribes/publishes updates to/from server
  83. func MessageQueue(ctx context.Context, network string) {
  84. ncutils.Log("netclient go routine started for " + network)
  85. var cfg config.ClientConfig
  86. cfg.Network = network
  87. cfg.ReadConfig()
  88. ncutils.Log("daemon started for network:" + network)
  89. client := SetupMQTT(&cfg)
  90. if cfg.DebugOn {
  91. if token := client.Subscribe("#", 0, nil); token.Wait() && token.Error() != nil {
  92. log.Fatal(token.Error())
  93. }
  94. ncutils.Log("subscribed to all topics for debugging purposes")
  95. }
  96. if token := client.Subscribe(fmt.Sprintf("update/%s/%s", cfg.Node.Network, cfg.Node.ID), 0, mqtt.MessageHandler(NodeUpdate)); token.Wait() && token.Error() != nil {
  97. log.Fatal(token.Error())
  98. }
  99. if cfg.DebugOn {
  100. ncutils.Log(fmt.Sprintf("subscribed to node updates for node %s update/%s/%s \n", cfg.Node.Name, cfg.Node.Network, cfg.Node.ID))
  101. }
  102. if token := client.Subscribe(fmt.Sprintf("peers/%s/%s", cfg.Node.Network, cfg.Node.ID), 0, mqtt.MessageHandler(UpdatePeers)); token.Wait() && token.Error() != nil {
  103. log.Fatal(token.Error())
  104. }
  105. if cfg.DebugOn {
  106. ncutils.Log(fmt.Sprintf("subscribed to peer updates for node %s peers/%s/%s \n", cfg.Node.Name, cfg.Node.Network, cfg.Node.ID))
  107. }
  108. var id string
  109. for _, server := range cfg.NetworkSettings.DefaultServerAddrs {
  110. if server.IsLeader {
  111. id = server.ID
  112. }
  113. if server.Address != "" {
  114. if token := client.Subscribe("serverkeepalive/"+id, 0, mqtt.MessageHandler(ServerKeepAlive)); token.Wait() && token.Error() != nil {
  115. log.Fatal(token.Error())
  116. }
  117. if cfg.DebugOn {
  118. ncutils.Log("subscribed to server keepalives")
  119. }
  120. } else {
  121. ncutils.Log("leader not defined for network" + cfg.Network)
  122. }
  123. }
  124. defer client.Disconnect(250)
  125. go MonitorKeepalive(ctx, client, &cfg)
  126. go Checkin(ctx, &cfg, network)
  127. <-ctx.Done()
  128. ncutils.Log("shutting down daemon")
  129. }
  130. // All -- mqtt message hander for all ('#') topics
  131. var All mqtt.MessageHandler = func(client mqtt.Client, msg mqtt.Message) {
  132. ncutils.Log("default message handler -- received message but not handling")
  133. ncutils.Log("Topic: " + string(msg.Topic()))
  134. //ncutils.Log("Message: " + string(msg.Payload()))
  135. }
  136. // NodeUpdate -- mqtt message handler for /update/<NodeID> topic
  137. func NodeUpdate(client mqtt.Client, msg mqtt.Message) {
  138. //potentiall blocking i/o so do this in a go routine
  139. go func() {
  140. var newNode models.Node
  141. var cfg config.ClientConfig
  142. var network = parseNetworkFromTopic(msg.Topic())
  143. cfg.Network = network
  144. cfg.ReadConfig()
  145. data, dataErr := decryptMsg(&cfg, msg.Payload())
  146. if dataErr != nil {
  147. return
  148. }
  149. err := json.Unmarshal([]byte(data), &newNode)
  150. if err != nil {
  151. ncutils.Log("error unmarshalling node update data" + err.Error())
  152. return
  153. }
  154. ncutils.Log("received message to update node " + newNode.Name)
  155. // see if cache hit, if so skip
  156. var currentMessage = read(newNode.Network, lastNodeUpdate)
  157. if currentMessage == string(data) {
  158. return
  159. }
  160. insert(newNode.Network, lastNodeUpdate, string(data))
  161. //check if interface name has changed if so delete.
  162. if cfg.Node.Interface != newNode.Interface {
  163. if err = wireguard.RemoveConf(cfg.Node.Interface, true); err != nil {
  164. ncutils.PrintLog("could not delete old interface "+cfg.Node.Interface+": "+err.Error(), 1)
  165. }
  166. }
  167. newNode.PullChanges = "no"
  168. //ensure that OS never changes
  169. newNode.OS = runtime.GOOS
  170. cfg.Node = newNode
  171. switch newNode.Action {
  172. case models.NODE_DELETE:
  173. if err := RemoveLocalInstance(&cfg, cfg.Network); err != nil {
  174. ncutils.PrintLog("error deleting local instance: "+err.Error(), 1)
  175. return
  176. }
  177. if token := client.Unsubscribe("update/"+newNode.ID, "update/peers/"+newNode.ID); token.Wait() && token.Error() != nil {
  178. ncutils.PrintLog("error unsubscribing during node deletion", 1)
  179. }
  180. return
  181. case models.NODE_UPDATE_KEY:
  182. if err := UpdateKeys(&cfg, client); err != nil {
  183. ncutils.PrintLog("err updating wireguard keys: "+err.Error(), 1)
  184. }
  185. case models.NODE_NOOP:
  186. default:
  187. }
  188. //Save new config
  189. if err := config.Write(&cfg, cfg.Network); err != nil {
  190. ncutils.PrintLog("error updating node configuration: "+err.Error(), 1)
  191. }
  192. nameserver := cfg.Server.CoreDNSAddr
  193. privateKey, err := wireguard.RetrievePrivKey(newNode.Network)
  194. if err != nil {
  195. ncutils.Log("error reading PrivateKey " + err.Error())
  196. return
  197. }
  198. file := ncutils.GetNetclientPathSpecific() + cfg.Node.Interface + ".conf"
  199. if err := wireguard.UpdateWgInterface(file, privateKey, nameserver, newNode); err != nil {
  200. ncutils.Log("error updating wireguard config " + err.Error())
  201. return
  202. }
  203. ncutils.Log("applyWGQuickConf to " + file)
  204. err = wireguard.ApplyWGQuickConf(file)
  205. if err != nil {
  206. ncutils.Log("error restarting wg after node update " + err.Error())
  207. return
  208. }
  209. //deal with DNS
  210. if newNode.DNSOn == "yes" {
  211. ncutils.Log("setting up DNS")
  212. if err = local.UpdateDNS(cfg.Node.Interface, cfg.Network, cfg.Server.CoreDNSAddr); err != nil {
  213. ncutils.Log("error applying dns" + err.Error())
  214. }
  215. } else {
  216. ncutils.Log("settng DNS off")
  217. _, err := ncutils.RunCmd("/usr/bin/resolvectl revert "+cfg.Node.Interface, true)
  218. if err != nil {
  219. ncutils.Log("error applying dns" + err.Error())
  220. }
  221. }
  222. }()
  223. }
  224. // UpdatePeers -- mqtt message handler for /update/peers/<NodeID> topic
  225. func UpdatePeers(client mqtt.Client, msg mqtt.Message) {
  226. go func() {
  227. var peerUpdate models.PeerUpdate
  228. var network = parseNetworkFromTopic(msg.Topic())
  229. var cfg = config.ClientConfig{}
  230. cfg.Network = network
  231. cfg.ReadConfig()
  232. data, dataErr := decryptMsg(&cfg, msg.Payload())
  233. if dataErr != nil {
  234. return
  235. }
  236. err := json.Unmarshal([]byte(data), &peerUpdate)
  237. if err != nil {
  238. ncutils.Log("error unmarshalling peer data")
  239. return
  240. }
  241. // see if cache hit, if so skip
  242. var currentMessage = read(peerUpdate.Network, lastPeerUpdate)
  243. if currentMessage == string(data) {
  244. return
  245. }
  246. insert(peerUpdate.Network, lastPeerUpdate, string(data))
  247. ncutils.Log("update peer handler")
  248. var shouldReSub = shouldResub(cfg.Node.NetworkSettings.DefaultServerAddrs, peerUpdate.ServerAddrs)
  249. if shouldReSub {
  250. Resubscribe(client, &cfg)
  251. cfg.Node.NetworkSettings.DefaultServerAddrs = peerUpdate.ServerAddrs
  252. file := ncutils.GetNetclientPathSpecific() + cfg.Node.Interface + ".conf"
  253. err = wireguard.UpdateWgPeers(file, peerUpdate.Peers)
  254. if err != nil {
  255. ncutils.Log("error updating wireguard peers" + err.Error())
  256. return
  257. }
  258. ncutils.Log("applyWGQuickConf to " + file)
  259. err = wireguard.ApplyWGQuickConf(file)
  260. if err != nil {
  261. ncutils.Log("error restarting wg after peer update " + err.Error())
  262. return
  263. }
  264. }
  265. }()
  266. }
  267. // MonitorKeepalive - checks time last server keepalive received. If more than 3+ minutes, notify and resubscribe
  268. func MonitorKeepalive(ctx context.Context, client mqtt.Client, cfg *config.ClientConfig) {
  269. var id string
  270. for _, servAddr := range cfg.NetworkSettings.DefaultServerAddrs {
  271. if servAddr.IsLeader {
  272. id = servAddr.ID
  273. }
  274. }
  275. for {
  276. select {
  277. case <-ctx.Done():
  278. return
  279. case <-time.After(time.Second * 150):
  280. if time.Since(keepalive[id]) > time.Second*200 { // more than 3+ minutes
  281. Resubscribe(client, cfg)
  282. }
  283. }
  284. }
  285. }
  286. // ServerKeepAlive -- handler to react to keepalive messages published by server
  287. func ServerKeepAlive(client mqtt.Client, msg mqtt.Message) {
  288. serverid := string(msg.Payload())
  289. keepalive[serverid] = time.Now()
  290. ncutils.Log("keepalive from server")
  291. }
  292. // Resubscribe --- handles resubscribing if needed
  293. func Resubscribe(client mqtt.Client, cfg *config.ClientConfig) error {
  294. if err := config.ModConfig(&cfg.Node); err == nil {
  295. ncutils.Log("resubbing on network " + cfg.Node.Network)
  296. client.Disconnect(250)
  297. client = SetupMQTT(cfg)
  298. if token := client.Subscribe("update/"+cfg.Node.ID, 0, NodeUpdate); token.Wait() && token.Error() != nil {
  299. log.Fatal(token.Error())
  300. }
  301. if cfg.DebugOn {
  302. ncutils.Log("subscribed to node updates for node " + cfg.Node.Name + " update/" + cfg.Node.ID)
  303. }
  304. if token := client.Subscribe("update/peers/"+cfg.Node.ID, 0, UpdatePeers); token.Wait() && token.Error() != nil {
  305. log.Fatal(token.Error())
  306. }
  307. ncutils.Log("finished re subbing")
  308. return nil
  309. } else {
  310. ncutils.Log("could not mod config when re-subbing")
  311. return err
  312. }
  313. }
  314. // UpdateKeys -- updates private key and returns new publickey
  315. func UpdateKeys(cfg *config.ClientConfig, client mqtt.Client) error {
  316. ncutils.Log("received message to update keys")
  317. //potentiall blocking i/o so do this in a go routine
  318. key, err := wgtypes.GeneratePrivateKey()
  319. if err != nil {
  320. ncutils.Log("error generating privatekey " + err.Error())
  321. return err
  322. }
  323. file := ncutils.GetNetclientPathSpecific() + cfg.Node.Interface + ".conf"
  324. if err := wireguard.UpdatePrivateKey(file, key.String()); err != nil {
  325. ncutils.Log("error updating wireguard key " + err.Error())
  326. return err
  327. }
  328. cfg.Node.PublicKey = key.PublicKey().String()
  329. PublishNodeUpdate(cfg)
  330. if err := config.ModConfig(&cfg.Node); err != nil {
  331. ncutils.Log("error updating local config " + err.Error())
  332. }
  333. return nil
  334. }
  335. // Checkin -- go routine that checks for public or local ip changes, publishes changes
  336. // if there are no updates, simply "pings" the server as a checkin
  337. func Checkin(ctx context.Context, cfg *config.ClientConfig, network string) {
  338. for {
  339. select {
  340. case <-ctx.Done():
  341. ncutils.Log("Checkin cancelled")
  342. return
  343. //delay should be configuraable -> use cfg.Node.NetworkSettings.DefaultCheckInInterval ??
  344. case <-time.After(time.Second * 60):
  345. ncutils.Log("Checkin running")
  346. //read latest config
  347. cfg.ReadConfig()
  348. if cfg.Node.Roaming == "yes" && cfg.Node.IsStatic != "yes" {
  349. extIP, err := ncutils.GetPublicIP()
  350. if err != nil {
  351. ncutils.PrintLog("error encountered checking ip addresses: "+err.Error(), 1)
  352. }
  353. if cfg.Node.Endpoint != extIP && extIP != "" {
  354. ncutils.PrintLog("endpoint has changed from "+cfg.Node.Endpoint+" to "+extIP, 1)
  355. cfg.Node.Endpoint = extIP
  356. PublishNodeUpdate(cfg)
  357. }
  358. intIP, err := getPrivateAddr()
  359. if err != nil {
  360. ncutils.PrintLog("error encountered checking ip addresses: "+err.Error(), 1)
  361. }
  362. if cfg.Node.LocalAddress != intIP && intIP != "" {
  363. ncutils.PrintLog("local Address has changed from "+cfg.Node.LocalAddress+" to "+intIP, 1)
  364. cfg.Node.LocalAddress = intIP
  365. PublishNodeUpdate(cfg)
  366. }
  367. } else {
  368. localIP, err := ncutils.GetLocalIP(cfg.Node.LocalRange)
  369. if err != nil {
  370. ncutils.PrintLog("error encountered checking ip addresses: "+err.Error(), 1)
  371. }
  372. if cfg.Node.Endpoint != localIP && localIP != "" {
  373. ncutils.PrintLog("endpoint has changed from "+cfg.Node.Endpoint+" to "+localIP, 1)
  374. cfg.Node.Endpoint = localIP
  375. PublishNodeUpdate(cfg)
  376. }
  377. }
  378. Hello(cfg, network)
  379. ncutils.Log("Checkin complete")
  380. }
  381. }
  382. }
  383. // PublishNodeUpdates -- saves node and pushes changes to broker
  384. func PublishNodeUpdate(cfg *config.ClientConfig) {
  385. if err := config.Write(cfg, cfg.Network); err != nil {
  386. ncutils.Log("error saving configuration" + err.Error())
  387. }
  388. data, err := json.Marshal(cfg.Node)
  389. if err != nil {
  390. ncutils.Log("error marshling node update " + err.Error())
  391. }
  392. if err = publish(cfg, fmt.Sprintf("update/%s", cfg.Node.ID), data); err != nil {
  393. ncutils.Log(fmt.Sprintf("error publishing endpoint update, %v \n", err))
  394. }
  395. }
  396. // Hello -- ping the broker to let server know node is alive and doing fine
  397. func Hello(cfg *config.ClientConfig, network string) {
  398. if err := publish(cfg, fmt.Sprintf("ping/%s", cfg.Node.ID), []byte("hello world!")); err != nil {
  399. ncutils.Log(fmt.Sprintf("error publishing ping, %v \n", err))
  400. }
  401. }
  402. func publish(cfg *config.ClientConfig, dest string, msg []byte) error {
  403. // setup the keys
  404. trafficPrivKey, err := auth.RetrieveTrafficKey(cfg.Node.Network)
  405. if err != nil {
  406. return err
  407. }
  408. serverPubKey, err := ncutils.ConvertBytesToKey(cfg.Node.TrafficKeys.Server)
  409. if err != nil {
  410. return err
  411. }
  412. client := SetupMQTT(cfg)
  413. defer client.Disconnect(250)
  414. encrypted, err := ncutils.BoxEncrypt(msg, serverPubKey, trafficPrivKey)
  415. if err != nil {
  416. return err
  417. }
  418. if token := client.Publish(dest, 0, false, encrypted); token.Wait() && token.Error() != nil {
  419. return token.Error()
  420. }
  421. return nil
  422. }
  423. func parseNetworkFromTopic(topic string) string {
  424. return strings.Split(topic, "/")[1]
  425. }
  426. func decryptMsg(cfg *config.ClientConfig, msg []byte) ([]byte, error) {
  427. // setup the keys
  428. diskKey, keyErr := auth.RetrieveTrafficKey(cfg.Node.Network)
  429. if keyErr != nil {
  430. return nil, keyErr
  431. }
  432. serverPubKey, err := ncutils.ConvertBytesToKey(cfg.Node.TrafficKeys.Server)
  433. if err != nil {
  434. return nil, err
  435. }
  436. return ncutils.BoxDecrypt(msg, serverPubKey, diskKey)
  437. }
  438. func shouldResub(currentServers, newServers []models.ServerAddr) bool {
  439. if len(currentServers) != len(newServers) {
  440. return true
  441. }
  442. for _, srv := range currentServers {
  443. if !ncutils.ServerAddrSliceContains(newServers, srv) {
  444. return true
  445. }
  446. }
  447. return false
  448. }