daemon.go 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370
  1. package functions
  2. import (
  3. "context"
  4. "crypto/ed25519"
  5. "crypto/rand"
  6. "crypto/tls"
  7. "crypto/x509"
  8. "errors"
  9. "fmt"
  10. "log"
  11. "os"
  12. "os/signal"
  13. "strings"
  14. "sync"
  15. "syscall"
  16. "time"
  17. mqtt "github.com/eclipse/paho.mqtt.golang"
  18. "github.com/gravitl/netmaker/logger"
  19. "github.com/gravitl/netmaker/mq"
  20. "github.com/gravitl/netmaker/netclient/auth"
  21. "github.com/gravitl/netmaker/netclient/config"
  22. "github.com/gravitl/netmaker/netclient/daemon"
  23. "github.com/gravitl/netmaker/netclient/global_settings"
  24. "github.com/gravitl/netmaker/netclient/local"
  25. "github.com/gravitl/netmaker/netclient/ncutils"
  26. "github.com/gravitl/netmaker/netclient/wireguard"
  27. ssl "github.com/gravitl/netmaker/tls"
  28. "golang.zx2c4.com/wireguard/wgctrl/wgtypes"
  29. )
  30. var messageCache = new(sync.Map)
  31. var serverSet map[string]bool
  32. var mqclient mqtt.Client
  33. const lastNodeUpdate = "lnu"
  34. const lastPeerUpdate = "lpu"
  35. type cachedMessage struct {
  36. Message string
  37. LastSeen time.Time
  38. }
  39. // Daemon runs netclient daemon from command line
  40. func Daemon() error {
  41. logger.Log(0, "netclient daemon started -- version:", ncutils.Version)
  42. UpdateClientConfig()
  43. if err := ncutils.SavePID(); err != nil {
  44. return err
  45. }
  46. // reference required to eliminate unused statticcheck
  47. serverSet = make(map[string]bool)
  48. serverSet["dummy"] = false
  49. // set ipforwarding on startup
  50. err := local.SetIPForwarding()
  51. if err != nil {
  52. logger.Log(0, err.Error())
  53. }
  54. // == add waitgroup and cancel for checkin routine ==
  55. wg := sync.WaitGroup{}
  56. quit := make(chan os.Signal, 1)
  57. reset := make(chan os.Signal, 1)
  58. signal.Notify(quit, syscall.SIGTERM, os.Interrupt)
  59. signal.Notify(reset, syscall.SIGHUP)
  60. cancel := startGoRoutines(&wg)
  61. for {
  62. select {
  63. case <-quit:
  64. cancel()
  65. logger.Log(0, "shutting down netclient daemon")
  66. wg.Wait()
  67. logger.Log(0, "shutdown complete")
  68. return nil
  69. case <-reset:
  70. logger.Log(0, "received reset")
  71. cancel()
  72. wg.Wait()
  73. logger.Log(0, "restarting daemon")
  74. cancel = startGoRoutines(&wg)
  75. }
  76. }
  77. }
  78. func startGoRoutines(wg *sync.WaitGroup) context.CancelFunc {
  79. ctx, cancel := context.WithCancel(context.Background())
  80. wg.Add(1)
  81. go Checkin(ctx, wg)
  82. serverSet := make(map[string]bool)
  83. networks, _ := ncutils.GetSystemNetworks()
  84. for _, network := range networks {
  85. logger.Log(3, "initializing network", network)
  86. cfg := config.ClientConfig{}
  87. cfg.Network = network
  88. cfg.ReadConfig()
  89. if err := wireguard.ApplyConf(&cfg.Node, cfg.Node.Interface, ncutils.GetNetclientPathSpecific()+cfg.Node.Interface+".conf"); err != nil {
  90. logger.Log(0, "failed to start ", cfg.Node.Interface, "wg interface", err.Error())
  91. }
  92. if cfg.PublicIPService != "" {
  93. global_settings.PublicIPServices[network] = cfg.PublicIPService
  94. }
  95. server := cfg.Server.Server
  96. if !serverSet[server] {
  97. // == subscribe to all nodes for each on machine ==
  98. serverSet[server] = true
  99. logger.Log(1, "started daemon for server ", server)
  100. err := local.SetNetmakerDomainRoute(cfg.Server.API)
  101. if err != nil {
  102. logger.Log(0, "error setting route for netmaker: "+err.Error())
  103. }
  104. wg.Add(1)
  105. go messageQueue(ctx, wg, &cfg)
  106. }
  107. }
  108. return cancel
  109. }
  110. // UpdateKeys -- updates private key and returns new publickey
  111. func UpdateKeys(nodeCfg *config.ClientConfig, client mqtt.Client) error {
  112. logger.Log(0, "interface:", nodeCfg.Node.Interface, "received message to update wireguard keys for network ", nodeCfg.Network)
  113. key, err := wgtypes.GeneratePrivateKey()
  114. if err != nil {
  115. logger.Log(0, "network:", nodeCfg.Node.Network, "error generating privatekey ", err.Error())
  116. return err
  117. }
  118. file := ncutils.GetNetclientPathSpecific() + nodeCfg.Node.Interface + ".conf"
  119. if err := wireguard.UpdatePrivateKey(file, key.String()); err != nil {
  120. logger.Log(0, "network:", nodeCfg.Node.Network, "error updating wireguard key ", err.Error())
  121. return err
  122. }
  123. if storeErr := wireguard.StorePrivKey(key.String(), nodeCfg.Network); storeErr != nil {
  124. logger.Log(0, "network:", nodeCfg.Network, "failed to save private key", storeErr.Error())
  125. return storeErr
  126. }
  127. nodeCfg.Node.PublicKey = key.PublicKey().String()
  128. PublishNodeUpdate(nodeCfg)
  129. return nil
  130. }
  131. // == Private ==
  132. // sets MQ client subscriptions for a specific node config
  133. // should be called for each node belonging to a given server
  134. func setSubscriptions(client mqtt.Client, nodeCfg *config.ClientConfig) {
  135. if token := client.Subscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID), 0, mqtt.MessageHandler(NodeUpdate)); token.WaitTimeout(mq.MQ_TIMEOUT*time.Second) && token.Error() != nil {
  136. if token.Error() == nil {
  137. logger.Log(0, "network:", nodeCfg.Node.Network, "connection timeout")
  138. } else {
  139. logger.Log(0, "network:", nodeCfg.Node.Network, token.Error().Error())
  140. }
  141. return
  142. }
  143. logger.Log(3, fmt.Sprintf("subscribed to node updates for node %s update/%s/%s", nodeCfg.Node.Name, nodeCfg.Node.Network, nodeCfg.Node.ID))
  144. if token := client.Subscribe(fmt.Sprintf("peers/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID), 0, mqtt.MessageHandler(UpdatePeers)); token.Wait() && token.Error() != nil {
  145. logger.Log(0, "network", nodeCfg.Node.Network, token.Error().Error())
  146. return
  147. }
  148. logger.Log(3, fmt.Sprintf("subscribed to peer updates for node %s peers/%s/%s", nodeCfg.Node.Name, nodeCfg.Node.Network, nodeCfg.Node.ID))
  149. }
  150. // on a delete usually, pass in the nodecfg to unsubscribe client broker communications
  151. // for the node in nodeCfg
  152. func unsubscribeNode(client mqtt.Client, nodeCfg *config.ClientConfig) {
  153. client.Unsubscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID))
  154. var ok = true
  155. if token := client.Unsubscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID)); token.WaitTimeout(mq.MQ_TIMEOUT*time.Second) && token.Error() != nil {
  156. if token.Error() == nil {
  157. logger.Log(1, "network:", nodeCfg.Node.Network, "unable to unsubscribe from updates for node ", nodeCfg.Node.Name, "\n", "connection timeout")
  158. } else {
  159. logger.Log(1, "network:", nodeCfg.Node.Network, "unable to unsubscribe from updates for node ", nodeCfg.Node.Name, "\n", token.Error().Error())
  160. }
  161. ok = false
  162. }
  163. if token := client.Unsubscribe(fmt.Sprintf("peers/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID)); token.WaitTimeout(mq.MQ_TIMEOUT*time.Second) && token.Error() != nil {
  164. if token.Error() == nil {
  165. logger.Log(1, "network:", nodeCfg.Node.Network, "unable to unsubscribe from peer updates for node", nodeCfg.Node.Name, "\n", "connection timeout")
  166. } else {
  167. logger.Log(1, "network:", nodeCfg.Node.Network, "unable to unsubscribe from peer updates for node", nodeCfg.Node.Name, "\n", token.Error().Error())
  168. }
  169. ok = false
  170. }
  171. if ok {
  172. logger.Log(1, "network:", nodeCfg.Node.Network, "successfully unsubscribed node ", nodeCfg.Node.ID, " : ", nodeCfg.Node.Name)
  173. }
  174. }
  175. // sets up Message Queue and subsribes/publishes updates to/from server
  176. // the client should subscribe to ALL nodes that exist on server locally
  177. func messageQueue(ctx context.Context, wg *sync.WaitGroup, cfg *config.ClientConfig) {
  178. defer wg.Done()
  179. logger.Log(0, "network:", cfg.Node.Network, "netclient message queue started for server:", cfg.Server.Server)
  180. err := setupMQTT(cfg)
  181. if err != nil {
  182. logger.Log(0, "unable to connect to broker", cfg.Server.Server, err.Error())
  183. return
  184. }
  185. //defer mqclient.Disconnect(250)
  186. <-ctx.Done()
  187. logger.Log(0, "shutting down message queue for server", cfg.Server.Server)
  188. }
  189. // NewTLSConf sets up tls configuration to connect to broker securely
  190. func NewTLSConfig(server string) (*tls.Config, error) {
  191. file := ncutils.GetNetclientServerPath(server) + ncutils.GetSeparator() + "root.pem"
  192. certpool := x509.NewCertPool()
  193. ca, err := os.ReadFile(file)
  194. if err != nil {
  195. logger.Log(0, "could not read CA file", err.Error())
  196. }
  197. ok := certpool.AppendCertsFromPEM(ca)
  198. if !ok {
  199. logger.Log(0, "failed to append cert")
  200. }
  201. clientKeyPair, err := tls.LoadX509KeyPair(ncutils.GetNetclientServerPath(server)+ncutils.GetSeparator()+"client.pem", ncutils.GetNetclientPath()+ncutils.GetSeparator()+"client.key")
  202. if err != nil {
  203. logger.Log(0, "could not read client cert/key", err.Error())
  204. return nil, err
  205. }
  206. certs := []tls.Certificate{clientKeyPair}
  207. return &tls.Config{
  208. RootCAs: certpool,
  209. ClientAuth: tls.NoClientCert,
  210. ClientCAs: nil,
  211. Certificates: certs,
  212. InsecureSkipVerify: false,
  213. }, nil
  214. }
  215. // setupMQTT creates a connection to broker and returns client
  216. // this function is primarily used to create a connection to publish to the broker
  217. func setupMQTT(cfg *config.ClientConfig) error {
  218. opts := mqtt.NewClientOptions()
  219. server := cfg.Server.Server
  220. port := cfg.Server.MQPort
  221. opts.AddBroker("ssl://" + server + ":" + port)
  222. tlsConfig, err := NewTLSConfig(server)
  223. if err != nil {
  224. logger.Log(0, "failed to get TLS config for", server, err.Error())
  225. return err
  226. }
  227. opts.SetTLSConfig(tlsConfig)
  228. opts.SetClientID(ncutils.MakeRandomString(23))
  229. opts.SetDefaultPublishHandler(All)
  230. opts.SetAutoReconnect(true)
  231. opts.SetConnectRetry(true)
  232. opts.SetConnectRetryInterval(time.Second << 2)
  233. opts.SetKeepAlive(time.Minute >> 1)
  234. opts.SetWriteTimeout(time.Minute)
  235. opts.SetOnConnectHandler(func(client mqtt.Client) {
  236. networks, err := ncutils.GetSystemNetworks()
  237. if err != nil {
  238. logger.Log(0, "error retriving networks", err.Error())
  239. }
  240. for _, network := range networks {
  241. var currNodeCfg config.ClientConfig
  242. currNodeCfg.Network = network
  243. currNodeCfg.ReadConfig()
  244. setSubscriptions(client, &currNodeCfg)
  245. }
  246. })
  247. opts.SetOrderMatters(true)
  248. opts.SetResumeSubs(true)
  249. opts.SetConnectionLostHandler(func(c mqtt.Client, e error) {
  250. logger.Log(0, "network:", cfg.Node.Network, "detected broker connection lost for", cfg.Server.Server)
  251. })
  252. mqclient = mqtt.NewClient(opts)
  253. log.Println(mqclient)
  254. var connecterr error
  255. for count := 0; count < 3; count++ {
  256. connecterr = nil
  257. if token := mqclient.Connect(); !token.WaitTimeout(30*time.Second) || token.Error() != nil {
  258. logger.Log(0, "unable to connect to broker, retrying ...")
  259. if token.Error() == nil {
  260. connecterr = errors.New("connect timeout")
  261. } else {
  262. connecterr = token.Error()
  263. }
  264. if err := checkBroker(cfg.Server.Server, cfg.Server.MQPort); err != nil {
  265. logger.Log(0, "could not connect to broker", cfg.Server.Server, err.Error())
  266. }
  267. }
  268. }
  269. if connecterr != nil {
  270. reRegisterWithServer(cfg)
  271. //try after re-registering
  272. if token := mqclient.Connect(); !token.WaitTimeout(30*time.Second) || token.Error() != nil {
  273. return errors.New("unable to connect to broker")
  274. }
  275. }
  276. return nil
  277. }
  278. func reRegisterWithServer(cfg *config.ClientConfig) {
  279. logger.Log(0, "connection issue detected.. attempt connection with new certs and broker information")
  280. key, err := ssl.ReadKeyFromFile(ncutils.GetNetclientPath() + ncutils.GetSeparator() + "client.key")
  281. if err != nil {
  282. _, *key, err = ed25519.GenerateKey(rand.Reader)
  283. if err != nil {
  284. log.Fatal("could not generate new key")
  285. }
  286. }
  287. RegisterWithServer(key, cfg)
  288. daemon.Restart()
  289. }
  290. // publishes a message to server to update peers on this peer's behalf
  291. func publishSignal(nodeCfg *config.ClientConfig, signal byte) error {
  292. if err := publish(nodeCfg, fmt.Sprintf("signal/%s", nodeCfg.Node.ID), []byte{signal}, 1); err != nil {
  293. return err
  294. }
  295. return nil
  296. }
  297. func parseNetworkFromTopic(topic string) string {
  298. return strings.Split(topic, "/")[1]
  299. }
  300. // should only ever use node client configs
  301. func decryptMsg(nodeCfg *config.ClientConfig, msg []byte) ([]byte, error) {
  302. if len(msg) <= 24 { // make sure message is of appropriate length
  303. return nil, fmt.Errorf("recieved invalid message from broker %v", msg)
  304. }
  305. // setup the keys
  306. diskKey, keyErr := auth.RetrieveTrafficKey(nodeCfg.Node.Network)
  307. if keyErr != nil {
  308. return nil, keyErr
  309. }
  310. serverPubKey, err := ncutils.ConvertBytesToKey(nodeCfg.Node.TrafficKeys.Server)
  311. if err != nil {
  312. return nil, err
  313. }
  314. return ncutils.DeChunk(msg, serverPubKey, diskKey)
  315. }
  316. // == Message Caches ==
  317. func insert(network, which, cache string) {
  318. var newMessage = cachedMessage{
  319. Message: cache,
  320. LastSeen: time.Now(),
  321. }
  322. messageCache.Store(fmt.Sprintf("%s%s", network, which), newMessage)
  323. }
  324. func read(network, which string) string {
  325. val, isok := messageCache.Load(fmt.Sprintf("%s%s", network, which))
  326. if isok {
  327. var readMessage = val.(cachedMessage) // fetch current cached message
  328. if readMessage.LastSeen.IsZero() {
  329. return ""
  330. }
  331. if time.Now().After(readMessage.LastSeen.Add(time.Hour * 24)) { // check if message has been there over a minute
  332. messageCache.Delete(fmt.Sprintf("%s%s", network, which)) // remove old message if expired
  333. return ""
  334. }
  335. return readMessage.Message // return current message if not expired
  336. }
  337. return ""
  338. }