daemon.go 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363
  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/local"
  24. "github.com/gravitl/netmaker/netclient/ncutils"
  25. "github.com/gravitl/netmaker/netclient/wireguard"
  26. ssl "github.com/gravitl/netmaker/tls"
  27. "github.com/kr/pretty"
  28. "golang.zx2c4.com/wireguard/wgctrl/wgtypes"
  29. )
  30. var messageCache = new(sync.Map)
  31. var serverSet map[string]bool
  32. const lastNodeUpdate = "lnu"
  33. const lastPeerUpdate = "lpu"
  34. type cachedMessage struct {
  35. Message string
  36. LastSeen time.Time
  37. }
  38. // Daemon runs netclient daemon from command line
  39. func Daemon() error {
  40. logger.Log(0, "netclient daemon started -- version:", ncutils.Version)
  41. UpdateClientConfig()
  42. if err := ncutils.SavePID(); err != nil {
  43. return err
  44. }
  45. // reference required to eliminate unused statticcheck
  46. serverSet = make(map[string]bool)
  47. serverSet["dummy"] = false
  48. // set ipforwarding on startup
  49. err := local.SetIPForwarding()
  50. if err != nil {
  51. logger.Log(0, err.Error())
  52. }
  53. // == add waitgroup and cancel for checkin routine ==
  54. wg := sync.WaitGroup{}
  55. quit := make(chan os.Signal, 1)
  56. reset := make(chan os.Signal, 1)
  57. signal.Notify(quit, syscall.SIGTERM, os.Interrupt)
  58. signal.Notify(reset, syscall.SIGHUP)
  59. cancel := startGoRoutines(&wg)
  60. for {
  61. select {
  62. case <-quit:
  63. cancel()
  64. logger.Log(0, "shutting down netclient daemon")
  65. wg.Wait()
  66. logger.Log(0, "shutdown complete")
  67. return nil
  68. case <-reset:
  69. logger.Log(0, "received reset")
  70. cancel()
  71. wg.Wait()
  72. logger.Log(0, "restarting daemon")
  73. cancel = startGoRoutines(&wg)
  74. }
  75. }
  76. }
  77. func startGoRoutines(wg *sync.WaitGroup) context.CancelFunc {
  78. defer wg.Done()
  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. server := cfg.Server.Server
  93. pretty.Println(serverSet, server, serverSet[server], !serverSet[server])
  94. if val, ok := serverSet[server]; ok {
  95. pretty.Println(val, ok)
  96. } else {
  97. pretty.Println("not ok", val, ok)
  98. }
  99. if !serverSet[server] {
  100. // == subscribe to all nodes for each on machine ==
  101. serverSet[server] = true
  102. logger.Log(1, "started daemon for server ", server)
  103. wg.Add(1)
  104. go messageQueue(ctx, wg, &cfg)
  105. }
  106. }
  107. return cancel
  108. }
  109. // UpdateKeys -- updates private key and returns new publickey
  110. func UpdateKeys(nodeCfg *config.ClientConfig, client mqtt.Client) error {
  111. logger.Log(0, "received message to update wireguard keys for network ", nodeCfg.Network)
  112. key, err := wgtypes.GeneratePrivateKey()
  113. if err != nil {
  114. logger.Log(0, "error generating privatekey ", err.Error())
  115. return err
  116. }
  117. file := ncutils.GetNetclientPathSpecific() + nodeCfg.Node.Interface + ".conf"
  118. if err := wireguard.UpdatePrivateKey(file, key.String()); err != nil {
  119. logger.Log(0, "error updating wireguard key ", err.Error())
  120. return err
  121. }
  122. if storeErr := wireguard.StorePrivKey(key.String(), nodeCfg.Network); storeErr != nil {
  123. logger.Log(0, "failed to save private key", storeErr.Error())
  124. return storeErr
  125. }
  126. nodeCfg.Node.PublicKey = key.PublicKey().String()
  127. PublishNodeUpdate(nodeCfg)
  128. return nil
  129. }
  130. // == Private ==
  131. // sets MQ client subscriptions for a specific node config
  132. // should be called for each node belonging to a given server
  133. func setSubscriptions(client mqtt.Client, nodeCfg *config.ClientConfig) {
  134. 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 {
  135. if token.Error() == nil {
  136. logger.Log(0, "connection timeout")
  137. } else {
  138. logger.Log(0, token.Error().Error())
  139. }
  140. return
  141. }
  142. logger.Log(3, fmt.Sprintf("subscribed to node updates for node %s update/%s/%s", nodeCfg.Node.Name, nodeCfg.Node.Network, nodeCfg.Node.ID))
  143. if token := client.Subscribe(fmt.Sprintf("peers/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID), 0, mqtt.MessageHandler(UpdatePeers)); token.Wait() && token.Error() != nil {
  144. logger.Log(0, token.Error().Error())
  145. return
  146. }
  147. logger.Log(3, fmt.Sprintf("subscribed to peer updates for node %s peers/%s/%s", nodeCfg.Node.Name, nodeCfg.Node.Network, nodeCfg.Node.ID))
  148. }
  149. // on a delete usually, pass in the nodecfg to unsubscribe client broker communications
  150. // for the node in nodeCfg
  151. func unsubscribeNode(client mqtt.Client, nodeCfg *config.ClientConfig) {
  152. client.Unsubscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID))
  153. var ok = true
  154. 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 {
  155. if token.Error() == nil {
  156. logger.Log(1, "unable to unsubscribe from updates for node ", nodeCfg.Node.Name, "\n", "connection timeout")
  157. } else {
  158. logger.Log(1, "unable to unsubscribe from updates for node ", nodeCfg.Node.Name, "\n", token.Error().Error())
  159. }
  160. ok = false
  161. }
  162. 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 {
  163. if token.Error() == nil {
  164. logger.Log(1, "unable to unsubscribe from peer updates for node ", nodeCfg.Node.Name, "\n", "connection timeout")
  165. } else {
  166. logger.Log(1, "unable to unsubscribe from peer updates for node ", nodeCfg.Node.Name, "\n", token.Error().Error())
  167. }
  168. ok = false
  169. }
  170. if ok {
  171. logger.Log(1, "successfully unsubscribed node ", nodeCfg.Node.ID, " : ", nodeCfg.Node.Name)
  172. }
  173. }
  174. // sets up Message Queue and subsribes/publishes updates to/from server
  175. // the client should subscribe to ALL nodes that exist on server locally
  176. func messageQueue(ctx context.Context, wg *sync.WaitGroup, cfg *config.ClientConfig) {
  177. defer wg.Done()
  178. logger.Log(0, "netclient message queue started for server: ", cfg.Server.Server)
  179. client, err := setupMQTT(cfg, false)
  180. if err != nil {
  181. logger.Log(0, "unable to connect to broker", cfg.Server.Server, err.Error())
  182. return
  183. }
  184. defer client.Disconnect(250)
  185. <-ctx.Done()
  186. logger.Log(0, "shutting down message queue for server ", cfg.Server.Server)
  187. }
  188. // NewTLSConf sets up tls configuration to connect to broker securely
  189. func NewTLSConfig(server string) (*tls.Config, error) {
  190. file := ncutils.GetNetclientServerPath(server) + ncutils.GetSeparator() + "root.pem"
  191. certpool := x509.NewCertPool()
  192. ca, err := os.ReadFile(file)
  193. if err != nil {
  194. logger.Log(0, "could not read CA file ", err.Error())
  195. }
  196. ok := certpool.AppendCertsFromPEM(ca)
  197. if !ok {
  198. logger.Log(0, "failed to append cert")
  199. }
  200. clientKeyPair, err := tls.LoadX509KeyPair(ncutils.GetNetclientServerPath(server)+ncutils.GetSeparator()+"client.pem", ncutils.GetNetclientPath()+ncutils.GetSeparator()+"client.key")
  201. if err != nil {
  202. logger.Log(0, "could not read client cert/key ", err.Error())
  203. return nil, err
  204. }
  205. certs := []tls.Certificate{clientKeyPair}
  206. return &tls.Config{
  207. RootCAs: certpool,
  208. ClientAuth: tls.NoClientCert,
  209. ClientCAs: nil,
  210. Certificates: certs,
  211. InsecureSkipVerify: false,
  212. }, nil
  213. }
  214. // setupMQTT creates a connection to broker and returns client
  215. // this function is primarily used to create a connection to publish to the broker
  216. func setupMQTT(cfg *config.ClientConfig, publish bool) (mqtt.Client, error) {
  217. opts := mqtt.NewClientOptions()
  218. server := cfg.Server.Server
  219. port := cfg.Server.MQPort
  220. opts.AddBroker("ssl://" + server + ":" + port)
  221. tlsConfig, err := NewTLSConfig(server)
  222. if err != nil {
  223. logger.Log(0, "failed to get TLS config for", server, err.Error())
  224. return nil, err
  225. }
  226. opts.SetTLSConfig(tlsConfig)
  227. opts.SetClientID(ncutils.MakeRandomString(23))
  228. opts.SetDefaultPublishHandler(All)
  229. opts.SetAutoReconnect(true)
  230. opts.SetConnectRetry(true)
  231. opts.SetConnectRetryInterval(time.Second << 2)
  232. opts.SetKeepAlive(time.Minute >> 1)
  233. opts.SetWriteTimeout(time.Minute)
  234. opts.SetOnConnectHandler(func(client mqtt.Client) {
  235. if !publish {
  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. })
  248. opts.SetOrderMatters(true)
  249. opts.SetResumeSubs(true)
  250. opts.SetConnectionLostHandler(func(c mqtt.Client, e error) {
  251. logger.Log(0, "detected broker connection lost for", cfg.Server.Server)
  252. })
  253. client := mqtt.NewClient(opts)
  254. for token := client.Connect(); !token.WaitTimeout(30*time.Second) || token.Error() != nil; token = client.Connect() {
  255. logger.Log(0, "unable to connect to broker, retrying ...")
  256. var err error
  257. if token.Error() == nil {
  258. err = errors.New("connect timeout")
  259. } else {
  260. err = token.Error()
  261. }
  262. if err := checkBroker(cfg.Server.Server, cfg.Server.MQPort); err != nil {
  263. return nil, err
  264. }
  265. logger.Log(0, "could not connect to broker", cfg.Server.Server, err.Error())
  266. if strings.Contains(err.Error(), "connectex") || strings.Contains(err.Error(), "connect timeout") {
  267. reRegisterWithServer(cfg)
  268. }
  269. }
  270. return client, nil
  271. }
  272. func reRegisterWithServer(cfg *config.ClientConfig) {
  273. logger.Log(0, "connection issue detected.. attempt connection with new certs and broker information")
  274. key, err := ssl.ReadKeyFromFile(ncutils.GetNetclientPath() + ncutils.GetSeparator() + "client.key")
  275. if err != nil {
  276. _, *key, err = ed25519.GenerateKey(rand.Reader)
  277. if err != nil {
  278. log.Fatal("could not generate new key")
  279. }
  280. }
  281. RegisterWithServer(key, cfg)
  282. daemon.Restart()
  283. }
  284. // publishes a message to server to update peers on this peer's behalf
  285. func publishSignal(nodeCfg *config.ClientConfig, signal byte) error {
  286. if err := publish(nodeCfg, fmt.Sprintf("signal/%s", nodeCfg.Node.ID), []byte{signal}, 1); err != nil {
  287. return err
  288. }
  289. return nil
  290. }
  291. func parseNetworkFromTopic(topic string) string {
  292. return strings.Split(topic, "/")[1]
  293. }
  294. // should only ever use node client configs
  295. func decryptMsg(nodeCfg *config.ClientConfig, msg []byte) ([]byte, error) {
  296. if len(msg) <= 24 { // make sure message is of appropriate length
  297. return nil, fmt.Errorf("recieved invalid message from broker %v", msg)
  298. }
  299. // setup the keys
  300. diskKey, keyErr := auth.RetrieveTrafficKey(nodeCfg.Node.Network)
  301. if keyErr != nil {
  302. return nil, keyErr
  303. }
  304. serverPubKey, err := ncutils.ConvertBytesToKey(nodeCfg.Node.TrafficKeys.Server)
  305. if err != nil {
  306. return nil, err
  307. }
  308. return ncutils.DeChunk(msg, serverPubKey, diskKey)
  309. }
  310. // == Message Caches ==
  311. func insert(network, which, cache string) {
  312. var newMessage = cachedMessage{
  313. Message: cache,
  314. LastSeen: time.Now(),
  315. }
  316. messageCache.Store(fmt.Sprintf("%s%s", network, which), newMessage)
  317. }
  318. func read(network, which string) string {
  319. val, isok := messageCache.Load(fmt.Sprintf("%s%s", network, which))
  320. if isok {
  321. var readMessage = val.(cachedMessage) // fetch current cached message
  322. if readMessage.LastSeen.IsZero() {
  323. return ""
  324. }
  325. if time.Now().After(readMessage.LastSeen.Add(time.Hour * 24)) { // check if message has been there over a minute
  326. messageCache.Delete(fmt.Sprintf("%s%s", network, which)) // remove old message if expired
  327. return ""
  328. }
  329. return readMessage.Message // return current message if not expired
  330. }
  331. return ""
  332. }