daemon.go 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360
  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. "golang.zx2c4.com/wireguard/wgctrl/wgtypes"
  28. )
  29. var messageCache = new(sync.Map)
  30. var serverSet map[string]bool
  31. const lastNodeUpdate = "lnu"
  32. const lastPeerUpdate = "lpu"
  33. type cachedMessage struct {
  34. Message string
  35. LastSeen time.Time
  36. }
  37. // Daemon runs netclient daemon from command line
  38. func Daemon() error {
  39. logger.Log(0, "netclient daemon started -- version:", ncutils.Version)
  40. UpdateClientConfig()
  41. if err := ncutils.SavePID(); err != nil {
  42. return err
  43. }
  44. // reference required to eliminate unused statticcheck
  45. serverSet = make(map[string]bool)
  46. serverSet["dummy"] = false
  47. // set ipforwarding on startup
  48. err := local.SetIPForwarding()
  49. if err != nil {
  50. logger.Log(0, err.Error())
  51. }
  52. // == add waitgroup and cancel for checkin routine ==
  53. wg := sync.WaitGroup{}
  54. quit := make(chan os.Signal, 1)
  55. reset := make(chan os.Signal, 1)
  56. signal.Notify(quit, syscall.SIGTERM, os.Interrupt)
  57. signal.Notify(reset, syscall.SIGHUP)
  58. cancel := startGoRoutines(&wg)
  59. for {
  60. select {
  61. case <-quit:
  62. cancel()
  63. logger.Log(0, "shutting down netclient daemon")
  64. wg.Wait()
  65. logger.Log(0, "shutdown complete")
  66. return nil
  67. case <-reset:
  68. logger.Log(0, "received reset")
  69. cancel()
  70. wg.Wait()
  71. logger.Log(0, "restarting daemon")
  72. cancel = startGoRoutines(&wg)
  73. }
  74. }
  75. }
  76. func startGoRoutines(wg *sync.WaitGroup) context.CancelFunc {
  77. ctx, cancel := context.WithCancel(context.Background())
  78. wg.Add(1)
  79. go Checkin(ctx, wg)
  80. serverSet := make(map[string]bool)
  81. networks, _ := ncutils.GetSystemNetworks()
  82. for _, network := range networks {
  83. logger.Log(3, "initializing network", network)
  84. cfg := config.ClientConfig{}
  85. cfg.Network = network
  86. cfg.ReadConfig()
  87. if err := wireguard.ApplyConf(&cfg.Node, cfg.Node.Interface, ncutils.GetNetclientPathSpecific()+cfg.Node.Interface+".conf"); err != nil {
  88. logger.Log(0, "failed to start ", cfg.Node.Interface, "wg interface", err.Error())
  89. }
  90. server := cfg.Server.Server
  91. if !serverSet[server] {
  92. // == subscribe to all nodes for each on machine ==
  93. serverSet[server] = true
  94. logger.Log(1, "started daemon for server ", server)
  95. wg.Add(1)
  96. go messageQueue(ctx, wg, &cfg)
  97. }
  98. }
  99. return cancel
  100. }
  101. // UpdateKeys -- updates private key and returns new publickey
  102. func UpdateKeys(nodeCfg *config.ClientConfig, client mqtt.Client) error {
  103. logger.Log(0, "interface:", nodeCfg.Node.Interface, "received message to update wireguard keys for network ", nodeCfg.Network)
  104. key, err := wgtypes.GeneratePrivateKey()
  105. if err != nil {
  106. logger.Log(0, "network:", nodeCfg.Node.Network, "error generating privatekey ", err.Error())
  107. return err
  108. }
  109. file := ncutils.GetNetclientPathSpecific() + nodeCfg.Node.Interface + ".conf"
  110. if err := wireguard.UpdatePrivateKey(file, key.String()); err != nil {
  111. logger.Log(0, "network:", nodeCfg.Node.Network, "error updating wireguard key ", err.Error())
  112. return err
  113. }
  114. if storeErr := wireguard.StorePrivKey(key.String(), nodeCfg.Network); storeErr != nil {
  115. logger.Log(0, "network:", nodeCfg.Network, "failed to save private key", storeErr.Error())
  116. return storeErr
  117. }
  118. nodeCfg.Node.PublicKey = key.PublicKey().String()
  119. PublishNodeUpdate(nodeCfg)
  120. return nil
  121. }
  122. // == Private ==
  123. // sets MQ client subscriptions for a specific node config
  124. // should be called for each node belonging to a given server
  125. func setSubscriptions(client mqtt.Client, nodeCfg *config.ClientConfig) {
  126. 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 {
  127. if token.Error() == nil {
  128. logger.Log(0, "network:", nodeCfg.Node.Network, "connection timeout")
  129. } else {
  130. logger.Log(0, "network:", nodeCfg.Node.Network, token.Error().Error())
  131. }
  132. return
  133. }
  134. logger.Log(3, fmt.Sprintf("subscribed to node updates for node %s update/%s/%s", nodeCfg.Node.Name, nodeCfg.Node.Network, nodeCfg.Node.ID))
  135. if token := client.Subscribe(fmt.Sprintf("peers/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID), 0, mqtt.MessageHandler(UpdatePeers)); token.Wait() && token.Error() != nil {
  136. logger.Log(0, "network", nodeCfg.Node.Network, token.Error().Error())
  137. return
  138. }
  139. logger.Log(3, fmt.Sprintf("subscribed to peer updates for node %s peers/%s/%s", nodeCfg.Node.Name, nodeCfg.Node.Network, nodeCfg.Node.ID))
  140. }
  141. // on a delete usually, pass in the nodecfg to unsubscribe client broker communications
  142. // for the node in nodeCfg
  143. func unsubscribeNode(client mqtt.Client, nodeCfg *config.ClientConfig) {
  144. client.Unsubscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID))
  145. var ok = true
  146. 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 {
  147. if token.Error() == nil {
  148. logger.Log(1, "network:", nodeCfg.Node.Network, "unable to unsubscribe from updates for node ", nodeCfg.Node.Name, "\n", "connection timeout")
  149. } else {
  150. logger.Log(1, "network:", nodeCfg.Node.Network, "unable to unsubscribe from updates for node ", nodeCfg.Node.Name, "\n", token.Error().Error())
  151. }
  152. ok = false
  153. }
  154. 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 {
  155. if token.Error() == nil {
  156. logger.Log(1, "network:", nodeCfg.Node.Network, "unable to unsubscribe from peer updates for node", nodeCfg.Node.Name, "\n", "connection timeout")
  157. } else {
  158. logger.Log(1, "network:", nodeCfg.Node.Network, "unable to unsubscribe from peer updates for node", nodeCfg.Node.Name, "\n", token.Error().Error())
  159. }
  160. ok = false
  161. }
  162. if ok {
  163. logger.Log(1, "network:", nodeCfg.Node.Network, "successfully unsubscribed node ", nodeCfg.Node.ID, " : ", nodeCfg.Node.Name)
  164. }
  165. }
  166. // sets up Message Queue and subsribes/publishes updates to/from server
  167. // the client should subscribe to ALL nodes that exist on server locally
  168. func messageQueue(ctx context.Context, wg *sync.WaitGroup, cfg *config.ClientConfig) {
  169. defer wg.Done()
  170. logger.Log(0, "network:", cfg.Node.Network, "netclient message queue started for server:", cfg.Server.Server)
  171. client, err := setupMQTT(cfg, false)
  172. if err != nil {
  173. logger.Log(0, "unable to connect to broker", cfg.Server.Server, err.Error())
  174. return
  175. }
  176. defer client.Disconnect(250)
  177. <-ctx.Done()
  178. logger.Log(0, "shutting down message queue for server", cfg.Server.Server)
  179. }
  180. // NewTLSConf sets up tls configuration to connect to broker securely
  181. func NewTLSConfig(server string) (*tls.Config, error) {
  182. file := ncutils.GetNetclientServerPath(server) + ncutils.GetSeparator() + "root.pem"
  183. certpool := x509.NewCertPool()
  184. ca, err := os.ReadFile(file)
  185. if err != nil {
  186. logger.Log(0, "could not read CA file", err.Error())
  187. }
  188. ok := certpool.AppendCertsFromPEM(ca)
  189. if !ok {
  190. logger.Log(0, "failed to append cert")
  191. }
  192. clientKeyPair, err := tls.LoadX509KeyPair(ncutils.GetNetclientServerPath(server)+ncutils.GetSeparator()+"client.pem", ncutils.GetNetclientPath()+ncutils.GetSeparator()+"client.key")
  193. if err != nil {
  194. logger.Log(0, "could not read client cert/key", err.Error())
  195. return nil, err
  196. }
  197. certs := []tls.Certificate{clientKeyPair}
  198. return &tls.Config{
  199. RootCAs: certpool,
  200. ClientAuth: tls.NoClientCert,
  201. ClientCAs: nil,
  202. Certificates: certs,
  203. InsecureSkipVerify: false,
  204. }, nil
  205. }
  206. // setupMQTT creates a connection to broker and returns client
  207. // this function is primarily used to create a connection to publish to the broker
  208. func setupMQTT(cfg *config.ClientConfig, publish bool) (mqtt.Client, error) {
  209. opts := mqtt.NewClientOptions()
  210. server := cfg.Server.Server
  211. port := cfg.Server.MQPort
  212. opts.AddBroker("ssl://" + server + ":" + port)
  213. tlsConfig, err := NewTLSConfig(server)
  214. if err != nil {
  215. logger.Log(0, "failed to get TLS config for", server, err.Error())
  216. return nil, err
  217. }
  218. opts.SetTLSConfig(tlsConfig)
  219. opts.SetClientID(ncutils.MakeRandomString(23))
  220. opts.SetDefaultPublishHandler(All)
  221. opts.SetAutoReconnect(true)
  222. opts.SetConnectRetry(true)
  223. opts.SetConnectRetryInterval(time.Second << 2)
  224. opts.SetKeepAlive(time.Minute >> 1)
  225. opts.SetWriteTimeout(time.Minute)
  226. opts.SetOnConnectHandler(func(client mqtt.Client) {
  227. if !publish {
  228. networks, err := ncutils.GetSystemNetworks()
  229. if err != nil {
  230. logger.Log(0, "error retriving networks", err.Error())
  231. }
  232. for _, network := range networks {
  233. var currNodeCfg config.ClientConfig
  234. currNodeCfg.Network = network
  235. currNodeCfg.ReadConfig()
  236. setSubscriptions(client, &currNodeCfg)
  237. }
  238. }
  239. })
  240. opts.SetOrderMatters(true)
  241. opts.SetResumeSubs(true)
  242. opts.SetConnectionLostHandler(func(c mqtt.Client, e error) {
  243. logger.Log(0, "network:", cfg.Node.Network, "detected broker connection lost for", cfg.Server.Server)
  244. })
  245. client := mqtt.NewClient(opts)
  246. var connecterr error
  247. for count := 0; count < 3; count++ {
  248. connecterr = nil
  249. if token := client.Connect(); !token.WaitTimeout(30*time.Second) || token.Error() != nil {
  250. logger.Log(0, "unable to connect to broker, retrying ...")
  251. if token.Error() == nil {
  252. connecterr = errors.New("connect timeout")
  253. } else {
  254. connecterr = token.Error()
  255. }
  256. if err := checkBroker(cfg.Server.Server, cfg.Server.MQPort); err != nil {
  257. logger.Log(0, "could not connect to broker", cfg.Server.Server, err.Error())
  258. }
  259. }
  260. }
  261. if connecterr != nil {
  262. reRegisterWithServer(cfg)
  263. //try after re-registering
  264. if token := client.Connect(); !token.WaitTimeout(30*time.Second) || token.Error() != nil {
  265. return client, errors.New("unable to connect to broker")
  266. }
  267. }
  268. return client, nil
  269. }
  270. func reRegisterWithServer(cfg *config.ClientConfig) {
  271. logger.Log(0, "connection issue detected.. attempt connection with new certs and broker information")
  272. key, err := ssl.ReadKeyFromFile(ncutils.GetNetclientPath() + ncutils.GetSeparator() + "client.key")
  273. if err != nil {
  274. _, *key, err = ed25519.GenerateKey(rand.Reader)
  275. if err != nil {
  276. log.Fatal("could not generate new key")
  277. }
  278. }
  279. RegisterWithServer(key, cfg)
  280. daemon.Restart()
  281. }
  282. // publishes a message to server to update peers on this peer's behalf
  283. func publishSignal(nodeCfg *config.ClientConfig, signal byte) error {
  284. if err := publish(nodeCfg, fmt.Sprintf("signal/%s", nodeCfg.Node.ID), []byte{signal}, 1); err != nil {
  285. return err
  286. }
  287. return nil
  288. }
  289. func parseNetworkFromTopic(topic string) string {
  290. return strings.Split(topic, "/")[1]
  291. }
  292. // should only ever use node client configs
  293. func decryptMsg(nodeCfg *config.ClientConfig, msg []byte) ([]byte, error) {
  294. if len(msg) <= 24 { // make sure message is of appropriate length
  295. return nil, fmt.Errorf("recieved invalid message from broker %v", msg)
  296. }
  297. // setup the keys
  298. diskKey, keyErr := auth.RetrieveTrafficKey(nodeCfg.Node.Network)
  299. if keyErr != nil {
  300. return nil, keyErr
  301. }
  302. serverPubKey, err := ncutils.ConvertBytesToKey(nodeCfg.Node.TrafficKeys.Server)
  303. if err != nil {
  304. return nil, err
  305. }
  306. return ncutils.DeChunk(msg, serverPubKey, diskKey)
  307. }
  308. // == Message Caches ==
  309. func insert(network, which, cache string) {
  310. var newMessage = cachedMessage{
  311. Message: cache,
  312. LastSeen: time.Now(),
  313. }
  314. messageCache.Store(fmt.Sprintf("%s%s", network, which), newMessage)
  315. }
  316. func read(network, which string) string {
  317. val, isok := messageCache.Load(fmt.Sprintf("%s%s", network, which))
  318. if isok {
  319. var readMessage = val.(cachedMessage) // fetch current cached message
  320. if readMessage.LastSeen.IsZero() {
  321. return ""
  322. }
  323. if time.Now().After(readMessage.LastSeen.Add(time.Hour * 24)) { // check if message has been there over a minute
  324. messageCache.Delete(fmt.Sprintf("%s%s", network, which)) // remove old message if expired
  325. return ""
  326. }
  327. return readMessage.Message // return current message if not expired
  328. }
  329. return ""
  330. }