sshd.go 11 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423
  1. package sshd
  2. import (
  3. "context"
  4. "errors"
  5. "os"
  6. "strings"
  7. "sync"
  8. "sync/atomic"
  9. "time"
  10. mqtt "github.com/eclipse/paho.mqtt.golang"
  11. "hnyfkj.com.cn/rtu/linux/baseapp"
  12. "hnyfkj.com.cn/rtu/linux/netmgrd"
  13. "hnyfkj.com.cn/rtu/linux/utils/jsonrpc2"
  14. "hnyfkj.com.cn/rtu/linux/utils/shell"
  15. "hnyfkj.com.cn/rtu/linux/utils/singletask"
  16. )
  17. const MODULE_NAME = "YFKJ_SSHD"
  18. var (
  19. coupler *MQTTCoupler
  20. )
  21. const (
  22. MqttQos1 byte = 1 //// 消息至少送达一次
  23. FastInterval = 1 * time.Second //// 快速检测时间间隔
  24. SlowInterval = 5 * time.Second //// 慢速检测时间间隔
  25. ExecutorCheckInterval = 2 * time.Second // 执行器回收检测
  26. ExecutorTimeout = 6 * time.Second // 执行器超时时间
  27. )
  28. var (
  29. ErrBrokerAddressEmpty = errors.New("mqtt server address is empty")
  30. ErrIMEINotAvailable = errors.New("device imei is not available")
  31. )
  32. type MQTTCoupler struct {
  33. ctx context.Context
  34. cancel context.CancelFunc
  35. broker, username, password string
  36. client mqtt.Client /// MQTT客户端
  37. isConnected atomic.Bool /// 标记是否已连接MQTT的Broker服务
  38. imei string // 设备唯一标识
  39. subTopic string // 订阅指令主题:/yfkj/device/rpc/imei/cmd
  40. pubTopic string // 发布应答主题:/yfkj/device/rpc/imei/ack
  41. ///////// 本地执行器, 允许多客户端, 同一客户端串行的执行指令
  42. executorMap map[string]*clientExecutor
  43. executorMapMu sync.Mutex
  44. // 注册本地的远程方法, 连接成功后用于让客户端能够主动下发指令
  45. registerRpcMeths *singletask.OnceTask // 注册方法, 单实例
  46. }
  47. type executorState int
  48. const (
  49. execIdle executorState = iota // 空闲状态时, 可安全回收
  50. execRunning // 正在执行时, 不允许回收
  51. execClosing // 表明执行器正在关闭中..
  52. )
  53. type clientExecutor struct {
  54. id string /////////////////// 客户端唯一ID
  55. executor *shell.Executor /////////////////// 本地的执行器
  56. mu sync.Mutex /////////////////// 同ID串行执行
  57. lastPing time.Time /////////////////// 用于超时回收
  58. state executorState /////////////////// 执行器的状态
  59. }
  60. func ModuleInit(mqttBroker, mqttUsername, mqttPassword string) bool {
  61. if mqttBroker == "" {
  62. baseapp.Logger.Errorf("[%s] 初始化远程运维模块失败: %v!!", MODULE_NAME, ErrBrokerAddressEmpty)
  63. return false
  64. }
  65. ctx, cancel := context.WithCancel(context.Background())
  66. coupler = &MQTTCoupler{
  67. ctx: ctx,
  68. cancel: cancel,
  69. broker: mqttBroker,
  70. username: mqttUsername,
  71. password: mqttPassword,
  72. executorMap: make(map[string]*clientExecutor),
  73. registerRpcMeths: &singletask.OnceTask{},
  74. }
  75. if err := coupler.init2(); err != nil {
  76. baseapp.Logger.Errorf("[%s] 初始化远程运维模块失败: %v!!", MODULE_NAME, err)
  77. return false
  78. }
  79. go coupler.startExecutorReaper(ExecutorCheckInterval, ExecutorTimeout)
  80. go coupler.keepOnline()
  81. return true
  82. }
  83. func ModuleExit() {
  84. if coupler != nil {
  85. coupler.cancel()
  86. }
  87. }
  88. func (c *MQTTCoupler) init2() error {
  89. imeiBytes, _ := os.ReadFile("/var/device_imei.txt")
  90. c.imei = strings.TrimSpace(string(imeiBytes))
  91. if c.imei == netmgrd.ErrUnknownModemTypeMsg || c.imei == "" {
  92. return ErrIMEINotAvailable
  93. }
  94. baseapp.Logger.Infof("[%s] ☺✔设备IMEI: %s", MODULE_NAME, c.imei)
  95. template := "/yfkj/device/rpc/imei"
  96. c.subTopic = strings.ReplaceAll(template+"/cmd", "imei", c.imei)
  97. c.pubTopic = strings.ReplaceAll(template+"/ack", "imei", c.imei)
  98. opts := mqtt.NewClientOptions().
  99. AddBroker(c.broker).
  100. SetUsername(c.username).SetPassword(c.password).
  101. SetConnectRetry(false).SetAutoReconnect(false).SetCleanSession(true).
  102. SetKeepAlive(10*time.Second).SetPingTimeout(5*time.Second). // Ping心跳间隔, 超时时间
  103. SetOrderMatters(false). /*离线遗愿消息*/ SetWill(c.pubTopic, string(`{"jsonrpc": "2.0", "method": "logout"}`), MqttQos1, false)
  104. opts.OnConnect = func(client mqtt.Client) {
  105. if !c.isConnected.Swap(true) {
  106. baseapp.Logger.Infof("[%s] MQTT Broker连接成功", MODULE_NAME)
  107. c.registerRpcMeths.Run(c.instRPCMethods, true) // 注册本地的RPC方法, 供远端调用, 单实例运行
  108. }
  109. }
  110. opts.OnConnectionLost = func(client mqtt.Client, err error) {
  111. if c.isConnected.Swap(false) {
  112. baseapp.Logger.Warnf("[%s] MQTT Broker连接丢失: %v!", MODULE_NAME, err)
  113. }
  114. }
  115. c.client = mqtt.NewClient(opts)
  116. return nil
  117. }
  118. func (c *MQTTCoupler) keepOnline() {
  119. t := time.NewTimer(FastInterval)
  120. defer t.Stop()
  121. for {
  122. select {
  123. case <-c.ctx.Done():
  124. return
  125. case <-t.C:
  126. t.Reset(c.tick())
  127. } // end select
  128. } // end for
  129. }
  130. func (c *MQTTCoupler) tick() time.Duration {
  131. if c.isConnected.Load() {
  132. return FastInterval
  133. }
  134. if err := c.connect(); err != nil {
  135. baseapp.Logger.Errorf("[%s] MQTT Broker连接失败: %v!!", MODULE_NAME, err)
  136. }
  137. return SlowInterval
  138. }
  139. func (c *MQTTCoupler) connect() error {
  140. if c.client.IsConnected() {
  141. return nil
  142. }
  143. token := c.client.Connect()
  144. select {
  145. case <-c.ctx.Done():
  146. return nil
  147. case <-token.Done():
  148. }
  149. return token.Error()
  150. }
  151. func (c *MQTTCoupler) instRPCMethods() {
  152. t := time.NewTicker(time.Second)
  153. defer t.Stop()
  154. for {
  155. if !c.isConnected.Load() || c.ctx.Err() != nil {
  156. return
  157. }
  158. token := c.client.Subscribe(c.subTopic, MqttQos1, c.handleRequests)
  159. select {
  160. case <-c.ctx.Done():
  161. return
  162. case <-token.Done():
  163. }
  164. if token.Error() == nil {
  165. baseapp.Logger.Infof("[%s] 本地RPC方法已注册, 等待远端调用...", MODULE_NAME)
  166. break
  167. }
  168. select {
  169. case <-c.ctx.Done():
  170. return
  171. case <-t.C:
  172. continue
  173. }
  174. }
  175. }
  176. func (c *MQTTCoupler) handleRequests(client mqtt.Client, msg mqtt.Message) {
  177. go c.execOneCmd(msg)
  178. }
  179. func (c *MQTTCoupler) execOneCmd(msg mqtt.Message) {
  180. str := string(msg.Payload())
  181. baseapp.Logger.Debugf("[%s] 收到一个RPC请求: %s", MODULE_NAME, str)
  182. var resp *jsonrpc2.Response // 预先定义一个空的应答
  183. var clientID string // 该客户端的|唯一标识|
  184. var ce *clientExecutor // 该客户端的本地执行器
  185. var exists bool // 判断执行器是否已存在
  186. req, err := jsonrpc2.ParseRequest(str)
  187. if err != nil || req.ID == nil /* 不接受通知类型的消息 */ {
  188. resp = jsonrpc2.BuildError(nil, jsonrpc2.ErrParse, "")
  189. goto retp
  190. }
  191. clientID, err = extractClientID(req.Params)
  192. if err != nil {
  193. resp = jsonrpc2.BuildError(req, jsonrpc2.ErrInvalidParams, err.Error())
  194. goto retp
  195. }
  196. c.executorMapMu.Lock()
  197. ce, exists = c.executorMap[clientID]
  198. if !exists {
  199. if len(c.executorMap) >= 3 {
  200. c.executorMapMu.Unlock()
  201. resp = jsonrpc2.BuildError(req, -32000, "connection refused: server has reached maximum client capacity (3/3)")
  202. goto retp
  203. }
  204. ce = &clientExecutor{
  205. id: clientID,
  206. executor: shell.NewExecutor(),
  207. state: execIdle,
  208. }
  209. c.executorMap[clientID] = ce
  210. baseapp.Logger.Infof("[%s] 客户端 %s 登录成功", MODULE_NAME, clientID)
  211. }
  212. c.executorMapMu.Unlock()
  213. ce.mu.Lock()
  214. ce.lastPing = time.Now()
  215. ce.mu.Unlock()
  216. switch req.Method {
  217. // Call-1: 心跳, 链路检测,"ping-pong"测试
  218. case "executor.ping":
  219. resp = buildResp(req, "pong", nil)
  220. goto retp
  221. // Call-2:在本地shell中执行远程下发的指令
  222. case "executor.exec":
  223. ce.mu.Lock()
  224. params, err := extractShellExecuteParams(req.Params)
  225. if err != nil {
  226. ce.mu.Unlock()
  227. resp = jsonrpc2.BuildError(req, jsonrpc2.ErrParse, err.Error())
  228. goto retp
  229. }
  230. if ce.state == execClosing {
  231. ce.mu.Unlock()
  232. resp = jsonrpc2.BuildError(req, -32001, "executor closed")
  233. goto retp
  234. }
  235. if ce.state == execRunning {
  236. ce.mu.Unlock()
  237. resp = jsonrpc2.BuildError(req, -32002, "executor busy")
  238. goto retp
  239. }
  240. ce.state = execRunning
  241. ce.mu.Unlock()
  242. start := time.Now()
  243. if true { //////// 记录执行日志-执行前
  244. baseapp.Logger.Infof("[%s][▷ EXEC] client=%s cmd=%q dir=%q timeout=%ds",
  245. MODULE_NAME, clientID, params.Cmd, params.Dir, params.Timeout)
  246. }
  247. result, err := ce.executor.Exec(params) // 本地执行用户指令
  248. cost := time.Since(start)
  249. if err != nil { // 记录执行日志-执行后
  250. baseapp.Logger.Warnf("[%s][✖ EXEC] client=%s cost=%v cmd=%q err=%v",
  251. MODULE_NAME, clientID, cost, params.Cmd, err)
  252. } else {
  253. baseapp.Logger.Infof("[%s][✔ EXEC] client=%s cost=%v cmd=%q",
  254. MODULE_NAME, clientID, cost, params.Cmd)
  255. }
  256. ce.mu.Lock()
  257. if ce.state != execClosing {
  258. ce.state = execIdle
  259. ce.lastPing = time.Now()
  260. }
  261. ce.mu.Unlock()
  262. resp = buildResp(req, result, err)
  263. goto retp
  264. // Call-3:中断本地shell的执行,等价Ctrl+C
  265. case "executor.interrupt":
  266. ce.mu.Lock()
  267. running := (ce.state == execRunning)
  268. ce.mu.Unlock()
  269. if !running {
  270. resp = jsonrpc2.BuildError(req, -32003, "no running command")
  271. goto retp
  272. }
  273. err := ce.executor.Interrupt()
  274. resp = buildResp(req, "interrupted", err)
  275. goto retp
  276. // Call-4:客户端安全退出, 释放本地的执行器
  277. case "executor.close":
  278. err := ce.handleClose()
  279. c.executorMapMu.Lock()
  280. delete(c.executorMap, clientID)
  281. baseapp.Logger.Infof("[%s] 客户端 %s 退出成功", MODULE_NAME, clientID)
  282. c.executorMapMu.Unlock()
  283. resp = buildResp(req, "closed", err)
  284. goto retp
  285. // Call-?:无效, 远端调用了还不支持的-方法
  286. default:
  287. resp = jsonrpc2.BuildError(req, jsonrpc2.ErrMethodNotFound, "")
  288. goto retp
  289. }
  290. retp:
  291. text, err := resp.String()
  292. if err != nil {
  293. baseapp.Logger.Errorf("[%s] 转换RPC应答失败: %v!!", MODULE_NAME, err)
  294. return
  295. }
  296. token := c.client.Publish(c.pubTopic, MqttQos1, false, text)
  297. select {
  298. case <-c.ctx.Done():
  299. return
  300. case <-token.Done():
  301. }
  302. if err := token.Error(); err != nil {
  303. baseapp.Logger.Errorf("[%s] 发送RPC应答失败: %v!!", MODULE_NAME, err)
  304. }
  305. baseapp.Logger.Debugf("[%s] 发送一个RPC应答, 报文内容: %s", MODULE_NAME, text)
  306. }
  307. func (c *MQTTCoupler) startExecutorReaper(interval, timeout time.Duration) {
  308. ticker := time.NewTicker(interval)
  309. defer ticker.Stop()
  310. for {
  311. select {
  312. case <-c.ctx.Done():
  313. return
  314. case <-ticker.C:
  315. c.executorMapMu.Lock()
  316. for id, ce := range c.executorMap {
  317. ce.mu.Lock()
  318. expired := time.Since(ce.lastPing) > timeout
  319. idle := (ce.state == execIdle)
  320. ce.mu.Unlock()
  321. if expired && idle { // 超时且状态空闲时则回收
  322. ce.handleClose() //// 该函数不能阻塞, 否则锁
  323. delete(c.executorMap, id)
  324. baseapp.Logger.Infof("[%s] 客户端 %s 超时移除", MODULE_NAME, id)
  325. } // end if
  326. } // end for2
  327. c.executorMapMu.Unlock()
  328. } // end select
  329. } ////// end for1
  330. }
  331. func (ce *clientExecutor) handleClose() error {
  332. needInterrupt := false
  333. ce.mu.Lock()
  334. switch ce.state {
  335. case execIdle:
  336. ce.state = execClosing
  337. case execRunning:
  338. ce.state = execClosing
  339. needInterrupt = true
  340. case execClosing:
  341. ce.mu.Unlock()
  342. return nil
  343. }
  344. ce.mu.Unlock()
  345. var err error
  346. if needInterrupt {
  347. err = ce.executor.Interrupt() // 发送"Ctrl+C"
  348. }
  349. return err
  350. }