service_windows.go 8.5 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398
  1. package main
  2. import (
  3. "bytes"
  4. "context"
  5. "errors"
  6. "fmt"
  7. "io"
  8. "os"
  9. "os/exec"
  10. "path/filepath"
  11. "time"
  12. "github.com/sirupsen/logrus"
  13. "github.com/containerd/containerd/log"
  14. "github.com/spf13/pflag"
  15. "golang.org/x/sys/windows"
  16. "golang.org/x/sys/windows/svc"
  17. "golang.org/x/sys/windows/svc/debug"
  18. "golang.org/x/sys/windows/svc/eventlog"
  19. "golang.org/x/sys/windows/svc/mgr"
  20. )
  21. var (
  22. flServiceName *string
  23. flRegisterService *bool
  24. flUnregisterService *bool
  25. flRunService *bool
  26. oldStderr windows.Handle
  27. panicFile *os.File
  28. service *handler
  29. )
  30. const (
  31. // These should match the values in event_messages.mc.
  32. eventInfo = 1
  33. eventWarn = 1
  34. eventError = 1
  35. eventDebug = 2
  36. eventPanic = 3
  37. eventFatal = 4
  38. eventExtraOffset = 10 // Add this to any event to get a string that supports extended data
  39. )
  40. func installServiceFlags(flags *pflag.FlagSet) {
  41. flServiceName = flags.String("service-name", "docker", "Set the Windows service name")
  42. flRegisterService = flags.Bool("register-service", false, "Register the service and exit")
  43. flUnregisterService = flags.Bool("unregister-service", false, "Unregister the service and exit")
  44. flRunService = flags.Bool("run-service", false, "")
  45. _ = flags.MarkHidden("run-service")
  46. }
  47. type handler struct {
  48. tosvc chan bool
  49. fromsvc chan error
  50. daemonCli *DaemonCli
  51. }
  52. type etwHook struct {
  53. log *eventlog.Log
  54. }
  55. func (h *etwHook) Levels() []logrus.Level {
  56. return []logrus.Level{
  57. logrus.PanicLevel,
  58. logrus.FatalLevel,
  59. logrus.ErrorLevel,
  60. logrus.WarnLevel,
  61. logrus.InfoLevel,
  62. logrus.DebugLevel,
  63. }
  64. }
  65. func (h *etwHook) Fire(e *logrus.Entry) error {
  66. var (
  67. etype uint16
  68. eid uint32
  69. )
  70. switch e.Level {
  71. case logrus.PanicLevel:
  72. etype = windows.EVENTLOG_ERROR_TYPE
  73. eid = eventPanic
  74. case logrus.FatalLevel:
  75. etype = windows.EVENTLOG_ERROR_TYPE
  76. eid = eventFatal
  77. case logrus.ErrorLevel:
  78. etype = windows.EVENTLOG_ERROR_TYPE
  79. eid = eventError
  80. case logrus.WarnLevel:
  81. etype = windows.EVENTLOG_WARNING_TYPE
  82. eid = eventWarn
  83. case logrus.InfoLevel:
  84. etype = windows.EVENTLOG_INFORMATION_TYPE
  85. eid = eventInfo
  86. case logrus.DebugLevel:
  87. etype = windows.EVENTLOG_INFORMATION_TYPE
  88. eid = eventDebug
  89. default:
  90. return errors.New("unknown level")
  91. }
  92. // If there is additional data, include it as a second string.
  93. exts := ""
  94. if len(e.Data) > 0 {
  95. fs := bytes.Buffer{}
  96. for k, v := range e.Data {
  97. fs.WriteString(k)
  98. fs.WriteByte('=')
  99. fmt.Fprint(&fs, v)
  100. fs.WriteByte(' ')
  101. }
  102. exts = fs.String()[:fs.Len()-1]
  103. eid += eventExtraOffset
  104. }
  105. if h.log == nil {
  106. fmt.Fprintf(os.Stderr, "%s [%s]\n", e.Message, exts)
  107. return nil
  108. }
  109. var (
  110. ss [2]*uint16
  111. err error
  112. )
  113. ss[0], err = windows.UTF16PtrFromString(e.Message)
  114. if err != nil {
  115. return err
  116. }
  117. count := uint16(1)
  118. if exts != "" {
  119. ss[1], err = windows.UTF16PtrFromString(exts)
  120. if err != nil {
  121. return err
  122. }
  123. count++
  124. }
  125. return windows.ReportEvent(h.log.Handle, etype, 0, eid, 0, count, 0, &ss[0], nil)
  126. }
  127. func getServicePath() (string, error) {
  128. p, err := exec.LookPath(os.Args[0])
  129. if err != nil {
  130. return "", err
  131. }
  132. return filepath.Abs(p)
  133. }
  134. func registerService() error {
  135. p, err := getServicePath()
  136. if err != nil {
  137. return err
  138. }
  139. m, err := mgr.Connect()
  140. if err != nil {
  141. return err
  142. }
  143. defer m.Disconnect()
  144. c := mgr.Config{
  145. ServiceType: windows.SERVICE_WIN32_OWN_PROCESS,
  146. StartType: mgr.StartAutomatic,
  147. ErrorControl: mgr.ErrorNormal,
  148. Dependencies: []string{},
  149. DisplayName: "Docker Engine",
  150. }
  151. // Configure the service to launch with the arguments that were just passed.
  152. args := []string{"--run-service"}
  153. for _, a := range os.Args[1:] {
  154. if a != "--register-service" && a != "--unregister-service" {
  155. args = append(args, a)
  156. }
  157. }
  158. s, err := m.CreateService(*flServiceName, p, c, args...)
  159. if err != nil {
  160. return err
  161. }
  162. defer s.Close()
  163. err = s.SetRecoveryActions(
  164. []mgr.RecoveryAction{
  165. {Type: mgr.ServiceRestart, Delay: 15 * time.Second},
  166. {Type: mgr.ServiceRestart, Delay: 15 * time.Second},
  167. {Type: mgr.NoAction},
  168. },
  169. uint32(24*time.Hour/time.Second),
  170. )
  171. if err != nil {
  172. return err
  173. }
  174. return eventlog.Install(*flServiceName, p, false, eventlog.Info|eventlog.Warning|eventlog.Error)
  175. }
  176. func unregisterService() error {
  177. m, err := mgr.Connect()
  178. if err != nil {
  179. return err
  180. }
  181. defer m.Disconnect()
  182. s, err := m.OpenService(*flServiceName)
  183. if err != nil {
  184. return err
  185. }
  186. defer s.Close()
  187. eventlog.Remove(*flServiceName)
  188. err = s.Delete()
  189. if err != nil {
  190. return err
  191. }
  192. return nil
  193. }
  194. // initService is the entry point for running the daemon as a Windows
  195. // service. It returns an indication to stop (if registering/un-registering);
  196. // an indication of whether it is running as a service; and an error.
  197. func initService(daemonCli *DaemonCli) (bool, bool, error) {
  198. if *flUnregisterService {
  199. if *flRegisterService {
  200. return true, false, errors.New("--register-service and --unregister-service cannot be used together")
  201. }
  202. return true, false, unregisterService()
  203. }
  204. if *flRegisterService {
  205. return true, false, registerService()
  206. }
  207. if !*flRunService {
  208. return false, false, nil
  209. }
  210. // Check if we're running as a Windows service or interactively.
  211. isService, err := svc.IsWindowsService()
  212. if err != nil {
  213. return false, false, err
  214. }
  215. h := &handler{
  216. tosvc: make(chan bool),
  217. fromsvc: make(chan error),
  218. daemonCli: daemonCli,
  219. }
  220. var log *eventlog.Log
  221. if isService {
  222. log, err = eventlog.Open(*flServiceName)
  223. if err != nil {
  224. return false, false, err
  225. }
  226. }
  227. logrus.AddHook(&etwHook{log})
  228. logrus.SetOutput(io.Discard)
  229. service = h
  230. go func() {
  231. if isService {
  232. err = svc.Run(*flServiceName, h)
  233. } else {
  234. err = debug.Run(*flServiceName, h)
  235. }
  236. h.fromsvc <- err
  237. }()
  238. // Wait for the first signal from the service handler.
  239. err = <-h.fromsvc
  240. if err != nil {
  241. return false, false, err
  242. }
  243. return false, true, nil
  244. }
  245. func (h *handler) started() error {
  246. // This must be delayed until daemonCli initializes Config.Root
  247. err := initPanicFile(filepath.Join(h.daemonCli.Config.Root, "panic.log"))
  248. if err != nil {
  249. return err
  250. }
  251. h.tosvc <- false
  252. return nil
  253. }
  254. func (h *handler) stopped(err error) {
  255. log.G(context.TODO()).Debugf("Stopping service: %v", err)
  256. h.tosvc <- err != nil
  257. <-h.fromsvc
  258. }
  259. func (h *handler) Execute(_ []string, r <-chan svc.ChangeRequest, s chan<- svc.Status) (bool, uint32) {
  260. s <- svc.Status{State: svc.StartPending, Accepts: 0}
  261. // Unblock initService()
  262. h.fromsvc <- nil
  263. // Wait for initialization to complete.
  264. failed := <-h.tosvc
  265. if failed {
  266. log.G(context.TODO()).Debug("Aborting service start due to failure during initialization")
  267. return true, 1
  268. }
  269. s <- svc.Status{State: svc.Running, Accepts: svc.AcceptStop | svc.AcceptShutdown | svc.Accepted(windows.SERVICE_ACCEPT_PARAMCHANGE)}
  270. log.G(context.TODO()).Debug("Service running")
  271. Loop:
  272. for {
  273. select {
  274. case failed = <-h.tosvc:
  275. break Loop
  276. case c := <-r:
  277. switch c.Cmd {
  278. case svc.Cmd(windows.SERVICE_CONTROL_PARAMCHANGE):
  279. h.daemonCli.reloadConfig()
  280. case svc.Interrogate:
  281. s <- c.CurrentStatus
  282. case svc.Stop, svc.Shutdown:
  283. s <- svc.Status{State: svc.StopPending, Accepts: 0}
  284. h.daemonCli.stop()
  285. }
  286. }
  287. }
  288. removePanicFile()
  289. if failed {
  290. return true, 1
  291. }
  292. return false, 0
  293. }
  294. func initPanicFile(path string) error {
  295. var err error
  296. panicFile, err = os.OpenFile(path, os.O_CREATE|os.O_WRONLY|os.O_APPEND, 0o200)
  297. if err != nil {
  298. return err
  299. }
  300. st, err := panicFile.Stat()
  301. if err != nil {
  302. return err
  303. }
  304. // If there are contents in the file already, move the file out of the way
  305. // and replace it.
  306. if st.Size() > 0 {
  307. panicFile.Close()
  308. os.Rename(path, path+".old")
  309. panicFile, err = os.Create(path)
  310. if err != nil {
  311. return err
  312. }
  313. }
  314. // Update STD_ERROR_HANDLE to point to the panic file so that Go writes to
  315. // it when it panics. Remember the old stderr to restore it before removing
  316. // the panic file.
  317. h, err := windows.GetStdHandle(windows.STD_ERROR_HANDLE)
  318. if err != nil {
  319. return err
  320. }
  321. oldStderr = h
  322. err = windows.SetStdHandle(windows.STD_ERROR_HANDLE, windows.Handle(panicFile.Fd()))
  323. if err != nil {
  324. return err
  325. }
  326. // Reset os.Stderr to the panic file (so fmt.Fprintf(os.Stderr,...) actually gets redirected)
  327. os.Stderr = os.NewFile(panicFile.Fd(), "/dev/stderr")
  328. // Force threads that panic to write to stderr (the panicFile handle now), otherwise it will go into the ether
  329. logrus.SetOutput(os.Stderr)
  330. return nil
  331. }
  332. func removePanicFile() {
  333. if st, err := panicFile.Stat(); err == nil {
  334. if st.Size() == 0 {
  335. windows.SetStdHandle(windows.STD_ERROR_HANDLE, oldStderr)
  336. panicFile.Close()
  337. os.Remove(panicFile.Name())
  338. }
  339. }
  340. }