manager.go 8.2 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334
  1. package main
  2. // 挂载管理(docs/design/21-knowledge-sources.md §4.4、§4.5):
  3. // 平台用 sync 下发完整的登记列表;挂载进程挂上缺少的、卸载多余的、更新有变化的。
  4. // 挂载表变化时(设备插拔)立即重新计算状态,不需要定时检查。
  5. import (
  6. "encoding/json"
  7. "fmt"
  8. "log"
  9. "os"
  10. "os/exec"
  11. "path/filepath"
  12. "regexp"
  13. "sort"
  14. "sync"
  15. "time"
  16. "github.com/hanwen/go-fuse/v2/fuse"
  17. "golang.org/x/sys/unix"
  18. )
  19. var idPattern = regexp.MustCompile(`^kb_[0-9a-z]{1,64}$`)
  20. type mountSpec struct {
  21. ID string `json:"id"`
  22. Kind string `json:"kind"`
  23. Local *localSource `json:"local,omitempty"`
  24. }
  25. type mountStatus struct {
  26. ID string `json:"id"`
  27. State string `json:"state"`
  28. Message *string `json:"message"`
  29. Since string `json:"since"`
  30. // 当前的实际路径(本机目录)
  31. RealPath string `json:"realPath,omitempty"`
  32. }
  33. type mount struct {
  34. spec mountSpec
  35. point string
  36. src *source
  37. server *fuse.Server
  38. state string
  39. message string
  40. since time.Time
  41. realPath string
  42. }
  43. type manager struct {
  44. mu sync.Mutex
  45. sys system
  46. roots []string
  47. mntDir string
  48. stateFile string
  49. mounts map[string]*mount
  50. lastMiss time.Time
  51. // mountFn 便于测试替换
  52. mountFn func(point, name string, src *source) (*fuse.Server, error)
  53. }
  54. func newManager(sys system, roots []string, mntDir, stateFile string) *manager {
  55. return &manager{sys: sys, roots: roots, mntDir: mntDir, stateFile: stateFile, mounts: map[string]*mount{}, mountFn: mountFS}
  56. }
  57. func (m *manager) setState(mt *mount, state, msg string) {
  58. if mt.state != state || mt.message != msg {
  59. if mt.state != "" {
  60. log.Printf("%s:%s → %s %s", mt.spec.ID, mt.state, state, msg)
  61. }
  62. mt.state, mt.message, mt.since = state, msg, time.Now()
  63. }
  64. }
  65. // resolve 按登记找到源目录并检查,返回真实路径与设备号、inode 号
  66. func (m *manager) resolve(spec mountSpec) (string, uint64, uint64, *checkError) {
  67. if spec.Kind != "local" || spec.Local == nil {
  68. return "", 0, 0, &checkError{"mount_failed", "暂不支持这种资料库类型"}
  69. }
  70. p, cerr := locate(m.sys, *spec.Local)
  71. if cerr != nil {
  72. return "", 0, 0, cerr
  73. }
  74. real, cerr := checkSource(p, m.roots)
  75. if cerr != nil {
  76. return "", 0, 0, cerr
  77. }
  78. how := unix.OpenHow{Flags: unix.O_PATH | unix.O_DIRECTORY | unix.O_CLOEXEC, Resolve: unix.RESOLVE_NO_SYMLINKS | unix.RESOLVE_NO_MAGICLINKS}
  79. fd, err := unix.Openat2(unix.AT_FDCWD, real, &how)
  80. if err != nil {
  81. return "", 0, 0, &checkError{"path_missing", "无法打开目录:" + real}
  82. }
  83. defer unix.Close(fd)
  84. var st unix.Stat_t
  85. if err := unix.Fstat(fd, &st); err != nil {
  86. return "", 0, 0, &checkError{"path_missing", "无法读取目录:" + real}
  87. }
  88. return real, st.Dev, st.Ino, nil
  89. }
  90. // refreshLocked 重新计算一个挂载的状态
  91. func (m *manager) refreshLocked(mt *mount) {
  92. real, dev, ino, cerr := m.resolve(mt.spec)
  93. if cerr != nil {
  94. mt.src.close()
  95. mt.realPath = ""
  96. m.setState(mt, cerr.State, cerr.Message)
  97. return
  98. }
  99. mt.src.set(real, dev, ino)
  100. mt.realPath = real
  101. if mt.server == nil {
  102. m.setState(mt, "mount_failed", "挂载失败")
  103. return
  104. }
  105. m.setState(mt, "ok", "")
  106. }
  107. func (m *manager) refreshAll() {
  108. m.mu.Lock()
  109. defer m.mu.Unlock()
  110. for _, mt := range m.mounts {
  111. m.refreshLocked(mt)
  112. }
  113. }
  114. // onMiss 资料库不可用时有人访问:可能是设备刚接上(如自动挂载),限速重新检查
  115. func (m *manager) onMiss() {
  116. m.mu.Lock()
  117. if time.Since(m.lastMiss) < 2*time.Second {
  118. m.mu.Unlock()
  119. return
  120. }
  121. m.lastMiss = time.Now()
  122. m.mu.Unlock()
  123. go m.refreshAll()
  124. }
  125. func (m *manager) mountLocked(mt *mount) {
  126. if mt.server != nil {
  127. return
  128. }
  129. if err := os.MkdirAll(mt.point, 0o755); err != nil {
  130. log.Printf("%s:创建挂载点失败:%v", mt.spec.ID, err)
  131. return
  132. }
  133. srv, err := m.mountFn(mt.point, "kb:"+mt.spec.ID, mt.src)
  134. if err != nil {
  135. log.Printf("%s:挂载失败:%v", mt.spec.ID, err)
  136. return
  137. }
  138. mt.server = srv
  139. }
  140. func (m *manager) unmountLocked(mt *mount) {
  141. if mt.server != nil {
  142. if err := mt.server.Unmount(); err != nil {
  143. // 有文件正被打开:延迟卸载
  144. _ = exec.Command("fusermount3", "-uz", mt.point).Run()
  145. }
  146. mt.server = nil
  147. }
  148. mt.src.close()
  149. _ = os.Remove(mt.point)
  150. }
  151. // sync 按完整列表挂载、卸载、更新
  152. func (m *manager) sync(specs []mountSpec, save bool) ([]mountStatus, error) {
  153. want := map[string]mountSpec{}
  154. for _, s := range specs {
  155. if !idPattern.MatchString(s.ID) {
  156. return nil, fmt.Errorf("资料库 ID 格式不正确:%q", s.ID)
  157. }
  158. if s.Kind == "local" && (s.Local == nil || !filepath.IsAbs(s.Local.Path)) {
  159. return nil, fmt.Errorf("%s:缺少本机目录", s.ID)
  160. }
  161. want[s.ID] = s
  162. }
  163. m.mu.Lock()
  164. for id, mt := range m.mounts {
  165. if _, ok := want[id]; !ok {
  166. m.unmountLocked(mt)
  167. delete(m.mounts, id)
  168. log.Printf("%s:已卸载", id)
  169. }
  170. }
  171. for id, s := range want {
  172. mt := m.mounts[id]
  173. if mt == nil {
  174. mt = &mount{spec: s, point: filepath.Join(m.mntDir, id), src: newSource()}
  175. mt.src.onMiss = m.onMiss
  176. m.mounts[id] = mt
  177. log.Printf("%s:登记 %s", id, describeSpec(s))
  178. } else {
  179. mt.spec = s
  180. }
  181. m.mountLocked(mt)
  182. m.refreshLocked(mt)
  183. }
  184. m.mu.Unlock()
  185. if save {
  186. if err := m.saveState(specs); err != nil {
  187. log.Printf("保存 state.json 失败:%v", err)
  188. }
  189. }
  190. return m.status(), nil
  191. }
  192. func describeSpec(s mountSpec) string {
  193. if s.Local != nil {
  194. if s.Local.UUID != "" {
  195. return fmt.Sprintf("%s(UUID %s,设备内 %s)", s.Local.Path, s.Local.UUID, s.Local.RelPath)
  196. }
  197. return s.Local.Path
  198. }
  199. return s.Kind
  200. }
  201. func (m *manager) status() []mountStatus {
  202. m.mu.Lock()
  203. defer m.mu.Unlock()
  204. out := make([]mountStatus, 0, len(m.mounts))
  205. for id, mt := range m.mounts {
  206. st := mountStatus{ID: id, State: mt.state, Since: mt.since.UTC().Format(time.RFC3339), RealPath: mt.realPath}
  207. if mt.message != "" {
  208. msg := mt.message
  209. st.Message = &msg
  210. }
  211. out = append(out, st)
  212. }
  213. sort.Slice(out, func(i, j int) bool { return out[i].ID < out[j].ID })
  214. return out
  215. }
  216. func (m *manager) unmountAll() {
  217. m.mu.Lock()
  218. defer m.mu.Unlock()
  219. for id, mt := range m.mounts {
  220. m.unmountLocked(mt)
  221. delete(m.mounts, id)
  222. }
  223. }
  224. type stateFile struct {
  225. Mounts []mountSpec `json:"mounts"`
  226. }
  227. func (m *manager) saveState(specs []mountSpec) error {
  228. b, err := json.MarshalIndent(stateFile{Mounts: specs}, "", " ")
  229. if err != nil {
  230. return err
  231. }
  232. tmp := m.stateFile + ".tmp"
  233. if err := os.WriteFile(tmp, b, 0o600); err != nil {
  234. return err
  235. }
  236. return os.Rename(tmp, m.stateFile)
  237. }
  238. func (m *manager) loadState() []mountSpec {
  239. b, err := os.ReadFile(m.stateFile)
  240. if err != nil {
  241. return nil
  242. }
  243. var s stateFile
  244. if json.Unmarshal(b, &s) != nil {
  245. log.Printf("state.json 格式不正确,忽略")
  246. return nil
  247. }
  248. return s.Mounts
  249. }
  250. // cleanupStale 启动时清理 mnt/ 下残留的挂载(上次进程退出后失效)和多余的空目录
  251. func (m *manager) cleanupStale() {
  252. for _, e := range m.sys.mountinfo() {
  253. if filepath.Dir(e.Point) == m.mntDir {
  254. if err := exec.Command("fusermount3", "-uz", e.Point).Run(); err != nil {
  255. log.Printf("清理残留挂载 %s 失败:%v", e.Point, err)
  256. } else {
  257. log.Printf("已清理残留挂载 %s", e.Point)
  258. }
  259. }
  260. }
  261. ents, _ := os.ReadDir(m.mntDir)
  262. for _, e := range ents {
  263. if idPattern.MatchString(e.Name()) {
  264. _ = os.Remove(filepath.Join(m.mntDir, e.Name()))
  265. }
  266. }
  267. }
  268. // watchMountinfo 挂载表变化时(POLLPRI)重新计算状态
  269. func (m *manager) watchMountinfo(stop <-chan struct{}) {
  270. f, err := os.Open("/proc/self/mountinfo")
  271. if err != nil {
  272. log.Printf("无法监听挂载表:%v", err)
  273. return
  274. }
  275. defer f.Close()
  276. buf := make([]byte, 64*1024)
  277. drain := func() {
  278. _, _ = f.Seek(0, 0)
  279. for {
  280. n, err := f.Read(buf)
  281. if n == 0 || err != nil {
  282. return
  283. }
  284. }
  285. }
  286. drain()
  287. fds := []unix.PollFd{{Fd: int32(f.Fd()), Events: unix.POLLPRI}}
  288. for {
  289. select {
  290. case <-stop:
  291. return
  292. default:
  293. }
  294. n, err := unix.Poll(fds, 1000)
  295. if err != nil && err != unix.EINTR {
  296. log.Printf("监听挂载表出错:%v", err)
  297. time.Sleep(time.Second)
  298. continue
  299. }
  300. if n <= 0 || fds[0].Revents&(unix.POLLPRI|unix.POLLERR) == 0 {
  301. continue
  302. }
  303. drain()
  304. // 设备插入时往往连续变化几次,稍等一下再统一计算
  305. time.Sleep(300 * time.Millisecond)
  306. drain()
  307. m.refreshAll()
  308. }
  309. }