fix: harden windows gateway liveness (#658)

bin/hermes-web-ui.mjs

@@ -2,7 +2,7 @@
 import { spawn, execSync } from 'child_process'
 import { resolve, dirname, join } from 'path'
 import { fileURLToPath } from 'url'
-import { readFileSync, writeFileSync, unlinkSync, mkdirSync, openSync, chmodSync, statSync } from 'fs'
+import { readFileSync, writeFileSync, unlinkSync, mkdirSync, openSync, chmodSync, statSync, existsSync } from 'fs'
 import { randomBytes } from 'crypto'
 import { homedir } from 'os'
 
@@ -61,7 +61,18 @@ function getCliBin() {
 }
 
 function getWindowsShell() {
-  return process.env.ComSpec || 'cmd.exe'
+  const systemRoot = process.env.SystemRoot || 'C:\\Windows'
+  const candidates = [
+    process.env.ComSpec,
+    join(systemRoot, 'System32', 'WindowsPowerShell', 'v1.0', 'powershell.exe'),
+    join(systemRoot, 'System32', 'cmd.exe'),
+  ].filter(Boolean)
+
+  for (const candidate of candidates) {
+    if (existsSync(candidate)) return candidate
+  }
+
+  return 'cmd.exe'
 }
 
 function quoteForWindowsCommand(value) {
@@ -70,6 +81,11 @@ function quoteForWindowsCommand(value) {
 
 function spawnCli(command, args, options) {
   if (process.platform === 'win32') {
+    const lowerCommand = String(command).toLowerCase()
+    if (!lowerCommand.endsWith('.cmd') && !lowerCommand.endsWith('.bat')) {
+      return spawn(command, args, options)
+    }
+
     const commandLine = `${quoteForWindowsCommand(command)} ${args.map(arg => String(arg)).join(' ')}`
     return spawn(getWindowsShell(), ['/d', '/s', '/c', commandLine], options)
   }
@@ -123,20 +139,62 @@ function getPort() {
   return argPort ?? DEFAULT_PORT
 }
 
-function getPid() {
+function getListeningPids(port) {
+  if (!port || isNaN(port)) return []
+
   try {
-    return parseInt(readFileSync(PID_FILE, 'utf-8').trim())
+    if (process.platform === 'win32') {
+      const out = execSync('netstat -aon -p tcp', { encoding: 'utf-8' })
+      return [...new Set(out.split('\n')
+        .map(line => line.trim())
+        .filter(line => line.includes('LISTENING'))
+        .map(line => line.split(/\s+/))
+        .filter(parts => {
+          const address = parts[1] || ''
+          const listenPort = parseInt(address.split(':').pop(), 10)
+          return listenPort === port
+        })
+        .map(parts => parseInt(parts[parts.length - 1], 10))
+        .filter(pid => Number.isFinite(pid)))]
+    }
+
+    const out = execSync(`lsof -tiTCP:${port} -sTCP:LISTEN`, { encoding: 'utf-8' }).trim()
+    return [...new Set(out.split('\n').map(pid => parseInt(pid, 10)).filter(pid => Number.isFinite(pid)))]
   } catch {
-    return null
+    return []
   }
 }
 
+function recoverPidFromPort() {
+  const port = getPortFromArgs() ?? DEFAULT_PORT
+  for (const pid of getListeningPids(port)) {
+    if (isRunning(pid)) {
+      mkdirSync(PID_DIR, { recursive: true })
+      writePid(pid)
+      return pid
+    }
+  }
+  return null
+}
+
+function getPid() {
+  const recovered = recoverPidFromPort()
+  if (recovered) return recovered
+
+  try {
+    const pid = parseInt(readFileSync(PID_FILE, 'utf-8').trim())
+    if (pid && isRunning(pid)) return pid
+  } catch {}
+
+  return null
+}
+
 function isRunning(pid) {
   try {
     process.kill(pid, 0)
     return true
-  } catch {
-    return false
+  } catch (err) {
+    return err?.code === 'EPERM'
   }
 }
 
@@ -202,10 +260,16 @@ function startDaemon(port) {
   } catch { }
 
   const logStream = openSync(LOG_FILE, 'a')
+  const windowsShell = process.platform === 'win32' ? getWindowsShell() : null
+  const serverEnv = { ...process.env, NODE_ENV: 'production', PORT: String(port), AUTH_TOKEN: token }
+  if (windowsShell) {
+    serverEnv.SHELL = serverEnv.SHELL?.trim() || windowsShell
+    serverEnv.ComSpec = serverEnv.ComSpec?.trim() || windowsShell
+  }
   const child = spawn(process.execPath, [serverEntry], {
     detached: true,
     stdio: ['ignore', logStream, logStream],
-    env: { ...process.env, NODE_ENV: 'production', PORT: String(port), AUTH_TOKEN: token },
+    env: serverEnv,
     windowsHide: true,
   })
 
@@ -391,9 +455,19 @@ switch (command) {
   default:
     ensureNativeModules()
     const port = !isNaN(command) ? parseInt(command) : DEFAULT_PORT
+    const windowsShell = process.platform === 'win32' ? getWindowsShell() : null
+    const serverEnv = {
+      ...process.env,
+      NODE_ENV: 'production',
+      PORT: String(port),
+    }
+    if (windowsShell) {
+      serverEnv.SHELL = serverEnv.SHELL?.trim() || windowsShell
+      serverEnv.ComSpec = serverEnv.ComSpec?.trim() || windowsShell
+    }
     const child = spawn(process.execPath, [serverEntry], {
       stdio: 'inherit',
-      env: { ...process.env, NODE_ENV: 'production', PORT: String(port) },
+      env: serverEnv,
       windowsHide: true,
     })
     child.on('exit', (code) => process.exit(code ?? 1))

packages/server/src/services/hermes/gateway-manager.ts

@@ -200,15 +200,28 @@ export class GatewayManager {
     }
   }
 
-  /** 从 profile 的 gateway.pid 文件读取 PID（JSON 格式 { "pid": 12345 }） */
+  /** Read a profile gateway PID, falling back to runtime state when gateway.pid is missing. */
   private readPidFile(name: string): number | null {
-    const pidPath = join(this.profileDir(name), 'gateway.pid')
-    if (!existsSync(pidPath)) return null
+    const profilePath = this.profileDir(name)
+    const pidPath = join(profilePath, 'gateway.pid')
 
     try {
-      const content = readFileSync(pidPath, 'utf-8').trim()
+      if (existsSync(pidPath)) {
+        const content = readFileSync(pidPath, 'utf-8').trim()
+        const data = JSON.parse(content)
+        return typeof data.pid === 'number' ? data.pid : parseInt(data.pid, 10) || null
+      }
+    } catch {}
+
+    const statePath = join(profilePath, 'gateway_state.json')
+    if (!existsSync(statePath)) return null
+
+    try {
+      const content = readFileSync(statePath, 'utf-8').trim()
       const data = JSON.parse(content)
-      return typeof data.pid === 'number' ? data.pid : parseInt(data.pid, 10) || null
+      const pid = typeof data.pid === 'number' ? data.pid : parseInt(data.pid, 10) || null
+      const state = data?.gateway_state
+      return pid && (state === 'running' || state === 'starting') ? pid : null
     } catch {
       return null
     }
@@ -218,13 +231,13 @@ export class GatewayManager {
   // 进程 & 端口检测工具
   // ============================
 
-  /** 检查进程是否存活（发送信号 0，不实际杀死进程） */
+  /** Check process liveness without sending a terminating signal. */
   private isProcessAlive(pid: number): boolean {
     try {
       process.kill(pid, 0)
       return true
-    } catch {
-      return false
+    } catch (err: any) {
+      return err?.code === 'EPERM'
     }
   }
 

tests/server/gateway-manager.test.ts

@@ -0,0 +1,97 @@
+import { mkdtempSync, mkdirSync, rmSync, writeFileSync } from 'fs'
+import { tmpdir } from 'os'
+import { join } from 'path'
+import { afterEach, describe, expect, it, vi } from 'vitest'
+
+const originalHermesHome = process.env.HERMES_HOME
+const tempHomes: string[] = []
+
+function createHermesHome(): string {
+  const home = mkdtempSync(join(tmpdir(), 'hermes-web-ui-gateway-'))
+  tempHomes.push(home)
+  return home
+}
+
+async function createManager(home: string): Promise<any> {
+  process.env.HERMES_HOME = home
+  vi.resetModules()
+  const { GatewayManager } = await import('../../packages/server/src/services/hermes/gateway-manager')
+  return new GatewayManager('default') as any
+}
+
+afterEach(() => {
+  vi.restoreAllMocks()
+  vi.resetModules()
+  if (originalHermesHome === undefined) {
+    delete process.env.HERMES_HOME
+  } else {
+    process.env.HERMES_HOME = originalHermesHome
+  }
+
+  for (const home of tempHomes.splice(0)) {
+    rmSync(home, { recursive: true, force: true })
+  }
+})
+
+describe('GatewayManager Windows process recovery', () => {
+  it('treats EPERM from process.kill(pid, 0) as an alive process', async () => {
+    const manager = await createManager(createHermesHome())
+    ;(vi.spyOn(process, 'kill') as any).mockImplementation(() => {
+      const error = new Error('permission denied') as NodeJS.ErrnoException
+      error.code = 'EPERM'
+      throw error
+    })
+
+    expect(manager.isProcessAlive(12345)).toBe(true)
+  })
+
+  it('returns false for missing processes', async () => {
+    const manager = await createManager(createHermesHome())
+    ;(vi.spyOn(process, 'kill') as any).mockImplementation(() => {
+      const error = new Error('missing process') as NodeJS.ErrnoException
+      error.code = 'ESRCH'
+      throw error
+    })
+
+    expect(manager.isProcessAlive(12345)).toBe(false)
+  })
+
+  it('prefers gateway.pid when PID metadata exists', async () => {
+    const home = createHermesHome()
+    writeFileSync(join(home, 'gateway.pid'), JSON.stringify({ pid: 11111 }))
+    writeFileSync(join(home, 'gateway_state.json'), JSON.stringify({ pid: 22222, gateway_state: 'running' }))
+
+    const manager = await createManager(home)
+
+    expect(manager.readPidFile('default')).toBe(11111)
+  })
+
+  it('falls back to gateway_state.json when gateway.pid is missing', async () => {
+    const home = createHermesHome()
+    writeFileSync(join(home, 'gateway_state.json'), JSON.stringify({ pid: '22222', gateway_state: 'running' }))
+
+    const manager = await createManager(home)
+
+    expect(manager.readPidFile('default')).toBe(22222)
+  })
+
+  it('does not use gateway_state.json for stopped gateways', async () => {
+    const home = createHermesHome()
+    writeFileSync(join(home, 'gateway_state.json'), JSON.stringify({ pid: 22222, gateway_state: 'stopped' }))
+
+    const manager = await createManager(home)
+
+    expect(manager.readPidFile('default')).toBeNull()
+  })
+
+  it('uses profile-scoped gateway_state.json fallback', async () => {
+    const home = createHermesHome()
+    const profileHome = join(home, 'profiles', 'work')
+    mkdirSync(profileHome, { recursive: true })
+    writeFileSync(join(profileHome, 'gateway_state.json'), JSON.stringify({ pid: 33333, gateway_state: 'starting' }))
+
+    const manager = await createManager(home)
+
+    expect(manager.readPidFile('work')).toBe(33333)
+  })
+})