fix: P0+P1 critical bug fixes across controller (24 files)

Concurrency fixes:
- Deep-copy stacks in GetStack/GetStacks to prevent shared state mutation (C04)
- Add per-state mutex to watchdog pathProbeState (C05)
- Guard MetricsCollector.Start() with sync.Once against double-start (C06)
- Hold diskJobMu across entire raw mount operation (C07)
- Add mutex to SetEncryptionKey (C08), MigrateEncryption write lock (H03)
- Use sync.Once for sync.Stop() channel close (H08)
- Set syncing=true before releasing lock in TriggerSync (H09)
- Deep-copy lastDBDump/lastBackup in GetFullStatus (H11)
- Add WaitGroup for stderr goroutine in MigrateDrive (H19)
- Add mutex to SetBackupRunningCheck (M18)

Security fixes:
- Validate Bearer token against Hub API key in CSRF middleware (H16)
- Validate backup paths start with expected prefix in RemoveStack (M12)
- Guard uuid[:8] slice with length check (H20)
- Parse fstab fields exactly for mount target matching (H21)

Bug fixes:
- Use decrypted env vars for compose deploy (C01)
- Log decrypt failures in DecryptMap instead of swallowing (C02)
- Move Deployed=false inside lock in runComposeDeploy (C03)
- Fix activeDrives() to skip disconnected drives (H02)
- Fix Snapshot() stderr extraction from exec.ExitError (H01)
- Check unlockCmd.Run() error in restic (H01)
- Buffer template rendering via bytes.Buffer (H07)
- Thread context.Context through cloudflare client (H10)
- Fix leaf-name collision detection in cross-drive backup (H15)
- Add nil check for crossDriveRunner (H17)
- Use strings.TrimSpace instead of slice on command output (H18)
- Make SaveAppConfig atomic with write-to-tmp+rename (H04)
- Pass encKey on deploy failure SaveAppConfig (H05)
- Fix IPv6 address format in TCP health probe

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

controller/internal/stacks/deploy.go

@@ -298,7 +298,7 @@ func (m *Manager) runComposeDeploy(name, stackDir string, env map[string]string,
 
 	if composeErr != nil {
 		m.logger.Printf("[ERROR] Stack %s deploy failed after %.1fs: %v", name, time.Since(start).Seconds(), composeErr)
-		// Revert in-memory state
+		// Revert in-memory and disk state
 		m.mu.Lock()
 		if s, ok := m.stacks[name]; ok {
 			s.Deployed = false
@@ -306,10 +306,12 @@ func (m *Manager) runComposeDeploy(name, stackDir string, env map[string]string,
 			s.DeployError = composeErr.Error()
 			s.AppConfig = nil
 		}
-		m.mu.Unlock()
-		// Revert disk state — keep app.yaml for debugging but mark as not deployed
+		// Also revert the shared appCfg under lock (C03 fix)
 		appCfg.Deployed = false
-		_ = SaveAppConfig(stackDir, appCfg, nil, nil)
+		m.mu.Unlock()
+		// Save reverted state to disk with encryption (H05 fix)
+		meta := LoadMetadata(stackDir)
+		_ = SaveAppConfig(stackDir, appCfg, m.encKey, SensitiveEnvVars(&meta))
 		return
 	}
 
@@ -363,8 +365,10 @@ func (m *Manager) UpdateStackConfig(name string, values map[string]string) error
 		return fmt.Errorf("saving updated config: %w", err)
 	}
 
-	_, err := m.composeExecWithEnv(stackDir, appCfg.Env, "up", "-d")
-	if err != nil {
+	// Use stackEnv which loads decrypted values for docker compose (C01 fix).
+	// appCfg.Env may contain encrypted values from LoadAppConfig.
+	env := m.stackEnv(stackDir)
+	if _, err := m.composeExecCustomEnv(stackDir, env, "up", "-d"); err != nil {
 		return fmt.Errorf("restarting with new config: %w", err)
 	}
 
@@ -552,8 +556,15 @@ func SaveAppConfig(stackDir string, cfg *AppConfig, encKey []byte, sensitiveVars
 	path := filepath.Join(stackDir, "app.yaml")
 	header := "# Auto-generated by felhom-controller — do not edit locked fields manually\n"
 	content := header + string(data)
-	if err := os.WriteFile(path, []byte(content), 0600); err != nil {
-		return fmt.Errorf("writing %s: %w", path, err)
+
+	// Atomic write: write to .tmp then rename (H04 fix)
+	tmpPath := path + ".tmp"
+	if err := os.WriteFile(tmpPath, []byte(content), 0600); err != nil {
+		return fmt.Errorf("writing %s: %w", tmpPath, err)
+	}
+	if err := os.Rename(tmpPath, path); err != nil {
+		_ = os.Remove(tmpPath)
+		return fmt.Errorf("renaming %s to %s: %w", tmpPath, path, err)
 	}
 	return nil
 }