This article develops a methodology on observed-mode-dependent nonfragile control for networked control systems subject to denial-of-service (DoS) attacks, which takes the strong concealment of attack behavior into account. To describe complex switching phenomenon between different forms of attacks that the adversary may launch on communication channels, the semi-Markov chain (SMC) is introduced. DoS attacks with a stealth characteristic are analyzed from a hidden semi-Markov jump systems perspective by constructing a double-layer stochastic process consisting of an SMC and an observed mode sequence. With the aid of emission probability, an observed-mode-dependent nonfragile controller is designed, which can still stabilize the system even if actual attack modes are unavailable. By utilizing the Lyapunov function that depends on actual attack modes and observed ones, sufficient conditions to ensure the $\delta$-error mean square stability of the closed-loop system is deduced. Eventually, a simulation related to the mass-spring-damper system with a single channel and an example of the unmanned ground vehicle with two channels are studied, both of which demonstrate the feasibility and validity.