The Linux kernel network stack is a critical component of modern operating systems, widely deployed across platforms and often exposed to untrusted inputs. Its complex and stateful nature makes it a frequent target of security vulnerabilities, particularly those triggered by subtle protocol interactions. While existing fuzzers like syzkaller have demonstrated strong capabilities in discovering kernel bugs, they face challenges in exercising deep protocol logic due to the lack of coordinated inputs and protocol awareness. In this paper, we present TRON, a tool designed for fuzzing the Linux kernel network stack. By synthesizing syscall–packet input sequences based on protocol structure and incorporating runtime feedback, TRON enables the exploration of protocol-dependent state transitions and deep execution paths. Our approach addresses the fundamental challenges in dual-input fuzzing by integrating protocol knowledge with execution feedback. We evaluate TRON on four recent Linux kernel versions and compare it against syzkaller and kernelGPT. The results show that TRON improves branch coverage by 22.9% and 12.1% over syzkaller and kernelGPT, respectively, and discovers 25 previously unknown bugs, 7 of which have been fixed. These results demonstrate the effectiveness of protocol–system call input synthesis in enhancing network stack fuzzing and uncovering hard-to-reach bugs in kernel protocol implementations.