作者: Anil Kurmus , Robby Zippel
关键词:
摘要: Software security practitioners are often torn between choosing performance or security. In particular, OS kernels sensitive to the smallest regressions. This makes it difficult develop innovative kernel hardening mechanisms: they may inevitably incur some run-time overhead. Here, we propose building each function with and without hardening, within a single split kernel. this allows trusted processes be run under unmodified code, while system calls of untrusted directed hardened code. We show such no overhead when compared an deferring decision making use run-time. means distributors, administrators users can selectively enable according their needs: give examples cases. Although approach cannot directly applied arbitrary mechanisms, cases where can. Finally, our implementation in Linux requires few changes sources application source changes. Thus, is both maintainable easy use.