Container escape detection method based on heterogeneous observation chain

Abstract

Aiming at the problem of high false negative rate in container escape detection technologies, a real-time detecting method of heterogeneous observation was proposed.Firstly, the container escape behavior utilizing kernel vulnerabilities was modeled, and the critical attributes of the process were selected as observation points.A heterogeneous observation method was proposed with “privilege escalation” as the detection criterion.Secondly, the kernel module was adopted to capture the attribute information of the process in real time, and the process provenance graph was constructed.The scale of the provenance graph was reduced through container boundary identification technology.Finally, a heterogeneous observation chain was built based on the process attribute information, and the prototype system HOC-Detector was implemented.The experiments show that HOC-Detector can successfully detect all container escapes using kernel vulnerabilities in the test dataset, and the increased runtime overhead is less than 0.8%.

Keywords

• container escape;kernel vulnerability;open provenance model;heterogeneous observation chain