Analysing the Resolution of Security Bugs in Software Maintenance

Security bugs in software systems are often reported after incidents of malicious attacks. Developers often need to resolve these bugs quickly in order to maintain the security of such systems. Bug resolution includes two kinds of activities: triaging confirms that the bugs are indeed security problems, after which fixing involves making changes to the code. It is reported in the literature that, statistically, security bugs are reopened more often compared to others, which poses two new research questions: (a) Are developers “rushing” to triage security bugs too soon under the pressure of deadlines? (b) Do developers need to spend more time fixing security bugs to avoid frequent reopening? This thesis explores these questions in order to determine whether security bug fixing should take a higher priority than other bugs to avoid malicious attackers exploiting vulnerabilities before the problems are fixed, and whether security bug fixing should take a higher priority than other bugs. In this thesis a quantitative approach has been adopted by conducting statistical empirical studies to observe the behaviour of software developers engaged in dealing with security bugs. Firstly, the concept of "rush'' has been borrowed from the time management literature to refer to the behaviour of people delivering work under the pressure of deadlines. By observing how developers deliver bug resolution before the deadline of releases, the degree of rush has been measured as the ratio between the actual time spent by developers during triaging and the theoretical time the developers have by delaying the fixes until the next regular release. In this thesis, a suggest that delaying bug assignment helps find the right developer and gives the developer more time to prepare for the same workload with more relaxed planning constraints. Secondly, to analyse the complexity of security bug fixes, the fan-in complexity of functions relevant to security bugs has been measured, rather than simply measuring the time spent by the software developers on the fixing of such bugs. The first null hypothesis is tested using a Man-Whitney method on five software case studies, Samba, MozillaFirefox, RedHat, FreeBSD and Mozilla. The second null hypothesis is tested by comparing the results of fixing security and non-security bugs from the Samba and MozillaFirefox case studies. Statistically significant results suggest that security bugs are triaged in a rush compared to non-security bugs for RedHat, FreeBSD and Mozilla. In terms of fan-in, the results of the Samba and MozillaFirefox case studies suggest that security bugs are more complex to fix compared to non-security bugs