Jonathan Lindsay: Attacking the Windows Kernel

Most modern processors provide a supervisor mode that is intended to run privileged operating system services that provide resource management transparently or otherwise to non-privileged code. Although a lot of research has been conducted into exploiting bugs in user mode code for privilege escalation within the operating system defined boundaries as well as what can be done if one has arbitrary supervisor access (typically related to modern rootkit work), not a great deal of research has been done on the interface between supervisor and non-supervisor, and potential routes from one to the other.

The biggest problem arises when trying to protect the kernel from itself - for example, under the IA32 architecture implementation of Windows, the distinction between user mode and kernel mode from the user mode perspective is easily enforced through hardware based protection. However, as the kernel is running as supervisor, how does the kernel make distinctions between what it should be accessing? This would be irrelevant if the supervisor was not exposed to interaction with supervisee; but that would defeat the purpose of having a kernel.

This presentation is focussed on Windows and the Intel Architecture, and will briefly outline the current supervisor boundaries provided. Different attack vectors, along with relevant examples, will be provided to demonstrate how to attack the supervisor from the perspective of the supervised. There will then be an outline of what possible architectures could be used to mitigate such attacks, such as the research operating system Singularity.