USENIX security 2002 wrap up

as promised, here's a small wrap up of usenix security '02, focusing on
the Linux involvement. thanks for lwn, i hope all is well with everyone
there.

A Brief Wrapup of USENIX Security 02

Jose Nazario <jose@monkey.org>

I wanted to drop a quick note to the LWN readership about the just
completed USENIX Security 2002 conference. Held in San Francisco, where
next week LWE will be held, USENIX Security was held on the heels of
DefCon and the Black Hat Briefings, leading to a 1-2-3 conference wipeout
for some.

Linux and its security capabilities were heavily featured this year at
USENIX Security, as were plenty of other issues of interest to LWN
readers, such as wireless access point mappings and digital freedom
issues.

Linux's LSM kernel features, part of the Linux Security Module feature
kit, were presented by folks from WireX (makers of Immunix, StackGuard and
FormatGuard), NAI labs (part of the SELinux development team), and others.
Their paper gave an overview of the architecture, some example code, work
to bring other Linux security projects into the LSM architecture, and some
benchmarks. Overall an excellent report, showing how much work and
research has gone into the project.

The static analysis tool CQUAL was the subject of both an invited paper
and a WIP (work in progress report). Both groups used annotated kernel
code and the CQUAL tool to evaluate system calls and their misuse,
demonstrating the utility of this approach to finding bugs and fixing
security violations. In fact, one of the bugs found in the refereed paper
was fixed in 2.4.19.

Code analysis was again a topic of a paper, this time from Lhee and Chapin
of Syracuse University. Using a modified GCC compiler on Linux, they
showed their research which discussed using type assists to dynamically
detect buffer overflows. While they incur a noticable penalty in some
areas, the runtime performance hit is negligable, making this an avenue
worth further real world exploration.

The paper `Setuid Demystified' is worth a read by anyone using or
developing setuid applications or programming in security, which is to say
nearly everyone using UNIX. While people may think they understand the
mechanism (and implications) of setuid programs, this is a typically
poorly understood system. The call graphs for both setuid and setreuid
under Linux shows just how complex this system call is. This paper, by
Chen, Wagner, and Dean, is a tour-de-force of UNIX security, destined to
be often cited and a respected classic.

Similarily, MIT researchers Kirinsky, Bruening, and Amarsinghe discussed a
mechanism which provides secure program execution under Linux. Using an
architecture they call `shepherding', they show how the origin, target,
flow, and boundaries of programs can be monitored to provide secure
execution of applications. Their work used both Windows and Linux as a
test platform, showing the flexibility and portability of this idea.

Untrusted code was the subject of yet another great refereed paper from UC
Davis researchers Peterson, Bishop, and Pandey. Using a modified Linux
kernel, they build sanboxes for program execution and mechanisms for
programs to securely communicate between sandboxes. Their approach has a
lot of well thought out design, and their findings presented at this
year's conference show great promise.

Additional talks at the conference were also very interesting, showing a
wide range of research moving ahead. While some felt this year's
conference wasn't as diverse or energy filled as last year's, there are
plenty of worthwhile and interesting research projects bearing fruit for
practical applications. Several projects use Linux as their platform of
research and development, due in part to the availability of the source
code but also due to the wide range of software available for it. Many of
these projects are directly contributing their results back to the Linux
community.

This impact of the economy on the community was also very visible. The
vendor area of this year's conference was described as `anemic' by a
friend. Luckliy, the conference is largely academic, which suffers less
impact from the economy than do corporate conferences. Despite this,
everyone seemed to have a good time, especially the author.



___________________________
jose nazario, ph.d.			jose@monkey.org
					http://www.monkey.org/~jose/