From: Alex Holst <a@area51.dk>
Sent: 18 April 2002 03:49
To: dev@subversion.tigris.org
Subject:  Subversion and assurance.


Hi. I've been bribed with bananas again. This time the guilty party is
gstein who requested that I post a note with my thoughts about security
and assurance, and what steps can be taken to reduce the possible number
of security flaws in subversion 1.0.

First, a brief introduction: When people ask you, as a developer, about
security in Subversion, you might say Subversion is secure. Subversion
has access control, it supports SSL, committers need no system accounts,
and other nice things. These are _security_ features, not nessesarily
_secure_ features.

You may have access control, but what if the code implementing this
access control was written poorly, and contains a buffer overflow?  2
hours ago you worried about who could read or write to a document in
your repository. Now you discover that an attacker can execute arbitrary
code as the userid your service is running as. This is not ideal. 

Hence, we distinquish between "security features" and assurance. Brian
Snow, a technical director at the NSA, defines assurance as follows:

        "Confidence-building activities that demonstrate that a system
        possesses the desired properties and only these properties and
        that functions are implemented correctly. Assurance can be
        provided through a structured design process, documentation, and
        testing."

Assurance is what protects the user in the case of misuse or when faced
with malice. Today, cars come with safety functions such as seatbelts,
ABS breaks, airbags, etc, all of which means that you have a very good
chance of walking away from accidents. This was not so 50 years ago. I
strongly recommend listening to Brian Snow's full talk on assurance,
which is available as a RealPlayer stream from Blackhat.com:

<http://media.blackhat.com:5554/ramgen/blackhat/bh-usa-00/audio/bh-usa-00-brian-snow-audio.rm>

The two most important steps that Subversion can take are:

        Establish secure coding guidelines that are communicated to all
        developers and enforced by the project leads. 

        Improve the documentation: A diagram much like qmail's Big
        Picture which shows how code and data flows within the program.
        It allows for fast identification of security boundaries.

These steps will enable greatly improved looks into the Subversion code
for someone who has not spent the last few months getting familiar with
the Subversion code.

Additional steps include:

       Establish a QA section on the website containing documentation
       about the tests that are run against Subversion. 

       Document how new tests for both server and client can be written
       and encourage users who are in need of assurance to participate
       in the QA process. The tests against the server should
       specifically include things like attempting to break ACLs,
       attempt to issue legal commands in an inproper order, use very
       long strings for filenames and arguments, etc.

       The more you document, the more likely it is that someone with
       the knowledge to spot problems will take a look at what you have
       done.

Websites that help:

        "Secure Programming for Linux and UNIX" by David Wheeler
        http://www.dwheeler.com/secure-programs/

        Software Quality Assurance: Documentation and Review
        http://hissa.ncsl.nist.gov/publications/nistir4909/

Books that help:

        "Safer C" by Les Hatton
        "Solid Software" by Hatton, Howell & Pfleeger
        "Building Secure Software" by Viega & McGraw
        "Writing Secure Code" by Howard & LeBlanc
        "Writing Solid Software" by Maguire


I'll be delighted to answer any questions. Thanks for your time.

--
I prefer the dark of the night, after midnight and before four-thirty,
when it's more bare, more hollow.                  http://a.area51.dk/