mailing list archives
Re: Installation of software, and security. . .
From: Jason Coombs <jasonc () science org>
Date: Tue, 19 Jul 2005 07:15:57 -1000
Tim Nelson wrote:
On Sun, 17 Jul 2005, John Richard Moser wrote:
Yes, you hit the nail on the head with a jackhammer. One discussion on
autopackage was that the devs don't want to limit the API and thus want
the prepare, install, and uninstall to be a bash script supplied by the
package "so it can do anything." I hate this logic. Why does it need
to be able to do "anything"?
I think you're both right :). I agree that packages need to be able
to do anything, but it'd be nice if we could try to eliminate the pre
and post install scripts.
Developers think that installers need to be able to do anything because
the developers think of themselves as being trustworthy. The code
written for an installer doesn't do anything harmful and it can be
trusted, so why should it not have the ability to do anything that the
developer decides it needs to do?
All malicious attacks originate from the hands and minds of other
people, malicious people, therefore a typical developer cannot see any
harm in their own way of thinking or in their own installer. Even those
developers who perceive an unacceptable risk or intrinsic flaw in the
way that these things get built and deployed have a very hard time
seeing themselves as responsible for the harm caused by others.
The truth is that people who expressly allow systems that are harmful to
continue to exist can be held responsible for the damage that those
systems cause, regardless of the fact that the malicious actor who
initiates the specific harm in each instance is somebody else entirely.
See: Metro-Goldwyn-Mayer Studios Inc. v. Grokster, Ltd.
Thus, if you are a developer and you deploy software without giving
serious thought to the things that you could do to make the entire
process of software distribution and installation safer for everyone,
then you are part of the problem.
Hopefully everyone can now see that applying digital signatures to code
is a pointless exercise in somebody else's arbitrary business strategy
(i.e. VeriSign and other purveyors of so-called 'federated identity
solutions') and is not being used today as a means of achieving improved
information security. A very sad state of affairs, given that signed
code at least attempts to address these issues of security during the
software installation/distribution process, albeit today's
implementations as a rule are very poorly-conceived.
We would all receive vastly-improved installation security if every
software vendor would adopt a standard for code/data/installer
authentication (that does not require digital signatures but that could
optionally use them) based on a keyed hash algorithm and a low-cost
specialized electronic device that sits on the desktop or in the server
room alongside the box to which software is deployed and is used to
verify hashes and explain forensically what the installer intends to do
to configure the box and deploy the code and data to it.
Of course that's just the ideal improvement, which I personally believe
the industry could even train end-users to understand and use.
Particularly if the proposed device were to generate an installation key
that the user would be required to enter in order to install the
software. (Sure, greedy people would try to use this to increase license
revenue or improve controls over intellectual property and copyright;
they will just have to be fought back by those who understand that the
point is security not personal enrichment.)
Short of the ideal stand-alone embedded system this concept could also
be built as software-only. Does anyone care? Will anyone ever build it?
jasonc () science org