Reflections on Trusting Trust

[Colin Percival]

Kris Kennaway wrote:
> On Tue, Nov 29, 2005 at 03:43:11PM -0800, Colin Percival wrote:
>>Even before you get to that point, you have to worry about making sure
>>that the build clients are secure.  One possibility which worries me a
>>great deal is that a trojan in the build code for a low-profile port
>>(e.g., misc/my-port-which-nobody-else-uses) could allow an attacker to
>>gain control of a build client (and then insert trojans into packages
>>which are built there).
> 
> They're closed systems that I keep up-to-date with security fixes, but
> yes, this is something that we do not defend against.  As you note,
> it's not really practical to at the moment, so the best we can do is
> just keep it in mind and look for other things to fix.

Yes and no.  Fixing other potential security risks is good, but not if
it leads users to think that the packages are more trustworthy than they
really are.  In particular, if we started distributing signed packages,
I suspect that most people would assume that the signatures guaranteed
that the packages were good, rather than simply ensuring that the packages
hadn't been modified with after they were built.

If we're going to sign anything, we need to ensure not just that we're
signing what we think we're signing, but also that we're signing what the
*end users* think that we're signing.

Colin Percival