Risks of deterministic builds (was: Re: Truth matters when writing software and selecting leaders)

[Jacob Bachmeyer]

Martin wrote:
> On 4/4/21 11:38 PM, Jacob Bachmeyer wrote:
> > Martin wrote:
> > > In a perfect world if everything is reproducible than all the 
> > > compilations are deterministic. It means that for a given 
> > > environment your source code will always produce the same binaries. 
> > > Briefly DDC method is using mix of different environments in order 
> > > to analyze the binary patterns of the same source code.
> >
> > The downside of this is that we are right back to a binary 
> > monoculture, and an exploit that works on one machine would be 
> > trivially guaranteed to work everywhere.  We really need some kind 
> of
> > controlled randomization that allows provably equivalent executables 
> > to be produced, but such that exploits relying on hardcoded offsets 
> > will only work on a limited subset.
> I don't understand what you mean by "binary monoculture" in this 
> context can you elaborate more about it pls?

Exploits are easier to develop when hardcoded offsets, virtual 
addresses, etc. can be used.  In a "binary monoculture" environment, 
that is possible.  This contributes to and worsens security problems in 
proprietary software, which is almost always distributed as a single 
identical set of binaries.

Reproducible builds are useful for validating the compiler, but there is 
a potential downside in that they make any exploit that can be found in 
the reproducibly built program much more reliable, since everyone will 
have exactly identical binaries.  Note that this is an identical risk 
with binary distributions:  if you simply install the binaries form 
Debian, an exploit can be tuned to Debian's version of that binary and 
it will work on your machine.


-- Jacob