Hacker News Reader

Comment by huhtenberg | Hacker News Reader

Can official binaries be independently reproduced from published sources by members of the public?

If no, then an audit has little to no value as it still implies trusting the vendor not to fudge the binaries or, more broadly, be malicious.

View on HN

I think your judgment is too harsh.

Very few software deployment systems make it possible for binaries to be independently reproduced from published sources by the public. AFAIK, it's limited to systems like Nix, Guix, recent Debian, and other participants in the Reproducible Builds project.

However, even within those systems, if you are downloading a compiled binary instead of building it yourself, how can you be sure that you get the "right" binary every time? Does the binary download system periodically "challenge" the binary provider by building from source and comparing with the downloaded binary? If so, does it report its findings anywhere?

It seems to me that even within a software deployment system that enables users to reproduce binaries, you still end up trusting whoever runs the deployment system, because there are no methods of challenging the reproducibility in a meaningful way. The systems I mentioned above sign the binaries, which means that you implicitly trust the holder of the signing key to send you the right binary. But it doesn't mean anything about the relationship of the binary to some source code.

Having said that, if I am using some program by downloading binaries, I am trusting whoever provides the binaries. If I trust them, then a source code audit is valuable to me, even though I can't be sure the compiled binary is related to the source code.

There's NO value in 3rd party vouching for the security (read, quality) of some specific version of the software, because this opinion will be rendered null and void with the next software update.

There is some value in 3rd party verifying the system design (the architecture, the protocol, etc.) and general engineering practices in the company, but this still hinges on the need to trust this company not to be (or being coerced to be) malicious. TunnelBear hasn't established the latter, so - yes, there's little to no value in former. There is some marketing value in it though.

PS. Zimmerman's original secure VoIP project was rooted in the idea of reproducible builds. It was open source, but with a license that prohibited any use except for verifying binary builds. It was 20 (?) years ago.

Regarding Zimmerman's VoIP, Tarsnap does the same thing. The client source is available but you aren't allowed to use it for anything except building the client for the Tarsnap service.

"NO value" is a huge stretch IMO. Sure, it's entirely possible for gaping security holes to be introduced in future releases, but if past versions have been consistently vouched for as secure, that's still going to increase my confidence in future versions being secure. Or if I'm paranoid, then where possible I can just stick to a specific version which has been vouched for as secure.

The trouble with VPN providers is that even with reproducible client builds, it's much easier for them to intercept the traffic on their side. Plus there is a near-zero chance of detection, unlike on the client side where the binary can be decompiled.

I work for an ISP and believe deeply in online privacy. I've had the idea of offering up an as-private-as-I-can-make-it VPN service a few times, but I always end up at the same point: wondering how I could prove that the service wasn't doing anything malicious or nefarious -- "taps", Netflow data, etc. would all be easily available to me.

What would it take to convince you that a VPN service was trustworthy?

I feel like people who complain about this are people who wouldn't be satisfied with anything unless they rolled it themselves. Of course, you could purchase your own server, use OpenVPN, etc. But anything that you haven't touched yourself is just one more thing that's potentially malicious.

perfect is the enemy of good enough

Random audits from trusted third-parties would be a nice thing. Allow people to come in at any time and check the systems. Trust is better when distributed over multiple neutral parties.

In terms of features, allow clients to regularly change IP and don't log who is using what IP. Also mix client traffic with Tor exit nodes to add noise to the traffic.

Little to no value? There's a pretty big gap between trusting a vendor to not lie and trusting them to not make mistakes.

It's a moot point for a security vendor to establish its competence without establishing its trustworthiness.

Am I missing something? Nothing TunnelBear produces appears to be open source, so the obvious answer to your question is no. Members of the public can't build binaries at all.

> If no, then an audit has little to no value as it still implies trusting the vendor not to fudge the binaries or, more broadly, be malicious.

Given they provide a VPN service, trusting the binaries is only going to take you so far.

Reply on news.ycombinator.com