Hacker News Reader

Comment by lima | Hacker News Reader

The whole point of Signal is not having to trust them because the protocol is open source and peer-reviewed.

Doesn't matter at all. If your adversary is the CIA, you cannot trust your mobile phone OS, firmware and drivers not to read and report your memory contents.

I used to think that as well but IOMMU is a thing now. Read strcat's posts, they're really informative. He's the developer of GrapheneOS.

https://news.ycombinator.com/threads?id=strcat

Your IOMMU needs to be programmed, just like your MMU. Programming is provided by the OS and driver. Early IOMMU configuration is provided by the firmware. Check mate.

https://news.ycombinator.com/item?id=20151431

> It can't improve the security of firmware directly, but it can certainly improve the isolation of it by auditing and improving IOMMU configuration

The IOMMU that is subservient to the cell modem, where the latter runs code of unknown provenance with the ability for ad-hoc, targeted updates over the network?

Isn't the modem one of the things isolated by the IOMMU? How is it subservient to it?

What if they all are FLOSS or isolated from the OS like on Linux phones?

If they all were properly isolated there would be less of a problem, but proper isolation is hard. If they all were FLOSS, there would also be the hope of solving this problem, as long as you can prove that you are really running the right FLOSS blobs.

But I know of no Linux smartphone that would really satisfy either of those conditions properly.

AFAIK my Librem 5 runs exclusively free software, except in (not perfectly but still) isolated WiFi and modem cards.

Yeah but have you viewed it? I haven't. Also what about iPhone blobs or similar?

I'm actually curious here as to what we can & can't prove as to being 'secure'.

There have been a number of third party audits and lots of academic research: https://community.signalusers.org/t/overview-of-third-party-...

Yeah, but the client and/or server are closed source aren't they? AFAIK Signal server source code nor servers themselves are audited.

So that means all bets are off: it's entirely possible to implement a perfect peer reviewed protocol in a client and/or server full of backdoors that entirely bypasses the protocol deliberately?

Clients and server are open source: https://github.com/signalapp

The protocol does not trust the server. Android has reproducible builds, iOS doesn't.

For certain things in signal you need to trust the server hardware and suspicious technologies like SGX. SGX is a prime target for a NSA backdoor. That some upper-layer software there is open source (with lots of caveats, e.g. you cannot verify which server version is really running and signal has a history of lagging behind on publishing their "open source" version of the server code) is a distraction at best.

Don't forget side-channels in SGX.

> For certain things in signal you need to trust the server hardware and suspicious technologies like SGX. SGX is a prime target for a NSA backdoor. That some upper-layer software there is open source (with lots of caveats, e.g. you cannot verify which server version is really running and signal has a history of lagging behind on publishing their "open source" version of the server code) is a distraction at best.

Before you worry about super expensive hard to remove hardware bugs, you should worry far more about the software running on that hardware. It is much more likely that theres bugs in the software than the hardware.

Yes, of course you are right. My comment was meant more along the lines of "even if the software were ultimately trustworthy because of open source, thousand eyes, careful development and fairy dust, those hardware problems would still be a way in".

Those "certain things" have nothing to do with message security. The SGX thing was an idea intended to provide a bit better protection against metadata leakage about who was talking to who. It turned out that there were other ways to get that sort of metadata. Does Signal even use the SGX stuff anymore?

Signal backup encryption works by trusting SGX. If you cannot trust SGX, you have to assume the attacker can decrypt your signal backups, of course including all media and messages.

>So that means all bets are off: it's entirely possible to implement a perfect peer reviewed protocol in a client and/or server full of backdoors that entirely bypasses the protocol deliberately?

No, that's not how cryptography works. Signal clients don't trust the servers, and it's entirely possible to review the client protocol and make sure that malicious server can't learn anything about the communication (except the dreaded metadata - who talks to whom). That's the point of E2E encryption.

Reply on news.ycombinator.com