Comment by lucb1e | Hacker News Reader

Comment by lucb1e | Hacker News Reader

> The code-sharing site Github uses encrypted-only access and, perhaps not intentionally, broke the pattern of Internet control in China.

Encrypted-only access is not a solution, it's a patch. Here is how China could break it if they wanted to:

1. Choose domains that you want to monitor in this way. E.g. github.com.

2. Requests to any IP for that domain on port 443 will return a certificate that the Chinese government issued using CNNIC. This is trusted by at least Mozilla: http://snag.gy/E1ftE.jpg (note to self: delete that ca)

3. Requests to any IP for that domain on port 80 will be converted to ssl. So the browser is connected over http to the ISP, and the ISP's servers will connect over https/ssl to Github.

In all instances, monitoring is possible. Forcing ssl is no solution. It does make monitoring harder though, and I suppose that is a good thing already, but all the Chinese government needs to do is throw money at it and the ISPs will be able to handle this.

If they do that, Mozilla, and most of the other browser vendors, will immediately revoke and blacklist that CA.

There was already a lot of debate about adding the CA, but the final consensus was, I believe, "as long as they don't actually abuse it, we should trust them; as soon as we have evidence of someone doing a MITM attack with that CA, we will blacklist it".

How many legitimate websites use that CA? Especially ones Chinese users will visit often? If the browser blocks the CA, those will all break.

Perhaps the browser could do something subtler, but it's not straightforward. Or we could make the https cert protocol more flexible, but changing protocols is hard.

Sourceforge and friends are known for injecting badware into binary installers for open-source software.

It will be interesting to see if the Chinese government makes their own alterations to the Firefox binary installers that pass through their network that add the CA back in.

Nobody checks their installers' GPG keys anyways

> Sourceforge and friends are known for injecting badware into binary installers for open-source software.

Wait... what???

GIMP no longer distributes binaries on sourceforge because sourceforge's installer "bundles third-party offers with Free Software packages. We do not want to support this kind of behavior, and have thus decided to abandon SourceForge."

http://www.gimp.org/

http://www.gluster.org/2013/08/how-far-the-once-mighty-sourc...

After reading that it doesn't look to me like they're sneaking into otherwise pure installers or doing it without project's knowledge. "When SourceForge introduced this, it bribed encouraged the top projects to participate by giving them a cut of the take. So these co-operating projects are also knowingly selling their users down the river."

For as long as I can remember Sourceforge's usability has sucked. That's reason enough to avoid it (but omg! free file hosting!) but the installer thing maybe isn't quite as deceptive and malwareish as your first message let on. It's just another facet of their lameness.

Kudos to Gimp for dumping them. Why does anyone stay with them at this point?

Aha, thanks for the clarification. It's good to know SF is only doing this under the consent of project leaders; I was under the impression that they were sneaking it under the radar.

I've actually been disappointed that downloading Firefox is only done over HTTP and I didn't see where SHASUMs for the builds were available. I actually tweeted at them about this and never got an answer: https://twitter.com/andrewdeandrade/status/40000123091669401...

You might be interested in the post we wrote about the Github MITM attack in China: https://en.greatfire.org/blog/2013/jan/china-github-and-man-...

Previously I would only have thought about distrusting a CA like CNNIC, but since the Snowden revelations, I feel like we all need to start questioning just about every CA out there.

The entire CA approach requires me to trust entities that I am not certain are truly trustworthy. After all they all fall under varied jurisdictions and at the end of they day they are owned and operated by individuals that can be bribed or coerced. Furthermore, we don't know what kind of practices each CA has in place to mitigate tampering and prevent subversion of the CA system.

With that in mind, is it possible to make a CA that is distributed (via a DHT) and publicly verifiable. The process to get a certificate from such a distributed CA could be a long, intense process, but at least it would be trustworthy to a degree no other CA could be.

Is Namecoin a viable alternative? Are there others?

I think the author wasn't suggesting to end surveillance, too, just censorship. Unless you're saying that the Chinese government could just do MITM attacks and serve the readers different stuff.

I see no reason why they couldn't, but good point. I had indeed replaced 'censorship' with 'surveillance' in my mind.

Surely, this is an (admittedly broad and difficult) issue with CAs and the trust element of SSL authentication. By the theory of SSL, an 'attacker' is not supposed to have control over a 'trusted' CA, so SSL is 'supposed to' protect against this.

I only thought of the Chinese govt issuing a certificate after writing half the post. Initially I was just going to comment on the easy possibility of MITM attacks. These would fail whenever anyone forced https into their address bar, but I think less than 5% of the people do that. That would mean just refusing connections for the <5% and still sniffing everyone else.

Strict transport security (http://en.m.wikipedia.org/wiki/HTTP_Strict_Transport_Securit...) is an attempt to prevent exactly this. However assuming that the user is always on a MITM'd connection, a preloaded list in the browser becomes necessary.

Yes, doing this attack after a site is already publicly known is hard because of HSTS (if Github uses it), but slowly caches will expire and browsers will reset themselves, and the amount of http traffic will increase a lot.

> By the theory of SSL, an 'attacker' is not supposed to have control over a 'trusted' CA, so SSL is 'supposed to' protect against this.

Every major government has a CA cert in your browser. SSL was obviously designed to be subverted in exactly this way. You won't even get a warning. Google pins their own certs in their own browser, but Moxy's Convergene.io or something like TACK would need to be implemented by Google and Mozilla for you to have a fighting chance.

Is convergence.io still a viable project? Last time I checked the github repo it looked to be abandoned.

Exactly. But the important point is that the current situation with CAs may be flawed, but since ultimately the user has control over which certificates to trust, it will be possible to use better trust models (like convergence.io) to eliminate censorship. After that, we can indeed use encryption to fight censorship.

Reply on news.ycombinator.com