Hacker News Reader

Comment by lifeisstillgood | Hacker News Reader

At some point are we going to think signing each IP packet is a good idea? I struggle to see how we can ever clean the internet without something on the order of "I expect packets from this list of servers certificate" (ok I know some malware would alter that list but that's a much smaller target area to defend)

I am just wondering if this level of unstoppable infection is just going to be it, or are we at the pre-cellular structure of life point in the internet?

View on HN

There's a project called SCION [1] that (among other things) does roughly this. In essence, participants announce their presence over a multicast-type protocol, and in order to send packets to anyone, you must have received a recent announcement from them.

It's a quite fascinating re-imagination of the Internet, solving many of its problems (and probably introducing a whole slew of new ones).

[1] https://www.scion-architecture.net/

> without something on the order of "I expect packets from this list of servers certificate"

Even now as it is it's worse than it should be: I can't control which pairs of (address,certificate) will be allowed to be accepted for specific sites. Instead, every browser vendor allows any "man in the middle" with the access to any CA (and CA's are known to be very bad(1)) to insert itself between my own server and my own client.

1) Read and weep: https://arstechnica.com/information-technology/2018/03/23000...

If you want secure browser access to some resource (for values of 'secure' where it matters more than your bank account but less than situations in which you wouldn't trust _any_ browser), you really need to remove certs from any commercial CA and install only the CA you need.

I know that it is possible to somehow achieve that, the thing is, it should be possible by default, so that I can simply say to e.g. my not-too-technical friend "this is my server, this is my cert, click there in your browser to compare the cert for my site before you connect and the browser will provably also not trust anybody else but your check."

This should be a basically available scenario for the secure connection, just like what we have in SSH. Don't believe "the users are too stupid" excuse. It's just an excuse:

https://golem.ph.utexas.edu/category/2014/10/new_evidence_of...

We simply shouldn't have to have the "trust in every crooked CA" when we connect to the servers we directly know.

If we get DNSSEC (or a less complicated alternative) working in widespread use then we could simply parse CAA records in the browser.

That way a website owner can whitelist only a very specific CA for their certificates.

Ideally with DNSSEC we could also get DANE and issue our own certificates and CA's would only be necessary as cosigner for OV, EV and similar.

The problem I describe and its easy solution are completely independent of DNSSEC.

DNSSEC is required to secure DNS responses which can be crucial to prevent "every crooked CA" to issue a cert for you. If CAA was interpreted by the browser, you could determine exactly which CA is trusted to issue certs for your site and DNSSEC would ensure that the browser gets the correct list of trusted CAs.

We did at one point, IPsec was the future that never happened. Protocol RFCs used to justify lack of crypto because ipsec was going to cover it soon[1]. Then VPN gateway vendors gave IPsec a hug of death and drowned out the end-to-end IPsec vision, TLS became popular and didn't require OS support, etc. There were problems with IPsec too of course, it was too complicated, OS support was colourful, no standard APIs for apps to configure or query IPsec status etc.

https://tools.ietf.org/html/rfc5406

I don't really see how that would help. The victim's computer expected to connect to the router, because the victim's computer was intentionally downloading and running dlls from the router.

I don't think it would help with the initial compromise of the router either. Buffer overflows (for example) can be exploited over https just as well as http, and that would be similar for other cryptographic strategies.

Does QUIC do this in a way?

Reply on news.ycombinator.com