Hacker News Reader

Comment by krupan | Hacker News Reader

I had a co-worker who was pushing for certs that were only valid for a day, or for hours or even minutes. It would solve the whole problem of revocation.

View on HN

When I've implemented JWT in the past (yes, I know, "don't use JWT"), I've opted for public/private signatures. The public cert I would publish on some URL for all consumers of the JWT to use. The rotation of certs there was a matter of hours. It would hold 3-4 of the last certs (depending on JWT lifetime).

What's wrong with using JWTs? What you have done should be industry standard and is enabled by the JWK spec. In fact several features of OpenID Connect are built upon signed JWTs from a well-known JWK store.

https://www.howmanydayssinceajwtalgnonevuln.com/

Having such a tempting footgun is a real problem.

The problem I think is that JWT is just a small puzzle piece of a great artwork that is identity and auth. It is overly complex (not without reason) and people don't put in enough effort to learn the whole picture (and it just isn't thought in schools) and just want to use something quickly for their current problem. Which is why they pick JWT and don't understand the host of related technologies - OAuth2, OIDC, JWK, JWE - just to name a few.

I don't see a reason why I would ever want to use auth=none for anything as at that point JWT is just base64 encoded json but I may be too deep in the identity world by now.

You have to draw the line somewhere. Daily certs would probably increase the load for CAs by a lot. Also, energy consumption for certificate management would increase.

How compute intensive is a CA, anyway? I’d think that a few Raspberry Pi-ish devices would be enough to do it for an entire enterprise (or maybe even one, but fault tolerance).

In general I would assume when you shorten certificate validity from 90-ish days to one the computational resources required for issuance will increase to about 90x of what it was before. So it would be a lot more expensive to run.

Also, unrelated to that argument but I think the most expensive part in a CA is not issuance but rather OCSP since an OCSP query could happen with every TLS connection (although stapling exists to solve that). OCSP would not be affected by a shortening of validity periods.

Is he not aware of OCSP Stapling? It's very much a solved issue, as the same CA authorization can be presented to multiple clients.

Openstack Anchor aimed for that. No problem with revocation if your certificates last a day, are fully automated, and issuing is validated by network https://wiki.openstack.org/wiki/Security/Projects/Anchor

So like Kerberos?

Reply on news.ycombinator.com