Hacker News Reader

Comment by blenderob | Hacker News Reader

> TOFU schemes aren't acceptable for the public web because the average user (1) isn't equipped to compare certificate fingerprints for their bank

This! Forget about average user. As a technical user too I don't know how I would compare fingerprints every single time without making a mistake. I could install software or write my own to do this on desktop but what would I do on cell phones?

And TOFU requires "trust" on first use. How do I make sure that if I should be trusting the website public key on first use? It doesn't seem like any easier to solve than PKI.

View on HN

This is the sleight of hand being employed when folks suggest TOFU mechanisms. The problem with any communication boils down to trust. The modern web PKI has a bunch of complexity and a plenty of rough edges in how it handles resolving that trust. TOFU is then proposed as a simpler solution with none of those pesky rough edges, but it doesn't have the rough edges because it leaves all the hard parts as an exercise for the reader.

It's a bit like suggesting that AES-GCM has risks so we ought to just switch to one-time-pads.

> How do I make sure that if I should be trusting the website public key on first use? It doesn't seem like any easier to solve than PKI.

Usually such questions get replied to with a recommendation of implementing DNSSEC. Which is also obviously PKI and in many ways worse than WebPKI.

It's the usual hilarious flow of "HTTPS is dogshit, so here's the SSH fingerprint you should trust instead, served over HTTPS of course".

SSH fingerprints can also be provided via DNS with the SSHFP[0] DNS record, which coupled with DNSSEC and supposing you trust the DNS root and intermediate entities (whether that's IANA/ICANN, or alternatives like OpenNIC or Namecoin) allows you to check the SSH server fingerprints without HTTPS. At some point you probably need to trust someone anyway.

Or you can always get the fingerprint out of band. If it's some friend granting you SSH access to their server, or a vendor, or whatever, you can ask them to write the fingerprint on a piece of paper and give it to you, with you checking the paper comes from them and then checking them.

[0]: https://datatracker.ietf.org/doc/html/rfc4255

Couldn't you just use DANE/TLSA at that point?

Reply on news.ycombinator.com