Building a Transparent Keyserver | Hacker News Reader

Building a Transparent Keyserver

words.filippo.io

80 pointsnoident26 comments

There are a couple things missing from this:

1. The monitoring client does not ensure that the checkpoint was created recently, so a malicious log can conceal malicious entries from monitors by serving an old checkpoint.

2. Though the age keyserver policy is not configured this way, the post suggests you could create a policy that requires only a minority of witnesses (e.g. 3 of 10) to cosign a checkpoint. If you do this, then monitors have to get checkpoints that are cosigned by at least 8 of the 10 witnesses. Otherwise, a malicious log could present one view to relying parties that is cosigned by one set of witnesses, and a different view to monitors that is cosigned by a different set of witnesses. There is currently no mechanism specified for monitors to get these extra cosignatures, so if you go with a minority policy you'll need to invent your own stuff in order for witnessing to actually accomplish anything.

Fixed (1) in https://github.com/FiloSottile/torchwood/commit/8b61ef967, thank you!

I'll add a note to the part of the article that mentions non-majority policies.

I wonder if they think of a deeper integration of this into the age binary. Currently the invocation looks extremely ugly:

    age -r $(go run filippo.io/torchwood/cmd/age-keylookup@main joe@example.com)

I assume once it's stabilized you'd swap the `go run` for just installing and using a binary, similar to what you're already doing with age.

Honestly not sure why I didn't do that once the tool had stabilized.

Switched to

    go install filippo.io/torchwood/cmd/age-keylookup@main
    age -r $(age-keylookup alice@example.com)

age is designed to be composable and very stable, and this shell combination works well enough, so it's unlikely we'll build it straight into age(1).

Offtopic but I really appreciate golang and so I am always on the lookout of modern alternatives and I found age and I found it to be brilliant for what its worth

But I was discussing it with some techies once and someone mentioned to me that it had less entropy (I think they mentioned 256 bits of entropy) whereas they wanted 512 bits of entropy which pgp supported

I can be wrong about what exactly they talked about since it was long time ago so pardon me if thats the case, but are there any "issues" that you know about in age?

Another thing regarding the transparent servers is that what really happens if the servers go down, do you have any thoughts of having fediverse-alike capabilities perhaps? And also are there any issues/limitations of the transparent keyserver that you wish to discuss

Also your work on age has been phenomenal so thank you for creating a tool like age!

> But I was discussing it with some techies once and someone mentioned to me that it had less entropy (I think they mentioned 256 bits of entropy) whereas they wanted 512 bits of entropy which pgp supported

> I can be wrong about what exactly they talked about since it was long time ago so pardon me if thats the case, but are there any "issues" that you know about in age?

Entropy bikeshedding is very popular for PGP / GnuPG enthusiasts, but it's silly.

age uses X25519, HKDF-SHA256, ChaCha20, and Poly1305. Soon it will also use ML-KEM-768 (post-quantum crypto!). This is all very secure crypto. If a quantum computer turns out to be infeasible to build on Earth, I predict none of these algorithms will be broken in our lifetime.

PGP supports RSA. That's enough reason to avoid it.

https://blog.trailofbits.com/2019/07/08/fuck-rsa/

If you want more reasons:

https://www.latacora.com/blog/2019/07/16/the-pgp-problem/

> PGP supports RSA. That's enough reason to avoid it.

I hate to break the narrative but age also supports RSA, for SSH compat:

https://man.archlinux.org/man/age.1#SSH_keys

That's only because SSH supports RSA. Mainstream usage of age with age public keys only supports X25519.

Eh. You don't really get to do this sleight of hand. If you're gonna rag on RSA support as a shibboleth for bad design, it's bad for GPG and bad for age. If it's direct evidence of bad design, age shouldn't have permitted it via their SSH key support.

I agree in principle, but I'm not looking at "what SSH dragged in". I'm looking at age as a pure isolated thing, according to the spec: https://github.com/C2SP/C2SP/blob/main/age.md

This transparency keyserver actually gives us an excellent opportunity to measure how many people use Curve25519 vs RSA, even with SSH support.

We should contrast this with actively valid public keys on a PGP keyserver in 2026 and see which uses modern crypto more. The results probably won't be surprising ;)

Those goalposts are really agile.

We've moved from "PGP supports RSA. That's enough reason to avoid it." to "We should contrast this with actively valid public keys on a PGP keyserver in 2026 and see which uses modern crypto more".

We aren't having the same discussion in both places, so no, it's not a fucking goalpost.

> The author pronounces it [aɡe̞] with a hard g, like GIF, and is always spelled lowercase.

Of all the words we could've used to explain how to pronounce something

>:)

It's pronounced "aggie".

> [GIF has a hard G]

Glad I preserved a tweet that commented on a subheadline at The Verge from when the creator of the GIF died:

Subheadline from The Verge: "It's pronounced 'jif'"

Tweet: "I guess he's with jod now"

https://toedtclassnotes.site44.com/#orgdf3fc45

As a monitor, how do you differentiate between the operator removing a poisoned key versus them adding a malicious key and then trying to hide that fact?

You don’t, but remember you monitor your own keys: if you know you didn’t upload a poisoned key and the log refuses to serve a key preimage for your email, you’ve caught it misbehaving.

Dunno, IMO you need to know the bits of what operator is running to fully trust the third party, eg run in an enclave and share attestation evidence and the source code. Otherwise, operator can just mimic the appearance of the log.

No, the point of the Merkle tree inclusion proofs and of the witness cosignatures is precisely that the operator can't show a different view of the log to different parties.

Filippo Valsorda discusses his server for storing age keys

At first glance I misread this as "stone age keys" and thought it was a dig at gpg

The good old SKS network achieves most or all of the advantages of key transparency in a simpler way by being append-only. An attacker could downgrade your PGP identity on one server but the rest would have the newest version you uploaded to the network.

There was a theory floating around back in 2018 that the append-only nature of the SKS network makes it effectively illegal due to the GDPR "right to erasure" but nothing came of that and the SKS network is still alive:

* https://spider.pgpkeys.eu/

The SKS network is append-only in aspiration. There is nothing like a Merkle tree stopping a server in the pool (or a MitM) from serving a fake key to a client. The whole point of tlogs is holding systems like that accountable. Also, the section on VRFs of the article addresses precisely the user removal issue.

A single SKS server can not serve a fake key, only a valid key that existed in the past. This might be done to maliciously unrevoke a key. The normal PGP key integrity prevents straight up forgeries.

Reply on news.ycombinator.com