PyPI Preventing Domain Resurrection Attacks

https://blog.pypi.org/posts/2025-08-18-preventing-domain-resurrections/

by pabs3 • 11 months ago

136 52 11 months ago

blog.pypi.org

ocdtrekkie5• 11 months ago

One of the places this worried me the most was Golang, where domains for repositories are used as a package source. Credit where credit's due, when I contacted Google about a developer who had passed before their domain expired, and they were able to lock out the domain from pulling anything new into their cache.

snickerdoodle12• 11 months ago

On the other hand it's pretty crazy you could just get a domain blacklisted like that

ocdtrekkie5• 11 months ago

It was quite possible to independently prove my report was genuine. I am confident they didn't take my word for it.

I give Google a lot of well-deserved criticism, but they handled this issue pretty well and quite promptly.

sneak2• 11 months ago

go’s lockfiles include cryptographic hashes which prevent even Google from maliciously replacing packages in the cache.

ocdtrekkie5• 11 months ago

A malicious actor could, however, quite trivially add one to the version number. ;)

junon2• 11 months ago

This doesn't address availability or discoverability though. There are no upsides other than a fleeting sense of "decentralization", without any of the securities of a real decentralized protocol.

tsimionescu• 11 months ago

For a given version, yes. But the security risk here is about publishing a new (malicious) version of a previously trusted package, by taking control of an expired domain.

Say I'm using mydomain.org/abc and it's a perfectly fine package. I'm currently using v1.25.7. Two months later, I see a new version of mydomain.org/abc, v1.25.8. I will likely update to it and use it relatively confidently. But, unbeknownst to me, mydomain.org's registration expired one month ago, and a malicious actor has grabbed it in the meantime; and they have now released a malicious version of this package. Nothing in Go's security model prevents this type of attack, except that Google's package "cache" (the package repo that Go downloads things from) can be configured by Google to not provide that malicious package.

This is a major design flaw with the silly idea of using an extremely imparmenent identifier - DNS names, which can only be leased for a limited time and are automatically auctioned off after they expire - as the main identifier for something as long lived as Go packages. No other package ecosystem has this terrible design concept - even when they do use DNS, they use it only for an initial authentication, not as the only identifier.

For example, in Java's Maven repo, to publish a package org.mydomain.abc from a Maven Central account, you need to prove that you have ownership of abc.mydomain.org once when associating this package. Afterwards, the package name remains associated with your Maven Central account, forever. If you lose access to that domain name and someone else obtains, they can't go through the same steps, Maven Central won't let them, since that package name is already tied to your account.

junon2• 11 months ago

This is why URI based imports make little sense to me, honestly.

WhyNotHugo• 11 months ago

go.mod keeps checksums, so this should come up easily for downstream consumers.

However, when devs change dependencies, an innocent `go mod …` might update to a new (compromised) version of a dependency.

sneak2• 11 months ago

All of this because we still haven’t solved durable long-term strong cryptographic identity on the internet.

(Actually we have, but nobody wants to make people install Metamask or similar to log into websites.)

zokier2• 11 months ago

Keybase had good concept originally and they even had some momentum for a sec, but then they took VC money and imploded. Considering how simple system it was, I feel it should be something we could implement again (without spof)

ocdtrekkie5• 11 months ago

I really liked Keybase, and do note that Keyoxide exists as an open-source project where annotations of proofs are made directly on the PGP keys and no longer requires a central point of failure. ;)

I will say there are downsides to cryptographic keys being required for updating a package: One of the biggest pain points we have in Sandstorm is we both verify authors using their PGP key (from Keybase, as it happens, it's an old project), and we have a keypair for each app which you need to update it.

The end result unfortunately is sometimes we have trivial one-line fixes for apps which are abandoned by the author, and it is very difficult to deploy said fix to end users. This can hypothetically manifest in the form of unfixable security problems. We do have a way to forcibly change the key, but it's intentionally a pain to do, and of course, is possible because ultimately the central authority has the ability to code the platform to circumvent that key anyways.

Which is to say, you do trade some security risks for other security risks, whether it's a better situation is an exercise for the reader.

0xCMP• 11 months ago

One of the founders is doing exactly that: https://foks.pub/

Thread with comments announcing it: https://news.ycombinator.com/item?id=44520419

duskwuff• 11 months ago

but then [Keybase] took VC money and imploded

My understanding is that they were acqui-hired by Zoom - probably for optics reasons, since Zoom was having a series of high-profile security incidents around then. Keybase development basically came to a standstill at that point, probably because the team got pulled away to work on Zoom projects.

dvtkrlbs• 11 months ago

I think that is mainly because crypto or anything crypto related is tainted bu scams etc. So a regular user just doesnt have the same trust. Another problem is crypto gets attackwd way more since it also involves real money.

int0x29• 11 months ago

I think they meant cryptography not crypto currencies. I will forever hate crypto currencies for taking the entire concept of cryptography as their name and getting it associated with scams.

thewebguyd2• 11 months ago

Flawed and doesn't scale now, but wasn't that the idea behind PGP and key signing parties? Nothing enforced, but the idea was someone was supposed to verify your identity irl before they signed your key.

0cf8612b2e1e• 11 months ago

I think SigSum is trying to gain some traction in this space.

https://www.sigsum.org/

tptacek• 11 months ago

Because it's a bad idea! Identity is much more complicated than any sane cryptographic protocol can express. You can see why instantly when people suggest things like Metamask as a solution to the problem: durable long-term strong cryptographic identity with no account recovery.

account42• 11 months ago

Which is of course because we have not solved the lost key problem.

jonhohle• 11 months ago

This isn’t limited to just domains, though it’s good to consider. What happens when Google, Microsoft, or anyone else recycles unused accounts? Unfortunately for them, it seems like email addresses have been so tightly coupled with identity that it seems unreasonable that an address could be reused after any period of account inactivity.

srini_reddy• 11 months ago

Google wont allow you to register with the deleted/suspended/inactive email address

nottorp3• 11 months ago

Why is Google notifying me that they'll delete the account associated with <some email address that i do own> lately then?

They'll delete the storage, or make that email available for new registrations?

giancarlostoro2• 11 months ago

I wish Microsoft had done this to my deleted Hotmail, instead some random person got my email, and potentially access to every website I ever registered to with said email...

giancarlostoro2• 11 months ago

I was angry when I realized my hotmail account I used from 2000 all the way through the 2010s got wiped by Microsoft. I was using gmail more, but still someone else got my old email and all the emails I used to receive along with it I'm sure.

blibble• 11 months ago

great!

this was one of the major problems with doing maven style domain based namespacing

namespaces when?!

michalpleban2• 11 months ago

The presented domain expiration model looks like .com and other global domains. What about .de, .uk and countless other country domains, each having its unique expiration policy? Will it work with them, or only with .com domains?

miketheman• 11 months ago

This largely depends on the ICANN policies and their definitions of Renewal and Registration Grace Periods. The Renewal period is variable, but the Registration Grace Period is pretty much 30 days everywhere.

https://www.icann.org/en/contracted-parties/consensus-polici...

Here's an example from denic.de: https://www.denic.de/en/domains/de-domains/domain-deletion#c...

TLDRisk2• 11 months ago

The ERRP only covers gTLDs, right? Have you seen any ICANN policies requiring ccTLDs to adopt the same grace periods. As far as I know, ccTLDs can do whatever they want.

michalpleban2• 11 months ago

ICANN policies only govern global domains. Country domains set their own policies; for example, .eu expiration period is 45 days, not 30.

WHO IS policies also vary wildly, for example .de domains do not show registration date in the WHO IS, so it's not possible to know if a domain was dropped and re-registered.

TLDRisk2• 11 months ago

Domain expiration is something that's more complicated than most people realize. As an example, I'm not sure if there's a definitive source stating that auto-renew grace is part of the ERRP. In my opinion it's not and the ERRP typically won't trigger for high value domains because they'll be sold or auctioned by the registrar instead.

As far as I know, the ERRP rules describe the policies for the deletion of expired domains, which isn't required, so working off the auto-renew grace status was a good choice here. Most people don't realize that expired domains aren't guaranteed to go through the ERRP lifecycle [1].

You should be aware that during the auto-renew period, the domain name may be available to third parties for registration, depending on your registrar's terms of service. You may also run the risk of having your domain name auctioned to a third party by your registrar during this period (depending on your terms of service)

The process described here sounds like a pretty reasonable approach to a hard problem. It won't catch everything, but it's a good approach and it's nice to see some effort put into the issue.

1. https://www.icann.org/resources/pages/registrant-about-errp-...

Dunedan• 11 months ago

dupe of https://news.ycombinator.com/item?id=44942432

selinkocalar• 11 months ago

This is exactly the kind of proactive security thinking we need more of. Supply chain attacks are getting more sophisticated, and preventing domain resurrection attacks is smart defensive strategy. Too many security incidents happen because someone thought "that'll never happen to us." PyPI is thinking ahead about attack vectors before they become widespread problems.

PyPI Preventing Domain Resurrection Attacks

Related Stories