I agree. This seems to be a hard problem, and broader than Perl's CPAN. PyPI, RubyGems, and Npm, for example, all face similar problems:

• Malicious modules made it into official PyPI repository
• Backdoored Ruby gems stole credentials, injected cryptomining code
• NPM blog on malicious modules
• Typosquatting package managers blog

• OT. Malicious software in PyPI by parv (2021) (Update)
• Malicious module on CPAN by choroba (2020) (Update)
• sometimes no Perl news is good news by zentara (2021) (Update)

• CPAN clients exposed to sig-related vulnerabilities by hippo (2021) (Update)
• Signature Verification Vulnerabilities in CPAN.pm, cpanminus and CPAN::Checksums - Hackeriet blog (Update)
• Addressing CPAN vulnerabilities related to checksums - blog by Neil Bowers (Update)
• Re^7: Meaning of XS object version (CPAN and Package Manager Security References) by me (2023) (Update)

It might be interesting to compare (and learn from) the security approaches taken by each of these similar mature open source repositories.

Update (2023)

> I'm not (yet) making heavy use of cpan or cpanm tools, and I'm still getting used to them

In case it helps, a detailed example of installing modules from CPAN securely on Linux, using cpan and cpanm, can now be found here.

Thanks to your question, I now keep a long list of Security References (don't want to disappoint the LanX ;-):

• Re: Security techniques every programmer should know (Security References)

Thanks for the idea. Shipping a root key with Perl and signing (perhaps via a few intermediaries) the developers' keys with it to let them sign their distributions would solve part of the Alice problem by providing a way for the user to verify that the package has been signed with a key known to the owner of the key (CPAN) which was signed by the root (Perl) key. I agree that having this system is better than not having a way to validate a developer's key and that this should protect from the evil mirror problem if the existing Perl installation is clean (and that latter assumption we just have to trust, see "Reflections on trusting trust" by Ken Thompson).

The current situation can be somewhat circumvented by only downloading packages from https://cpan.org/authors/id/... (effectively limiting oneself to a single mirror, which doesn't quite scale, and having to trust the HTTPS PKI, which is arguably not as good as a dedicated Perl PKI).

It should also be noted that typical supply chain attacks on modules involve typosquatting and taking over unmaintained modules, which, aside the author's public key visibly changing, wouldn't raise any louder alerts: the attacker would look the same as any other CPAN citizen, but the signed code supplied by them would also be malicious. But that's a social problem, likely impossible to solve by purely technical means.

But then, how would you know that the "shipped root key" is authentic?

Well, if you're using system perl (and you trust the operating system in the first place), than you can use the system perl provided key as a first resource. Otherwise, that depends on how you obtained perl.

Clearly, we're always reaching a point of no-trust. This is an unavoidable "chicken and egg" problem, because cryptography doesn't magically create trust, despite what many folks out there say about this. Making such assumptions might cause more trouble than not using cryptography at all.

For it to work we must know the complexities and limits involved.

Trust depends on the physical world, and cannot be virtually provided. So, however we use cryptography, it must be backed by physical bounds. That's why we have some reasonable trust on integrated circuit cards, while (hopefully) not so much on a credit card "security code", which could be somehow leaked.

That said, you should only trust a key if:

• you obtain it directly from its owner;
• you obtain it from someone you absolutely trust to have obtained it directly from its owner;
• you obtain it from someone you absolutely trust to have obtained it from someone else who you also absolutely trust to have obtained it directly from its owner;
• ... and so on (you get the picture). Needless to say, the bigger this trust chain, the more fragile it is.

If such a trust chain cannot be created, then you must be willing to accept some level of risk. Some measures can be taken to reduce that risk, but it'll exist nonethless. One such measure is to disseminate your public key over many channels, so that all of them would have to be compromised by an attacker. Again, I'll mention OpenBSD with signify on this:

... If you have either the chicken or the egg, you're all set. But what about people with neither?

There are no key servers for signify. No web of trust. Just keys. The good news is the keys are pretty small. As demonstrated. We can stick them just about everywhere, and we do. They're on the web site, they're on twitter, they're on the top side of CD. 56 base64 characters. You can read it out loud over the phone in under a minute. Wide dispersion makes it harder and harder to intercept all the ways you may get the key and increases the risk of detection should anybody try some funny business.

Hopefully, key rotation will help on that. If you can't verify the next key, then at least you know you had a compromised key. If you follow its trail (how you obtained it), you might help to strenghten the key distribution process.

return on_success() or die;