Skip to main content

1. Buy expired NPM maintainer email domains.
2. Re-create maintainer emails
3. Take over packages
4. Submit legitimate security patches that include package.json version bumps to malicious dependency you pushed
5. Enjoy world domination.
I just noticed "foreach" on npm is controlled by a single maintainer.

I also noticed they let their personal email domain expire, so I bought it before someone else did.

I now control "foreach" on NPM, and the 36826 projects that depend on it.
Why is it so much more often NPM we see in these disasters and so seldom Debian or even Gem or CPAN?
Because in # it takes multiple pairs of eyes to review and approve a package. Packages are signed and Debian Developers sign each other cryptographic keys after meeting in person and verifying passports. To become a Debian Developer one needs to prove commitment by contributing for years, receiving mentoring and passing interviews and exams.
It is theoretically possible to be a an active community member in the projects of your pypi/CPAN/rubygems/Debian dependencies.

Not so with the typical npm dependency tree. The culture is completely different. Your transitive dependences will typically count in the hundreds rather than a handful.

Arguably those hyper-prolific authors that maintain 390 published packages cannot even be active community members of their own packages, which could be said to be a contributing reason for that Tarr incident.
This entry was edited (3 months ago)
But are there any architectural differences? Central compilation, NMU etc?
It's entirely cultural when you're comparing pypi, rubygems and npm. For Debian as noted by @federico there are clear governance differences.
This entry was edited (3 months ago)
Tarr incident?
foreach sounds like a package that you shouldnt need with Array.prototype.forEach ​:blobfoxthonking:​
yes, that's true. It made it into ECMAScript 5.1

Now if you've for _some_ weird reason a system that requries some _older_ build target you get a polyfill.

That was provided by packages like this and should be helluvEOL nowadays. There are better suited and highly automated polyfills.

Anyway, the issue is very real. This happened before and will happen again.

It's also the very same for most language depending package managers out there and this is why version pinning is a thing.
So it could happen to PyPI (Python), RubyGems (Ruby), Crates (Rust), … too :-(
…and browser extensions and game mods. Heck, whatever allows to regain access to an account via mail basically.

No 2FA on your Google Dev account? Too bad πŸ™ƒ
Attacks on PyPI (the one I know best) and the others based on the fact that anybody can upload a new package have already happened and will keep happening.

With npn it happens more often because of the higher visibility, of the culture of tiny libraries that multiplies the attack surface by a few orders of magnitude and other social factors.
Personally I very much prefer to live in a world where there are two layers in the distribution of software (libraries).

One where everybody can upload, and I as a moderately competent software person can go to discover new things, review them (both as code and as maintainership situation), decide whether or not I want to trust them.

And another where I as a tired person who needs the software|library *now* can go and get things (and trust automatic updates) knowing that somebody has already given them at least a bit of review, that there are automatic systems in place to keep checking that it keeps working, that there are procedure in place to substitute the people involved in this review if they disappear, if there are security patches I will get them applied without having to upgrade to a completely new version at a random time and basically all the things that I would have to do personally to safely use in production code taken directly from the first layer.
The second layer is called "distros" :P
also, since with the first layer you have to re-audit with every update, you may as well vendor that dependency (as in, put a copy of a specific version in your repo), so arguably github could be enough as the first layer
Honestly having a central point for all things python which is just little more than a directory feels nicer than the alternative of having to go through github, condemning to oblivion everybody who wants to host elsewhere (including self-host).

Some kind of distributed directory would be even better, but I'm not the person who will write one any time soon :D
Also, my personal choice rather than vendoring would be to package and upload for debianΒΉ: 90+% of the work has already been done, I might as well do the last bit and make my work useful for everybody else.

ΒΉ because that's what I use and what runs on production. substitute with fedora, arch, whatever else may apply.
Each project can be installed with the required OS dependencies. In case you need to test something against an older Debian release you can just use a simple chroot or systemd-nspawn as a container. Less messy and more secure than docker.
That sounds like something I need to research more.

So far I only used chroot for repairing broken installations.
There's a series of articles starting from:
Most of the time you just need an ephemeral run akin to running chroot.
what do you plan to do with such power?
Preventing other people from using it is enough. That and using it as a chance to educate pepole on why thy can't trust NPM.
can we trust you more than we can trust npm?
I don't recommend trusting me... or any single individual, with this kind of power.

If someone asks me nicely with a rubber hose, I will be obliged to hand over access.

There is a reason the name of my company is "Distrust"

Distrust should lead to Distributed Trust.

Demand multisig code reviews, and multisig reproducibly built releases for anything that matters.

This website uses cookies to recognize revisiting and logged in users. You accept the usage of these cookies by continue browsing this website.