[Catalog-sig] [PSF-Members] Howto Guide for MITM attacks on PyPI
Lennart Regebro
regebro at gmail.com
Mon Feb 4 11:51:11 CET 2013
I cc:d catalog-sig, aiming to move the dicussion there.
On Mon, Feb 4, 2013 at 11:40 AM, Christian Heimes <christian at python.org> wrote:
> * Package creator provides her public key somehow (a PKI is tricky and
> hard to get right)
This breaks it. It can't be "somehow".
For example, I'm currently working on a project I call "Hovercraft".
It has four dependencies: Distribute/Setuptools, docutils, lxml and
svg.path.
I'm the author of svg.path, so including the Hovercraft package
itself, that's five packages with four sources and four different
public keys. If you need to go and find these public keys "somehow"
before pip will download and install the packages, pip will become
practically useless, as you for a practical use of it have to find
hundreds of separate public keys. It will be come almost practically
impossible to download and install packages securely.
Since pip in such a situation would be useless we would have to allow
pip to install packages without checking for signatures, which then
will be how everybody will use it, making that whole security feature
unused and useless.
So that doesn't work. PyPI *has* to be made reliable in as much as we
must be able to trust PyPI to either send us the correct file, or
trust it to give us information that we can verify that it is the
correct file, automatically. If it can't be made reliable then it has
to be replaced.
//Lennart
More information about the Catalog-SIG
mailing list