[python-committers] New Authenticode certificate

[Jeff Hardy]

On Wed, Feb 10, 2016 at 12:40 AM, M.-A. Lemburg <mal at egenix.com> wrote:

> On 09.02.2016 22:40, Steve Dower wrote:
> > On 09Feb2016 1030, M.-A. Lemburg wrote:
> >> On 09.02.2016 18:41, Jeff Hardy wrote:
> >>> On Mon, Feb 8, 2016 at 12:34 PM, M.-A. Lemburg <mal at egenix.com> wrote:
> >>>
> >>>> To everyone: We now have a PSF code signing certificate.
> >>>>
> >>>> I have sent the certificate to Steve for use in the Windows
> >>>> installers. If other developers need to create signed
> >>>> installers/code for Python, please let me know.
> >>>>
> >>>
> >>> Hi Marc-Andre,
> >>> Would it be possible to use it for IronPython as well?
> >>
> >> I don't know. Steve is using it as Authenticode certificate,
> >>
> >> [SNIP]
> >>
> >> It will certainly work for signing executables and msi
> >> installers.
> >>
> >> Perhaps Steve can help with this.
>

Yes, it would be signing the IronPython .exe's, MSI, and possibly NuGet
packages (although that part of the ecosystem is in flux and I have no idea
what's going on right now).


> >>
> >
> > There are three aspects to this: technical, political and security.
> >
> > Technically, yes IronPython could absolutely be signed with the same
> certificate.
> >
> > Politically, it requires the PSF to be willing to put their name to the
> safety of the signed
> > binaries and installers. Essentially, if/when something bad is done with
> or via something signed by
> > the PSF, there is an implied responsibility (no idea how legally
> enforceable it is). I am not in a
> > position to say whether or not this is okay for IronPython.
>
> Regardless of politics (the PSF wants to help where ever we can),
> we may only sign code with the PSF code signing certificate which
> the PSF has a right to distribute.
>
> I originally was under the impression that we do, but now that I
> wanted to check, I'm having trouble finding the copyright owners
> of the code.
>
> The license is the Apache license (but without copyright holder
> information), and the stdlib is part of the installers (which the
> PSF has distribution rights to), but the IronPython runtime itself
> only says: "Copyright (c) IronPython Team", so it's not clear what
> distribution rights the PSF would have.
>

We deliberately didn't so copyright assignment at the start to avoid
dealing with the MS lawyers too much, so the bulk of the code is (c)
Microsoft, the rest would be whoever wrote it. It's a nice, low-friction
system, as long as we don't change it. :)

If we had to move to PSF copyright assignment I'd be OK with it (and I
doubt other main contributors would have an issue) but the trick would be
tracking down all other contributors and getting their sign off, and also
getting MS to sign off on it (although the MS of today would probably be
more amenable than the MS of 5 years ago).

Alternatively, maybe the *binaries* can be (c) PSF, but the code copyrights
remain the same as they are. Not sure if that's a thing. Then the PSF would
have no issues distributing the binaries. I'm pretty sure the Apache
license is enough to give the PSF (though their representative) permission
to build binaries from the source and distribute them, but IANAL, etc.


>
> > Security-wise, it is very important to minimize the number of people who
> have access to the
> > certificate. Code signed with this certificate is basically given a free
> pass by most virus scanners
> > and security software.
>
> I don't think that's a true statement. Decent virus scanners
> will still scan the files for malicious content, even if signed.
>
> It's true that minimizing the possible attack surface is always
> preferred, though.
>
> > If we decide to start signing IronPython with the PSF certificate, I'd
> be most comfortable if I were
> > doing the builds to avoid sharing the certificate any further than
> needed. But that isn't going to
> > scale when all the other interpreters want equal treatment.
> >
> > I'm not sure exactly what the cost of the certificate is to the PSF, but
> it may be an expense
> > they're willing to take to get separate certs?
>
> We can only get one code signing certificate per organization from
> our certificate provider StartSSL.
>

I don't have an issue with Steve building them; the release process is
pretty much a single make step. It's a mild annoyance for each of us, but
it would only be for final releases, so only 2-3 times a year at most.

- Jeff
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/python-committers/attachments/20160216/2280c064/attachment.html>