[PATCH v7 06/13] x86: Add early SHA support for Secure Launch early measurements

[James Bottomley]

On Sat, 2023-11-11 at 18:19 +0000, Andrew Cooper wrote:
> On 11/11/2023 5:44 pm, Eric Biggers wrote:
> > On Fri, Nov 10, 2023 at 05:27:44PM -0500, Ross Philipson wrote:
> > >  arch/x86/boot/compressed/early_sha1.c   | 12 ++++
> > >  lib/crypto/sha1.c                       | 81
> > > +++++++++++++++++++++++++
> > It's surprising to still see this new use of SHA-1 after so many
> > people objected to it in the v6 patchset.  It's also frustrating
> > that the SHA-1 support is still being obfuscated by being combined
> > in one patch with SHA-2 support, perhaps in an attempt to conflate
> > the two algorithms and avoid having to give a rationale for the
> > inclusion of SHA-1.  Finally, new functions should not be added to
> > lib/crypto/sha1.c unless those functions have multiple users.
> 
> The rational was given.  Let me reiterate it.
> 
> There are real TPMs in the world that can't use SHA-2.  The use of
> SHA-1 is necessary to support DRTM on such systems, and there are
> real users of such configurations.

Given that TPM 2.0 has been shipping in bulk since Windows 10 (2015)
and is required for Windows 11 (2021), are there really such huge
numbers of TPM 1.2 systems involved in security functions?

> DRTM with SHA-1-only is a damnsight better than no DTRM, even if SHA-
> 1 is getting a little long in the tooth.

That's not the problem.  The problem is that sha1 is seen as a
compromised algorithm by NIST which began deprecating it in 2011 and is
now requiring it to be removed from all systems supplied to the US
government by 2030

https://www.nist.gov/news-events/news/2022/12/nist-retires-sha-1-cryptographic-algorithm

That means we have to control all uses of sha1 in the kernel and have
an option to build without it.  FIPS has an even tighter timetable: it
requires sha1 to be out by 2025.

> So unless you have a credible plan to upgrade every non-SHA-2 TPM in
> the world, you are deliberately breaking part of the usecase paying
> for the effort of trying to upstream DRTM support into Linux.

Given that most CSOs follow NIST and FIPS it seems a little strange
that there would be a huge demand for such an intricate security
protocol as Dynamic Launch on a system that can't be FIPS 140-3
certified.

James