[PATCH] extend IMA boot_aggregate with kernel measurements

[Roberto Sassu]

> From: Mimi Zohar [mailto:zohar at linux.ibm.com]
> Sent: Tuesday, June 16, 2020 8:11 PM
> On Tue, 2020-06-16 at 17:29 +0000, Roberto Sassu wrote:
> > > From: James Bottomley [mailto:jejb at linux.ibm.com]
> > > Sent: Friday, June 12, 2020 7:14 PM
> > > On Fri, 2020-06-12 at 15:11 +0000, Roberto Sassu wrote:
> > > > with recent patches, boot_aggregate can be calculated from non-SHA1
> > > > PCR banks. I would replace with:
> > > >
> > > > Extend cumulative digest over ...
> > > >
> > > > Given that with this patch boot_aggregate is calculated differently,
> > > > shouldn't we call it boot_aggregate_v2 and enable it with a new
> > > > option?
> > >
> > > So here's the problem: if your current grub doesn't do any TPM
> > > extensions (as most don't), then the two boot aggregates are the same
> > > because PCRs 8 and 9 are zero and there's a test that doesn't add them
> > > to the aggregate if they are zero.  For these people its a nop so we
> > > shouldn't force them to choose a different version of the same thing.
> > >
> > > If, however, you're on a distribution where grub is automatically
> > > measuring the kernel and command line into PCRs 8 and 9 (I think
> Fedora
> > > 32 does this), your boot aggregate will change.  It strikes me in that
> > > case we can call this a bug fix, since the boot aggregate isn't
> > > properly binding to the previous measurements without PCRs 8 and 9.
> In
> > > this case, do we want to allow people to select an option which doesn't
> > > properly bind the IMA log to the boot measurements?  That sounds like
> a
> > > security hole to me.
> > >
> > > However, since it causes a user visible difference in the grub already
> > > measures case, do you have a current use case that would be affected?
> > > As in are lots of people already running a distro with the TPM grub
> > > updates and relying on the old boot aggregate?
> >
> > I don't know how many people would be affected. However, if an
> > attestation tool processes both measurement lists from unpatched
> kernels
> > and patched kernels, keeping the same name would be a problem as it
> > cannot be determined from the measurement list how boot_aggregate
> > was calculated.
> >
> > Anyway, I agree this should be fixed. At least, I suggest to add a Fixes tag,
> > to ensure that this patch is applied to all stable kernels.
> 
> The boot aggregate on existing systems would be sha1.  Does it make
> sense to limit this change to larger digests?  Anyone backporting
> support for larger digests would also need to backport this change as
> well.

Yes, it would be a safe choice.

Roberto

HUAWEI TECHNOLOGIES Duesseldorf GmbH, HRB 56063
Managing Director: Li Peng, Li Jian, Shi Yanli