On Wed, 2021-03-24 at 09:14 -0700, James Bottomley wrote:
> On Tue, 2021-03-23 at 14:07 -0400, Mimi Zohar wrote:
> > On Tue, 2021-03-23 at 17:35 +0100, Ahmad Fatoum wrote:
> > > Hello Horia,
> > >
> > > On 21.03.21 21:48, Horia Geantă wrote:
> > > > On 3/16/2021 7:02 PM, Ahmad Fatoum wrote:
> > > > [...]
> > > > > +struct trusted_key_ops caam_trusted_key_ops = {
> > > > > + .migratable = 0, /* non-migratable */
> > > > > + .init = trusted_caam_init,
> > > > > + .seal = trusted_caam_seal,
> > > > > + .unseal = trusted_caam_unseal,
> > > > > + .exit = trusted_caam_exit,
> > > > > +};
> > > > caam has random number generation capabilities, so it's worth
> > > > using that
> > > > by implementing .get_random.
> > >
> > > If the CAAM HWRNG is already seeding the kernel RNG, why not use
> > > the kernel's?
> > >
> > > Makes for less code duplication IMO.
> >
> > Using kernel RNG, in general, for trusted keys has been discussed
> > before. Please refer to Dave Safford's detailed explanation for not
> > using it [1].
> >
> > [1]
> > https://lore.kernel.org/linux-integrity/BCA04D5D9A3B764C9B7405BBA4D4A3C035F2A38B@xxxxxxxxxxxxxxxxxxxxxxxx/
>
> I still don't think relying on one source of randomness to be
> cryptographically secure is a good idea. The fear of bugs in the
> kernel entropy pool is reasonable, but since it's widely used they're
> unlikely to persist very long. Studies have shown that some TPMs
> (notably the chinese manufactured ones) have suspicious failures in
> their RNGs:
>
> https://www.researchgate.net/publication/45934562_Benchmarking_the_True_Random_Number_Generator_of_TPM_Chips
>
> And most cryptograhpers recommend using a TPM for entropy mixing rather
> than directly:
>
> https://blog.cryptographyengineering.com/category/rngs/
>
> The TPMFail paper also shows that in spite of NIST certification
> things can go wrong with a TPM:
>
> https://tpm.fail/
We already had a lengthy discussion on replacing the TPM RNG with the
kernel RNG for trusted keys, when TEE was being introduced [2,3]. I'm
not interested in re-hashing that discussion here. The only
difference now is that CAAM is a new trust source. I suspect the same
concerns/issues persist, but at least in this case using the kernel RNG
would not be a regression.
[2] Pascal Van Leeuwen on mixing different sources of entropy and certification -
https://lore.kernel.org/linux-integrity/MN2PR20MB29732A856A40131A671F949FCA950@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx/
[3] Jarrko on "regression" and tpm_asym.c -
https://lore.kernel.org/linux-integrity/20191014190033.GA15552@xxxxxxxxxxxxxxx/
Mimi
