Re: [RFC PATCH v2 1/2] tpm, tpm_tis: Introduce TPM_IOC_SET_LOCALITY

["Jarkko Sakkinen" <jarkko@kernel.org>]

On Sat Nov 2, 2024 at 12:52 PM EET, Ard Biesheuvel wrote:
> > Short answer: I have no idea. I would not mind that but neither
> > the commit message for TPM give a clue on this. Actually, I *do
> > not care* if it is RO and RW but I'm neither good at guessing
> > random shit.
> >
>
> You were cc'ed on the rest of the series, no?

Yeah, but that does not make sysfs attribute having store operation less
confusing. At minimum 2/2 should replace the current sysfs patch, if
store operation is not required.

> Shall we clarify this first, before proposing patches that introduce
> new ioctls() and kernel command line parameters to a security
> sensitive subsystem?
>
> My reading of 19/20 is that the secure launch module sets the default
> locality, and given that it can be built as a module, setting the
> default locality needs to be exported to modules (but as I indicated,
> this should probably be in a TPM internal module namespace)
>
> If setting the default locality from user space is a requirement down
> the road, we can discuss it then. For now, let's not go off into the
> weeds and derail this series even more.

If sysfs store is not required after all, and only thing that touches
the locality is slmodule, tweaking 17/20's set operation to this would
be good enough for me:

int tpm_chip_set_locality(struct tpm_chip *chip, u8 locality)
{
	int ret;

	if (locality >= TPM_MAX_LOCALITY)
		return false;

	ret = tpm_try_get_ops(chip);
	if (ret)
		return ret;

	chip->default_locality = locality;

	tpm_put_ops(chip);
	return 0;
}
EXPORT_SYMBOL_GPL(tpm_chip_set_locality);

Now that I've worked on this issue I think also 15/20 and 16/20 are
pretty clear I can suggest some tweaks to the commit messages later to
make then more self-explatonery.

BR, Jarkko