Re: [PATCH 6/8] security: keys: trusted: add PCR policy to TPM2 keys

[James Bottomley <James.Bottomley@HansenPartnership.com>]

On Mon, 2019-12-09 at 10:18 +0000, David Woodhouse wrote:
> On Sat, 2019-12-07 at 21:12 -0800, James Bottomley wrote:
> > This commit adds the ability to specify a PCR lock policy to TPM2
> > keys.  There is a complexity in that the creator of the key must
> > chose either to use a PCR lock policy or to use authentication.  At
> > the current time they can't use both due to a complexity with the
> > way authentication works when policy registers are in use.  The way
> > to construct a pcrinfo statement for a key is simply to use the
> > TPMS_PCR_SELECT structure to specify the PCRs and follow this by a
> > hash of all their values in order of ascending PCR number.
> > 
> > For simplicity, we require the policy name hash and the hash used
> > for the PCRs to be the same.  Thus to construct a policy around the
> > value of the resettable PCR 16 using the sha1 bank, first reset the
> > pcr to zero giving a hash of all zeros as:
> > 
> > 6768033e216468247bd031a0a2d9876d79818f8f
> > 
> > Then the TPMS_PCR_SELECT value for PCR 16 is
> > 
> > 03000001
> > 
> > So create a new 32 byte key with a policy policy locking the key to
> > this value of PCR 16 with a parent key of 81000001 would be:
> > 
> > keyctl new 32 keyhandle=0x81000001 hash=sha1
> > pcrinfo=030000016768033e216468247bd031a0a2d9876d79818f8f" @u
> 
> OK... but I've love to see a more formal definition of this binary
> format, as part of the "standard" we allegedly have for the overall
> ASN.1 representation.

It's actually defined in the TPM2 command manual ... it's basically the
policy commands you send to the TPM ordered so they can be directly
hashed.

However, I agree a standards definition would be good.  This format
doesn't support TPM2_PolicyOr directly (and the command manual is
silent on how it should be supported), so that's going to have to be
defined in the standard anyway.

[...]
> > +int tpm2_encode_policy(struct tpm2_policies *pols, u8 **data, u32
> > *len)
> > +{
> > +	u8 *buf = kmalloc(2 * PAGE_SIZE, GFP_KERNEL);
> > +	u8 *work = buf + PAGE_SIZE, *ptr;
> > +	int i;
> > +
> > +	if (!buf)
> > +		return -ENOMEM;
> > +
> > +	for (i = 0; i < pols->count; i++) {
> > +		u8 *seq_len, *tag_len;
> > +		u32 cmd = pols->code[i];
> > +		int l;
> > +
> > +		/*
> > +		 * cheat a bit here: we know a policy is < 128
> > bytes,
> > +		 * so the sequence and cons tags will only be two
> > +		 * bytes long
> > +		 */
> > +		*work++ = _tag(UNIV, CONS, SEQ);
> > +		seq_len = work++;
> > +		*work++ = _tagn(CONT, CONS, 0);
> > +		tag_len = work++;
> > +		asn1_encode_integer(&work, cmd);
> > +		*tag_len = work - tag_len - 1;
> > +		*work++ = _tagn(CONT, CONS, 1);
> > +		tag_len = work++;
> > +		asn1_encode_octet_string(&work, pols->policies[i],
> > +					 pols->len[i]);
> > +		*tag_len = work - tag_len - 1;
> > +		l = work - seq_len - 1;
> > +		/* our assumption about policy length failed */
> > +		if (WARN(l > 127,
> > +			 "policy is too long: %d but must be <
> > 128", l)) {
> > +			kfree(buf);
> > +			return -EINVAL;
> > +		}
> > +		*seq_len = l;
> 
> 
> 
> You're not even using your own sequence encoding here, because it
> only works when you know the length in advance. How about setting
> *seq_len to 0x80 to start with, for an indeterminate length.

I already did that in the asn.1 patch, so I've updated this one to use
it.

> Then in the happy case where it is <128, just go back and fill it in
> as you currently do. Otherwise append 0x00 0x00 as the end marker.

That doesn't work ... the format of these octet strings is likely to
have two zeros together, so they *have* to be definite length encoded.

> None of this has to be DER, does it?

None of what?  The policy?  the DER format is already in use so we
can't change it.

> <Insert more whining about PAGE_SIZE assumptions and buffer
> overflows>

OK, OK, I fixed that too.

James