[PATCH 00/22] KEYS: Support TPM-wrapped key and crypto ops

[Denis Kenzior]

David,

On 09/20/2018 11:45 AM, David Woodhouse wrote:
> On Thu, 2018-09-20 at 09:26 +0200, Marcel Holtmann wrote:
>> Hi David,
>>
>>>>> Yes.  It shouldn't be much code, either.  You still have to check for X.509
>>>>> DER since the kernel currently supports that.
>>>>
>>>> For reasons of backward compatibility, correct?  The kernel also has
>>>> mscode.asn1 which we would need to support as well.  Since we can't break
>>>> compatibility then perhaps this doesn't buy us a whole lot in the end.
>>>
>>> Don't worry about mscode - that's not an asymmetric key parser.  That's only
>>> ever used directly from verify_pefile_signature().
>>>
>>> Currently, we have to retain support for DER-encoded X.509.
>>>
>>> But there's no reason we can't have a PEM parser that decodes the PEM and
>>> selects X.509, PKCS#8 or TPM based on the ascii header in that.  PKCS#8 and
>>> TPM don't need to take DER directly.
>>
>> since we have to support DER-encoded anyway, can we get the current
>> patches merged (with fixes to the commit messages for the openssl
>> examples if needed) and then work on PEM support inside the kernel.
>> For me these seems to be two independent features. And in the current
>> form the patches have been tested and used.
>>
>> Or let me ask this differently, are there any objections to merging
>> these patches with just DER support?
> 
> Let me rephrase that question slightly: Are we happy to have to make
> inferences from the ASN.1 structure, and in particular that a bare
> OCTET-STRING is a TPMv1 blob? I believe James ended up doing something
> somewhat more sensible for the TPMv2 blob so that might end up being
> OK...?
> 

I think it should be OK.  It is highly unlikely that we get another 
OCTET string type format, and even then we can actually peek inside and 
try to parse the TPM structure carried in the DER if there is indeed a 
conflict.

Also, after refreshing my memory of how PEM format is actually 
specified, I'm no longer convinced that having it in the kernel is such 
a good idea.  It would probably be made to work, but the PEM file format 
itself isn't that precise, at least compared to DER.

James, is there a document / official place with your TPMKey ASN.1 
format?  The only reference I can find is here:
https://kernel.googlesource.com/pub/scm/linux/kernel/git/jejb/openssl_tpm2_engine/+/93b2f4f3f0f2260292f54de4e6333219063c77b1/tpm2-asn.h

Regards,
-Denis