Re: calc_keyid_v2 producing different keyid for non-sha1 SKIDs

[Stefan Berger <stefanb@linux.ibm.com>]

On 4/26/21 6:14 PM, Vitaly Chikunov wrote:
> On Tue, Apr 27, 2021 at 01:01:48AM +0300, Vitaly Chikunov wrote:
>> Stefan,
>>
>> https://tools.ietf.org/html/rfc7093
>>
>> On Mon, Apr 26, 2021 at 04:21:26PM -0400, Stefan Berger wrote:
>>> On 4/26/21 3:37 PM, Vitaly Chikunov wrote:
>>>> Hi,
>>>>
>>>> I am reported that IMA signatures where SKID is not just sha1 of the
>>>> public key (but something different, for example different hash algo,
>>>> such as Streebog) have "wrong" keyid in the signature. This is because
>>>> a) kernel extracting keyid from the cert's subjectKeyIdentifier (SKID)
>>>> x509 extension, (or if this fails it takes just serial, perhaps, we can
>>>> disregard this corner case), it never does sha1 over the public key).
>>>
>>> Is it wrong for ecrdsa keys? What is the spec?
>> It seems, some CA provide by default certs with Streebog-256 hash as
>> drop-in replacement for SHA1, so their users forced to (re-)request the
>> certs with a compatible SHA1 SKID.
>>
>>> Here's the spec that describes using sha1 for the skid which seems to work
>>> like this for RSA and ECDSA keys from what I can tell:
>>>
>>> https://tools.ietf.org/html/rfc3280#section-4.2.1.2
>> Perhaps, you meant https://tools.ietf.org/html/rfc5280#section-4.2.1.2
>>
>>    "Other methods of generating unique numbers are also acceptable."
>>
>> Also, see https://tools.ietf.org/html/rfc7093
> And, I think all v2 signatures potentially affected.

I have been using evmctl successfully with RSA and ECDSA keys now and 
certificates created by **OpenSSL**. Problems may occur if the 
certificate-generating tool uses something else than a sha1 to calculate 
the subject key identifier (skid) and therefore the key id calculated by 
evmctl (with a sha1) does not match. For the non-working case one could 
pass in a keyidv2 that the user would have to determine from the 
certificate's subject key identifier's last 4 bytes.

It would be interesting to know which tools do not use a sha1 to 
calculate the subject key identifier or what types of keys those are so 
that one could give recommendations for tools to use. GnuTLS's certtool 
for example does not seem to use the same algorithm to calculate the 
skid, so I would not recommend using it for generating the certs to be 
used in conjunction with evmctl and IMA signatures.

Also, evmctl could for example use a different hash if for example tools 
creating certs for ecrdsa keys most often use a streebog hash. This 
would make it easier for the user not having to determine the keyid.

>
>>>> But, b) evmctl, when signing, uses just private key (not even knowing
>>>> certificate where SKID should be) and calculating sha1 of public key.
>>>> Thus, keyids could mismatch each other, and it's even not easy to fix
>>>> evmctl, because there is no cert at the time of signing.
>>>>
>>>> Perhaps, we should fix this somehow. For example, when signing,
>>>> introduce new option --cert, where SKID should be extracted. Does it
>>>> looks reasonable?
> Additionally, we could add `--keyid' option, so users could manually set
> keyid without extracting it from the cert file.


Agreed.

    Stefan