On 4/11/2018 1:36 AM, James Bottomley wrote:
> On Tue, 2018-04-10 at 23:01 +0100, Martin Townsend wrote:
>> Using openssl to get the signature in my x509 cert
>>
>> Signature Algorithm: sha256WithRSAEncryption
>> 68:82:cc:5d:f9:ee:fb:1a:77:72:a6:a9:c6:4c:cc:d7:f6:2a:
>> 17:a5:db:bf:5a:2b:8d:39:60:dc:a0:93:39:45:0f:bc:a7:e8:
>> 7f:6c:06:84:2d:f3:c1:94:0a:60:56:1c:50:78:dc:34:d1:87:
>>
>> So there's an extra 0x00 and the signature is 257 bytes so I guess
>> this is upsetting CAAM, just need to work out where it's coming from,
>> or whether it's valid and CAAM should be handling it.
>
> A signature is just a bignum so leading zeros are permitted because
> it's the same numeric value; however, there are normalization
> procedures that require stripping the leading zeros, say before doing a
> hash or other operation which would be affected by them.
>
> CAAM should definitely handle it on the "be liberal in what you accept"
> principle. The kernel should probably remove the leading zeros on the
> "be conservative in what you do" part of the same principle.
>
Looking at the generic SW implementation (crypto/rsa.c, rsa_verify()), leading
zeros are removed:
s = mpi_read_raw_from_sgl(req->src, req->src_len);
CAAM implementation of rsa is not doing this (though it is removing leading
zeros when reading public, private keys).
It has to be fixed. Thanks for the report.
>> I notice that in my stack trace I have pkcs1pad_verify which
>> suggests some sort of padding?
>
> Yes, RSA has various forms of padding because the information being
> encrypted is usually much smaller than the encryption unit; PKCS1 is
> the most common (although its now deprecated in favour of OAEP because
> of all the padding oracle problems).
>
RSA padding has been intentionally added as a template, wrapping "textbook"
(raw) RSA primitives.
For PKCS#1 v1.5, a template instantiation is called pkcs1pad(rsa, hash_alg).
Currently in kernel the only supported RSA padding scheme is PKCS#1 v1.5.
When implemented, another scheme - for e.g. OAEP - would be added in a similar
way, as a template: oaep(rsa, ...).
Horia
