RFR 8236470: Deal with ECDSA using ecdsa-with-SHA2 plus hash algorithm as AlgorithmId

[Weijun Wang]

> On Dec 24, 2019, at 12:11 AM, Xuelei Fan <xuelei.fan at oracle.com> wrote:
> 
> Will the update for getEncodedParams() has other compatibility impact? As it is a public method, I'm not sure if the update is necessary and will impact ECDSA encoding in other places.  Please double check it.

With the current code, ECDSA will always fail when trying to parse a parameters byte array containing a hash algorithm. Therefore this fix will not bring more error.

The only parameter for a ECDSA algorithm we can handle now is ECParameters (named curve). For ecdsa-with-SHA2, the parameter already contains the hash algorithm and I have no idea where to put the curve name. My fix just throws away the whole parameter. This should not be a problem since an EC key already knows its curve, and the best ECDSA can do is to reject the signature if the curve there does not match the key's curve. 

The final copy of https://tools.ietf.org/html/rfc5758 didn't mention ecdsa-with-SHA2 at all. It was removed in an earlier draft [1] and hopefully no one should be using it.

Thanks,
Max

[1] https://tools.ietf.org/rfcdiff?difftype=--hwdiff&url2=draft-ietf-pkix-sha2-dsa-ecdsa-05.txt

> 
> Otherwise, looks good to me.
> 
> Xuelei
> 
> On 12/21/2019 5:33 AM, Weijun Wang wrote:
>> I uploaded a less disruptive webrev at https://cr.openjdk.java.net/~weijun/8236470/webrev.01.
>> --Max
>>> On Dec 21, 2019, at 7:47 PM, Weijun Wang <weijun.wang at oracle.com> wrote:
>>> 
>>> Oh, I take back the "mutates the internal" words, it didn't. algName is only a local variable.
>>> 
>>> So the major problem of the current code is that while getName() is modified to return the "full" signature algorithm name, getEncodedParams() still returns it.
>>> 
>>> I realized my fix does have the problem that AlgorithmId.parse(input).encode() will be different from input, but making change at the root is still better than recognizing specifiedWithECDSA_oid and ignoring parameters everywhere.
>>> 
>>> Thanks,
>>> Max
>>> 
>>> 
>>>> On Dec 21, 2019, at 7:23 PM, Weijun Wang <weijun.wang at oracle.com> wrote:
>>>> 
>>>> Please take a review at
>>>> 
>>>>  http://cr.openjdk.java.net/~weijun/8236470/webrev.00/
>>>> 
>>>> The current implementation is not good at dealing with ECDSA specified by ecdsa-with-SHA2 plus a hash algorithm. While the AlgorithmId::getName is able to return the "full" signature algorithm name, it mutates the internal, cannot be guaranteed to be called, and leaves the parameters unchanged. This fix move the logic to AlgorithmId::parse and the class becomes practically immutable.
>>>> 
>>>> For Oracle internal reviewers: An update is made to the test also. The closed path has been wrong since the repo consolidation. We do have such certs in the closed area.
>>>> 
>>>> Thanks,
>>>> Max
>>>> 
>>>