Code review request, 7106773: 512 bits RSA key cannot work withSHA384 and SHA512

[Weijun Wang]

On 01/11/2012 06:55 PM, Xuelei Fan wrote:
> On 1/11/2012 6:42 PM, Weijun Wang wrote:
>>
>>
>> On 01/11/2012 06:02 PM, Xuelei Fan wrote:
>>> On 1/11/2012 5:50 PM, Weijun Wang wrote:
>>>> Hi Andrew
>>>>
>>>> Take a brief look at the webrev. Looks like this Lengthable thing is the
>>>> only change after your previous webrev. Please confirm.
>>>>
>>> Yes.
>>>
>>>> But I want something bigger. I would like to know if it is possible to
>>>> add this keysize() method deep down into the very basic Key interface.
>>>> If Key can have a method called getEncoded() I think this means it
>>>> normally has a concrete form and surely has a publicly acceptable
>>>> keysize() attribute. In JDK 8 we have default implementation for new
>>>> interface methods. Is this issue a good candidate?
>>>>
>>> As Key is an java interface, we may not be able to add one more method
>>> for compatibility reason. We may export the "Lengthable"/"Measurable"
>>> interface in JDK 8. It's possible to implement Lengthable in all
>>> sub-classes of Key in Oracle provider, but as would involve too many
>>> changes. As we need to backport this fix into JDK 7, I think we'd better
>>> consider the big picture in the future.
>>
>> Then I think the previous webrev is enough for JDK 7, and for JDK 8, we
>> simply add a new keysize() method to Key.
>>
> If we add one new method to Key interfaces. The providers based on JDK 7
> and previous releases would have to update their codes so as to
> implement this new method. As will result in serious compatibility issues.

I am talking about the new default method language feature in JDK 8 ([1]
Section 11, 12). Then the default impl of Key::keySize() returns -1,
default impl of SecretKey::keySize() returns getEncoded().length()*8, etc.

> 
> It is possible that we export the "Lengthable" interface, and have
> Oracle providers support this interface, and suggest other providers to
> use this interfaces.
> 
> The previous webrev hurt the performance a little because of reflections.

Thanks for reminding me this. Yes, those P11 and MSCAPI keys. This
webrev is still necessary, and the code changes are fine except for

1. SignatureAndHashAlgorithm.java:283, you left a System.out.println

2. KeyLength.java:58, more System.out.printlns

3. KeyLength.java:88, UnsupportedOperationException, necessary?

Thanks
Max

[1] http://cr.openjdk.java.net/~briangoetz/lambda/lambda-state-4.html

> 
> Xuelei
> 
>> Max
>>
>>>
>>>> At least, in KeyLength::getKeySize(), I would like to see "if (key
>>>> instanceof Lengthable)" to be the first check, and, if possible, the
>>>> only one needed, at least for keys from providers built in JDK.
>>>>
>>> It's OK to check it at first. But as we also need to support other
>>> providers, I think we'd better also check other types of instance.
>>>
>>> Thanks,
>>> Xuelei
>>>
>>>> Thanks
>>>> Max
>>>>
>>>>
>>>> On 01/11/2012 08:57 AM, Xuelei Fan wrote:
>>>>> "Measurable" looks like a better name. I will update the name in the
>>>>> next webrev after this round of code review:
>>>>>
>>>>> webrev: http://cr.openjdk.java.net/~xuelei/7106773/webrev.04/
>>>>>
>>>>> Thanks,
>>>>> Xuelei
>>>>>
>>>>> On 1/10/2012 11:47 PM, Vincent Ryan wrote:
>>>>>> On 01/10/12 03:19 PM, Xuelei Fan wrote:
>>>>>>> On 1/10/2012 11:09 PM, Weijun Wang wrote:
>>>>>>>> It's late night and I'll read it tomorrow. But can you choose another
>>>>>>>> word instead of Lengthable? Length is not a verb.
>>>>>>>>
>>>>>>> ;-) The name took me a lot of time, searching by google, dictionary, and
>>>>>>> any possible English translation. I have to agree that I failed to find
>>>>>>> a suitable name. I tried hardly to persuade myself that "lengthable" is
>>>>>>> also used by someother application code, so it might not too bad to use
>>>>>>> it here.
>>>>>>>
>>>>>>> With the word "lengthable", I want to express that the length is
>>>>>>> measurable. Any suggestion for the better one?
>>>>>>>
>>>>>>
>>>>>> Measurable ;-)
>>>>>>
>>>>>>
>>>>>>> Thanks,
>>>>>>> Xuelei
>>>>>>>
>>>>>>>> Max
>>>>>>>> ------------------------------------------------------------------------
>>>>>>>> 发件人: Xuelei Fan
>>>>>>>> 发送时间: 2012/1/10 22:51
>>>>>>>> 收件人: Weijun Wang
>>>>>>>> 抄送: OpenJDK
>>>>>>>> 主题: Re: Code review request, 7106773: 512 bits RSA key cannot work
>>>>>>>> withSHA384 and SHA512
>>>>>>>>
>>>>>>>> It has been around 50 days passed since the last day we talked about the
>>>>>>>> issue. Hope you can recall it from the deep memory. ;-)
>>>>>>>>
>>>>>>>> webrev: http://javaweb.us.oracle.com/~xufan/bugbios/7106773/webrev.04/
>>>>>>>>
>>>>>>>> In this update, as we agreed, a new Oracle private interface was
>>>>>>>> introduced: sun.security.util.Lengthable, and Lengthable.length() is
>>>>>>>> defined to get the length an object. sun.security.pkcs11.P11Key and
>>>>>>>> sun.security.mscapi.Key will implements the interface. As will easy and
>>>>>>>> speedup (comparing with reflection approach) the getting of key length
>>>>>>>> of those unextractable keys in hardware device.
>>>>>>>>
>>>>>>>> In the webrev, I should also include another two signed jars,
>>>>>>>> sunpkcs11.jar and sunmscapi.jar. I will include them when I get the
>>>>>>>> official signed jars.
>>>>>>>>
>>>>>>>> Thanks,
>>>>>>>> Xuelei
>>>>>>>>
>>>>>>>> On 11/22/2011 8:41 AM, Weijun Wang wrote:
>>>>>>>>> I really like this one.
>>>>>>>>>
>>>>>>>>> Thanks
>>>>>>>>> Max
>>>>>>>>>
>>>>>>>>> On 11/21/2011 08:05 PM, Xuelei Fan wrote:
>>>>>>>>>>>>     How about this approach? This looks very safe.
>>>>>>>>>>>>
>>>>>>>>>> I also prefer this approach, although it need more updates in PKCS11 and
>>>>>>>>>> MSCPI source code. If you vote for this approach, I will try to
>>>>>>>>>> implement it.
>>>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>>>
>>>>>
>>>
>