Code Review Request for 6996769: support AEAD ciphers

[Valerie (Yu-Ching) Peng]

Mike,

Thanks for the comments...
With the current API, it is still possible to have the IV generated at 
the provider level by calling Cipher.init(...) without GCM parameters.
This means that provider will generate IVs using provider-default iv 
length as well as provider-default value for tag length.
Of course, things may be a little awkward if you need to use custom tag 
length and a different iv length from what the provider uses as default.
SunJCE provider uses 128-bit tag length as well as 96-bit IVs as 
default. If there are demands to use other non-default values, then we 
should see if something can be done to facilitate these scenarios.

As for preventing the same IV from being re-used, I think it takes more 
than just adding a new constructor for GCMParameterSpec.
Generally, AlgorithmParameterSpec objects can be converted to 
AlgorithmParameters objects and vice versa. If the GCMParameterSpec only 
specifies iv length, would you expect the provider to automatically 
generate a new IV w/o explicit re-init? If yes, then what should the 
Cipher.getParameters() return, the new one or the old one? Regardless 
which way we choose, it introduces a window on when 
Cipher.getParameters() should be called.

Valerie

On 12/13/12 20:07, Michael StJohns wrote:
> Sorry for the late comment -
>
> You might want to consider section 9.1, first paragraph of SP800-38D which defines the GCM mode.  Basically, for FIPS validated implementations, to prevent accidental reuse, the IV needs to be generated inside the cryptographic boundary using one of the defined mechanism.   Obviously, that's on the ENCRYPT side since you need to match the IV on the DECRYPT side.
>
> I think you may need to add a GCMParameterSpec that allows you to specify either the IV or the IV length as a subclass of IvParameterSpec.
>
> Mike
>
>
>
>
> At 08:23 PM 12/7/2012, Valerie (Yu-Ching) Peng wrote:
>> Max,
>>
>> The webrev has been updated so that different key + iv values have to be used for AES/GCM encryption.
>> Latest version at: http://cr.openjdk.java.net/~valeriep/6996769/webrev.03/
>>
>> Please review and send me comments.
>> Thanks!
>> Valerie
>>
>> On 11/07/12 21:50, Valerie (Yu-Ching) Peng wrote:
>>> Max,
>>>
>>> Update: I removed the block (starting line 580 in CipherCore.java) for handling RC2 since it's never used.
>>>
>>> It turns out that the current impl in RC2Cipher always convert the AlgorithmParameters object to RC2ParameterSpec and only uses CipherCore.init(..., AlgorithmParameterSpec,...) method. Thus, I won't be adding a regression test, but rather simply document the current RC2Cipher behavior in CipherCore.java to clarify things up.
>>>
>>> The updated webrev is at:
>>> http://cr.openjdk.java.net/~valeriep/6996769/webrev.01/
>>>
>>> Xuelei brought up the issue of enforcing (Key+IV) uniqueness for GCM mode during this afternoon's meeting.
>>> I think more changes may be made after we decide what to do.
>>> So, there may be a webrev.02 coming... Just a heads up.
>>>
>>> Thanks!
>>> Valerie
>>>
>>> On 11/07/12 14:48, Valerie (Yu-Ching) Peng wrote:
>>>>> 580                  } else if (key.getAlgorithm().equals("RC2")) {
>>>>>
>>>>> This seems a bug fix. Is there a regression test for it?
>>>> I just noticed this problem when make the enhancement for GCM mode.
>>>> I will add a regression test for this.
>