[15] RFR JDK-8246077: Cloneable test in HmacCore seems questionable
Weijun Wang
weijun.wang at oracle.com
Tue Jun 16 01:18:01 UTC 2020
Ah yes, you're correct.
But because of this delayed selection, the existence of CloneableDelegate is not that useful unless user has specified a provider at the beginning. At first every Signature is a Delegate and thus not a Cloneable, you would have to clone it to make it Cloneable.
I noticed none of our impls are Cloneable (or did I miss one?). What is the motivation to update Signature in this code change?
Thanks,
Max
> On Jun 16, 2020, at 7:29 AM, Valerie Peng <valerie.peng at oracle.com> wrote:
>
> Hmm, on a second thought, I reverted back on this last suggestion. Signature class has this delayed provider selection mechanism, so the clone() method should always rely on the chosen signatureSpi obj.
>
> Thanks,
> Valerie
> On 6/15/2020 12:59 PM, Valerie Peng wrote:
>> Sure, sounds good. Webrev is updated in place at webrev.01 since the change is just one-line.
>>
>> Will proceed with integration once the mach5 tests finish.
>>
>> Thanks!
>> Valerie
>> On 6/14/2020 2:21 AM, Weijun Wang wrote:
>>> Looks fine to me. Maybe you can also use "if (this instanceof Cloneable)" in Signature$Delegate::clone.
>>>
>>> Thanks,
>>> Max
>>>
>>>> On Jun 11, 2020, at 3:45 AM, Valerie Peng <valerie.peng at oracle.com> wrote:
>>>>
>>>> Webrev updated at: http://cr.openjdk.java.net/~valeriep/8246077/webrev.01/
>>>>
>>>> Key changes in this revision are in the Delegate.of() method in java.security.MessageDigest class.
>>>>
>>>> Comments?
>>>>
>>>> Thanks,
>>>> Valerie
>>>> On 6/8/2020 1:42 PM, Valerie Peng wrote:
>>>>> "md instanceof Cloneable && md is not from PKCS11" is not entirely precise. What I mean is that for MessageDigestSpi impls from PKCS11 provider, we will need to call the clone() to know for sure whether cloning is supported. If we decide to employ these extra logic for saving clone() call, it's better to do it inside the MessageDigest.of(...) so the callers don't have to repeat the same logic. Comments?
>>>>>
>>>>> Valerie
>>>>>
>>>>> On 6/8/2020 1:34 PM, Valerie Peng wrote:
>>>>>> Hmm, I checked all MessageDigestSpi impls of JDK providers... The story is more complicated than we expect.
>>>>>>
>>>>>> For SUN provider which implement the digest spi, implementing Cloneable means it supports clone functionality.
>>>>>>
>>>>>> However, for SunPKCS11 provider which wraps native PKCS11 library, it always implements Cloneable interface, but when clone() is called, it will then perform the equivalent PKCS11 calls and throw CNSE if any PKCS11 error is observed.
>>>>>>
>>>>>> So, there is a possibility that the instanceof check and the clone() check leads to different result in this particular scenario.
>>>>>>
>>>>>> The chance of 3rd non-PKCS11 party provider whose MessageDigest/MessageDigestSpi impl implements Cloneable but throws CNSE when clone() is called should be very low? So, I think it should be sufficient to use "md instanceof Cloneable && md is not from PKCS11"?
>>>>>>
>>>>>> Valerie
>>>>>>
>>>>>> On 6/6/2020 9:10 AM, Xuelei Fan wrote:
>>>>>>> As the the Delegate class takes care of the Cloneable SPI implementation, it should be sufficient to use "md instanceof Cloneable" only. It is not a expected behavior that a provider implements Cloneable but does not really support it.
>>>>>>>
>>>>>>> Xuelei
>>>>>>>
>>>>>>> On 6/5/2020 10:54 PM, Weijun Wang wrote:
>>>>>>>> Is it possible to try "md instanceof Cloneable || md.clone() returns"? Hopefully this is fast enough in most cases and also has the chance to recognize more actually-cloneable ones.
>>>>>>>>
>>>>>>>> I'm also OK with simply using "md instanceof Cloneable".
>>>>>>>>
>>>>>>>> --Max
>>>>>>>>
>>>>>>>>> On Jun 6, 2020, at 12:02 PM, Valerie Peng <valerie.peng at oracle.com> wrote:
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> I am merely following the spec's recommendation of trying the clone() for cloneability check.
>>>>>>>>>
>>>>>>>>> If you both are ok with it and prefer the instanceof check, I can sure reverting back the changes in HmacCore and HandshakeHash classes.
>>>>>>>>>
>>>>>>>>> Valerie
>>>>>>>>>
>>>>>>>>> On 6/5/2020 2:04 AM, Seán Coffey wrote:
>>>>>>>>>> I share the same concern. clone() is a heavy weight operation in constructors that can be called alot during intensive crypto operations.
>>>>>>>>>>
>>>>>>>>>> Now that you've done work on Delegate class - why not also keep the (instanceof Cloneable) test ? That can serve as your fastpath for the default JDK configuration AFAIK.
>>>>>>>>>>
>>>>>>>>>> regards,
>>>>>>>>>> Sean.
>>>>>>>>>>
>>>>>>>>>> On 05/06/2020 00:16, Weijun Wang wrote:
>>>>>>>>>>>> 在 2020年6月5日,03:19,Valerie Peng <valerie.peng at oracle.com> 写道:
>>>>>>>>>>>>
>>>>>>>>>>>>> Can you give an example when these 2 rules have different results? Is this only true for those implementation that directly subclass MessageDigest?
>>>>>>>>>>>> Before this fix, even the Spi impl implements Cloneable the instanceof check will always fail because the wrapper class, i.e. MessageDigest.Delegate does not. However, if you call the clone() (made public by the MessageDigest class), it will succeed because Delegate.clone() checks to see if the spi object implements the Cloneable interface, if yes, it will proceed to call the spi clone(). So, for this scenario, the results are different, e.g. instanceof returns false, but clone() succeeds. This is just one example. Is this what you are asking?
>>>>>>>>>>> No.
>>>>>>>>>>>
>>>>>>>>>>> I understand this case, but this has already been fixed. Is there any other example? Or are you only follow the words in the spec? i.e. try clone() to see if it’s cloneable.
>>>>>>>>>>>
>>>>>>>>>>> I am worried that try clone() is much heavier than just check instanof Cloneable.
>>>>>>>>>>>
>>>>>>>>>>> Thanks,
>>>>>>>>>>> Max
More information about the security-dev
mailing list