RFR CSR 8202590: Customizing the generation of a PKCS12 keystore

[Michael StJohns]

On 5/9/2018 6:06 AM, Weijun Wang wrote:
> Hi Mike
>
> Your comments make sense. However,
>
> 1. I don't think it's easy in JDK to break a PBE algorithm into PBKDF + Cipher. For example, I cannot create a SecretKey from SecretKeyFactory.getInstance("PBEWithSHA1AndDESede") and use it to init a Cipher.getInstance("DESede"). I still have to use Cipher.getInstance("PBEWithSHA1AndDESede").

No - you'd instantiate a KDF using PBE, derive the secret key from the 
password, and use it with DESede.   Right now, you're hiding a KDF under 
the cover of turning a secret key spec into a secret key.

And right now, all of this is actually hidden under the KeyStore 
covers.    If you're going to use compound algorithm names, what I'd 
rather do is

    KeyStore ks = KeyStore.getInstance("PKCS12/PBKDF2/HMAC-SHA256/AES-KW");

Or  <keystoretype>/<kdfalg>/<kdfprf and mac alg>/<dataencryption alg>

And have the generic KeyStore.getInstance("PKCS12")  return what you 
specify in the preferences, or a general parser for things you're 
reading in.


>
> 2. If I read correctly, MacData in pkcs12 (https://tools.ietf.org/html/rfc7292#page-11) always uses the old style KDF (https://tools.ietf.org/html/rfc7292#appendix-B). If we have a "pbkdf" setting, it might mislead a user to believe it's also used by MacData generation.

You are - unfortunately - correct that Appendix B is still used for the 
KDF for macdata keys.  That appears to be a bug as you shouldn't be 
using the same key (password) for different algorithms and also appears 
to be a backwards compatibility item from the previous versions of 
PKCS12.  I'll ask the CFRG  what they think about it.

But I disagree with the "mislead a user".  As you note, Appendix B is 
specific that it's the only source for macdata KDFs, and you could 
include a note in the spec that this only applies to deriving data 
encryption keys rather than the macdata key.

>
> I looked at openssl and it also has independent -certpbe and -keypbe.
>
> Also, setting multiple algorithms with a preference order might not be what we wish for. The choice of algorithms is first about security and second about interoperability. By assigning *only* one algorithm for each usage, we are clear what the default algorithms are. For example, Oracle's JDK release will set them to match our crypto roadmap. Or, an application vendor or system administrator can choose their algorithms if it can be guaranteed that all parties using the application or inside a certain enterprise support the algorithms.

You've got a fair point - but I'd disagree that assigning default 
algorithms is about security, it really is about interoperability - 
making sure that the most number of people can handle the PKCS12 object 
without installing additional providers.   But seriously,

|keystore.pkcs12.certProtectionAlgorithm = PBEWithSHA1AndRC2_40|

is not a security choice!

But, after thinking about it a bit more, the preference list isn't 
useful without additional providers - and at that point you're probably 
generating stuff by specifying algorithms and such so I agree with no 
preference list.  Could you please, please, please not use weak 
algorithms as the default here though?

>
> Thanks
> Max
>
>> On May 5, 2018, at 10:50 PM, Michael StJohns <mstjohns at comcast.net> wrote:
>>
>> On 5/5/2018 3:38 AM, Weijun Wang wrote:
>>> Please take a review of
>>>
>>>     
>>> https://bugs.openjdk.java.net/browse/JDK-8202590
>>>
>>>
>>> This enhancement has two major purposes:
>>>
>>> 1. Provide a way to change encryption and Mac algorithms used in PKCS 12.
>>>
>>> 2. The ability to create a password-less PKCS 12 keystore containing unencrypted certificates and no Mac.
>>>
>>> Especially, the long paragraph in the spec on behavior of an existing keystore makes sure that once a password-less keystore is generated (with -Dkeystore.pkcs12.certProtectionAlgorithm=NONE and -Dkeystore.pkcs12.macAlgorithm=NONE), one can add new certificates to it without any special setting and keep it password-less.
>>>
>>> Thanks
>>> Max
>>>
>>>
>> I think you want to break this into two parts - the first part specifies the algorithm used to convert a password into key material. The second defines the algorithms used for protection for the various parts.
>> # password to key material scheme
>> .pbkdf=PBKDF2withHMAC-SHA256  (Form is base function with the PRF)
>> # PKCS12 macData
>> .macAlgorithm=HMAC-SHA256  # this is the algorithm for the PKCS12 macData component, if NONE, this component is not present
>> # protection scheme for PKCS8ShroudedKeyBagn if NONE, then a PKCS8KeyBag is produced instead.
>> .keyProtectionAlgorithm=AES-KWA
>> #protection scheme for certificates - produces an encryptedData object encrypted under the scheme, or a certBag object if "NONE" is specified
>> .certProtectionAlgorithm=NONE
>>
>>
>> Second, you probably want to do this as multi-choice entries in the java.security file ala providers:
>>
>> .pbkdf.0=PBKDF2withSHA256
>> .pbkdf.9=PBKDF1withSHA1 # the current default aka pbe
>>
>> So that you can specify a somewhat secure default, but still allow for providers that don't implement the stronger versions.
>>
>> This requires a bit more work in figuring out what the embedded OIDs should be, and there is always the chance of mismatch, but it turns out there is the chance of mismatch even in the proposed version if you have protection algorithms coming from two different PBE schemes.
>>
>> Specifying it this way is closer to the PKCS5 2.0 model rather than PKCS12 and matches the recommendations in the IETF's version of PKCS12.  You also *really* don't want to use two different KDFs with the same password.
>>
>> Mike
>>
>>
>>
>>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/security-dev/attachments/20180511/8b1f1991/attachment.htm>