[pycrypto] RSA exportKey question with bcrypt?

Legrandin helderijs at gmail.com
Mon Jul 15 13:52:56 PDT 2013 

Hi Kurt,

I hope I understand correctly this time.
The presence of a header like:

-----BEGIN RSA PRIVATE KEY-----
Proc-Type: 4,ENCRYPTED
DEK-Info: AES-256-CBC,16D792053CB9E5981B06E020900F86EA
[...]

indicates that the PEM envelope is encrypted and that the inner PKCS#8
structure is clear.

A header like:

-----BEGIN ENCRYPTED PRIVATE KEY-----
MIIFHzBJBgkqhkiG9w0BBQ0wPDAbBgkqhkiG9w0BBQwwDgQIHI1C+JhO35cCAgPo
[...]

indicates the opposite: the PEM envelope is clear and the inner PKCS#8
structure is encrypted. That was produced by a call like:

    >> rsa.exportKey(passphrase='boo', pkcs=8,
protection='PBKDF2WithHMAC-SHA1AndAES256-CBC')

I don't think you can have both PKCS#8-level encryption (with all its
nice properties, like ability to fine tune the algorithms and so on)
and nice human-readable headers in the PEM envelope (like DEK-Info)
describing the type of encryption that was performed.That would
totally confuse the receiver...

It' also worth streessing that PEM-level encryption is not really
specified anywhere other than in very old RFCs like RFC 1421, which
only define DES as algorithm and no password key derivation. Nowadays,
PEM-level encryption is best avoided, even if that means that the only
hint that the key is encrypted is the generic "BEGIN ENCRYPTED PRIVATE
KEY" line.

As described in the docstrings, specifying the 'protection' parameter
automatically implies
PKCS#8-level encryption, so the change of export format is actually
done on purpose.

> Also maybe more importantly would be the extra parameters for salt size and
> iteration count?

I agree it would be a nice addition (along with some support for bcrypt/scrypt).

> An afterthought, maybe it's time exportKey(), importKey() stay the same as
> 2.6 and have new functions that allow these extra combinations?

Which extra combinations? Salt size and iteration count you mean?
They could be passed as a dictionary, since they are
algorithm-specific parameters.

2013/7/15 Kurt Vogel <kvogel at mdcom.com>:
>
> Understood, but the format of the export changes when we add 'protection'
> parameter.
> Can we keep same format and have different headers, ex:
>
>> Proc-Type: 4,ENCRYPTED
>> DEK-Info: AES-256-CBC,16D792053CB9E5981B06E020900F86EA
>
> Because I notice we wrap the unencrypted PEM into a PBES2 which is encrypted
> there.
>
> Also maybe more importantly would be the extra parameters for salt size and
> iteration count?
>
> An afterthought, maybe it's time exportKey(), importKey() stay the same as
> 2.6 and have new functions that allow these extra combinations?
>
> Thanks,
> Kurt
>
>
>
> On Mon, Jul 15, 2013 at 2:26 AM, Legrandin <helderijs at gmail.com> wrote:
>>
>> Hi Kurt,
>>
>> In the PyCrypto 2.6 release, rsa.exportKey(passphrase='boo') generates
>> a TDES-encrypted private key, with encryption being done at the PEM
>> level.
>>
>> I don't think that behavior should change (e.g. we should not silently
>> switch to AES or even to the more robust PKCS#8-level encryption).
>>
>> 2013/7/15 Kurt Vogel <kvogel at mdcom.com>:
>> > And finally a comment/question/complaint :(
>> >
>> > If I use protection like this for ex:
>> >
>> > export = rsa.exportKey(passphrase='boo', pkcs=8,
>> > protection='PBKDF2WithHMAC-SHA1AndAES256-CBC')
>> >
>> > The exported key looks like this:
>> >
>> > -----BEGIN ENCRYPTED PRIVATE KEY-----
>> > MIIFHzBJBgkqhkiG9w0BBQ0wPDAbBgkqhkiG9w0BBQwwDgQIHI1C+JhO35cCAgPo
>> > MB0GCWCGSAFlAwQBKgQQ2FsezYUEaQLPHxk0z6+R4gSCBNDV++BsvKxxpo6uhUYw
>> > ...
>> >
>> > With export = rsa.exportKey(passphrase='boo'):
>> >
>> > -----BEGIN RSA PRIVATE KEY-----
>> > Proc-Type: 4,ENCRYPTED
>> > DEK-Info: DES-EDE3-CBC,CE7B6EC598ED0D10
>> >
>> > lPMvbYUypG+O4P/LilzGVQqP+6PMbnnLMP6eosyubcBqLtQxvMlvRRqgRu5CDApA
>> > ...
>> >
>> > The logic in exportKey() looks a bit convoluted, is this for some
>> > backward
>> > compatibility issue?  I would expect to see something like this, what
>> > ssh-key does:
>> >
>> > -----BEGIN RSA PRIVATE KEY-----
>> > Proc-Type: 4,ENCRYPTED
>> > DEK-Info: AES-256-CBC,16D792053CB9E5981B06E020900F86EA
>> >
>> > oL8O6n5v1S3cgGJIwrzrAq5TQIb7OeolGJpHXiyTUj1iStulgS5vAjkht0cgq53p
>> > ...
>> > ..
>> >
>> > Thanks,
>> > Kurt
>> >
>> >
>> > On Sun, Jul 14, 2013 at 11:40 PM, Kurt Vogel <kvogel at mdcom.com> wrote:
>> >>
>> >> While I'm on the subject and appears Dwayne is merging in pull requests
>> >> :)
>> >>
>> >> For RSA exportKey() think we could have **kwargs for extra prot_params
>> >> passed to
>> >>
>> >> PKCS8.wrap() like iteration_count and salt size?
>> >>
>> >>
>> >>
>> >> On Sun, Jul 14, 2013 at 9:34 PM, Kurt Vogel <kvogel at mdcom.com> wrote:
>> >>>
>> >>> Hi,
>> >>>
>> >>> Do you guys know roughly when this will go in?
>> >>>
>> >>> Also with import/export RSA keys can we support bcrypt?
>> >>>
>> >>> Does JCA and BouncyCastle use bcrypt, eg:
>> >>>
>> >>> 'BcryptWithHMAC-SHA1AndAES256-CBC'
>> >>>
>> >>> Thanks,
>> >>> Kurt
>> >>>
>> >>>
>> >>> On Fri, Jul 5, 2013 at 2:52 AM, Legrandin <helderijs at gmail.com> wrote:
>> >>> >
>> >>> > Hi Kurt , thanks a lot for providing feedback. It is much
>> >>> > appreciated.
>> >>> >
>> >>> > * I guess you refer to camel-casing used for several variables,
>> >>> > which
>> >>> > was due to my preference to stick to ASN.1 naming.
>> >>> >   I can work on that and make sure flake8 does not complain that
>> >>> > much.
>> >>> >
>> >>> > * Right. Code evolved at different points in time, and indeed it is
>> >>> > now hard to follow the path of the 'parameter' value. I will try to
>> >>> > fix that.
>> >>> >
>> >>> > * I used strings like 'PBKDF2WithHMAC-SHA1AndAES128-CBC' because
>> >>> > that
>> >>> > is the style used in JCA and BouncyCastle and a lot of people are
>> >>> > familiar with it.
>> >>> >   I am not very clear what the benefit enums might bring? One option
>> >>> > I
>> >>> > considered was the ability to provide 3 independent parameters
>> >>> >   instead of one (since protection mainly depends on type of KDF,
>> >>> > PRF,
>> >>> > and symmetric cipher) but at the end I guess most
>> >>> >   uses case are about the desire to protect the private key using a
>> >>> > password in a strong way, and the ability to tweak the various
>> >>> > parameters
>> >>> >   is not that relevant. Plus, exportKey() parameter list becomes to
>> >>> > long.
>> >>> >
>> >>> > * I am really ashamed to admit that I actually have 9 pull requests
>> >>> > open, not 2 so I am totally giving headaches to the maintainer. :-)
>> >>> >   It is of course only up to him to decide which features should go
>> >>> > in; given that he has not much time these days, it is likely that
>> >>> > only
>> >>> >   few features and bugfixes may go into any next release.
>> >>> >   The release merge window seems to roughly be once per year and I
>> >>> > find it is natural to have so many outstanding pull requests by now.
>> >>> >   To my defense, I can only say that the all pull requests cover one
>> >>> > feature only and that I try to keep them as independent as possible.
>> >>> >   Most of them apply cleanly to master (e.g. HKDF, CCM, PKCS#8, bug
>> >>> > fixes, etc).
>> >>> >   In some cases though, they do depend on an existing pull request
>> >>> > (as
>> >>> > in the case of DSA import/export depending on PKCS8 be applied
>> >>> > first),
>> >>> >   because keeping them separated is honestly too much work for me
>> >>> > *and* they are indeed extensions of other extensions.
>> >>> >
>> >>> > > Hi, I was looking at the pycrypto pull request
>> >>> > > https://github.com/dlitz/pycrypto/pull/32.  Just a few comments...
>> >>> > >
>> >>> > > * For readability can you pep8 format the code?
>> >>> > > * RSA, for import/export the protection parameter maybe rename to
>> >>> > > algo or
>> >>> > > wrap algo?  It evolves from: 'protection' to 'wrap_algo' to 'mode'
>> >>> > > as
>> >>> > > it
>> >>> > > goes down the call stack.
>> >>> > > * Also maybe make this parameter an enum/value?  Since the long
>> >>> > > string can
>> >>> > > be error prone, low level code would need to change anyway if it
>> >>> > > were
>> >>> > > either
>> >>> > > string or int if we support more modes.
>> >>> > > * And last but not least... I'm new to this email list and not
>> >>> > > sure
>> >>> > > how
>> >>> > > often pull requests are accepted but maybe you could reduce the
>> >>> > > amount of
>> >>> > > features going in?  I know you have another one, 51, after this...
>> >>> > > Maintainer may reluctant to do massive changes all at once?
>> >>> > >
>> >>> > > Anyway just ideas...
>> >>> > > Thanks for your time,
>> >>> > > Sincerely,
>> >>> > > Kurt
>> >>> > >
>> >>> > _______________________________________________
>> >>> > pycrypto mailing list
>> >>> > pycrypto at lists.dlitz.net
>> >>> > http://lists.dlitz.net/cgi-bin/mailman/listinfo/pycrypto
>> >>>
>> >>
>> >
>> >
>> > _______________________________________________
>> > pycrypto mailing list
>> > pycrypto at lists.dlitz.net
>> > http://lists.dlitz.net/cgi-bin/mailman/listinfo/pycrypto
>> >
>> _______________________________________________
>> pycrypto mailing list
>> pycrypto at lists.dlitz.net
>> http://lists.dlitz.net/cgi-bin/mailman/listinfo/pycrypto
>
>
>
> _______________________________________________
> pycrypto mailing list
> pycrypto at lists.dlitz.net
> http://lists.dlitz.net/cgi-bin/mailman/listinfo/pycrypto
>

More information about the pycrypto mailing list