openssh unix dev - Nov 2005 - PKCS#11 support for openssh

Hello OpenSSH developers,

A week ago I've posted a patch that enables openssh to work 
with PKCS#11 tokens.

I didn't receive any comments regarding the patch or reply 
to my questions.

In current software world, providing a security product that 
does not support standard interface for external 
cryptographic hardware makes the product obsolete.

Please comment my patch, so I can know how to proceed.

Best Regards,
Alon Bar-Lev.

On Tue, 1 Nov 2005, Alon Bar-Lev wrote:
> Hello OpenSSH developers,
>
> A week ago I've posted a patch that enables openssh to work with
PKCS#11
> tokens.
>
> I didn't receive any comments regarding the patch or reply to my
questions.
>
> In current software world, providing a security product that does not
support
> standard interface for external cryptographic hardware makes the product 
> obsolete.
>
> Please comment my patch, so I can know how to proceed.
I am busy at the moment, hopefully I will have time to look at it properly 
next week. Looking at it briefly, I was concerned about the assumption of 
X.509 support - we have no intention of including x.509 at present.

-d

Dan C wrote:
> Thanks for your quick reply.
> 
> On Mon, Nov 14, 2005 at 10:11:06PM +0200, Alon Bar-Lev wrote:
> 
>>Dan C wrote:
>>
>>>On Mon, Nov 14, 2005 at 09:54:46AM +0200, Alon Bar-Lev wrote:
>>>Unfortunately I'm unable to use the OpenSC PKCS#11 provider as
desired
>>>with my card, as the manufacturer (Aladdin's eToken) does not
use a
>>>compliant layout. They have however recently developed their own
>>>proprietary PKCS#11 module, so obviously I'm keen for OpenSSH
PKCS#11
>>>support.
>>
>>But I heard of success in using eToken with OpenSC and 
>>PKCS#11 provider...
>>Maybe you want to use the same content in Windows and 
>>Linux... Then I agree that it is not possible...
> 
> 
> Yeah, that's correct.
> 
> Previously I've been using a seperate card initialized with
OpenSC's
> pkcs15-init. But with Aladdin's upcoming PKCS#11 module, it would be
> nice to consolidate all of my key usage onto one card.
> 
> 
>>>I suspect that I'm not actually using your patch as intended
though.
>>>Should I still be able to, as previously with OpenSC - generate a
self
>>>signed certifcate with my existing RSA private key, import the pair
to
>>>my card and then reference the private key to log into my legacy SSH
>>>hosts?
>>
>>Well... You need the X.509 patch for your host... I think 
>>that smartcards should be used with X.509... I have a 
>>discussion regarding this issue with OpenSSH developers...
>>
>>http://marc.theaimsgroup.com/?l=openssh-unix-dev&m=113096115818802&w=2
>>
>>Roumen Petrov does not support self-signed certificate in 
>>his X.509 patch implementation... I've asked him to... He is 
>>thinking on it....
>>
>>So if you can use a certificate which is not self-signed... 
>>It would be the best... Until things will clear up.
> 
> 
> Ah, I see - then I haven't been using it as intended. The existing
> OpenSC support allows you to reference a private RSA key and a public
> key in the form of a certificate generated against the private key (to
> humour the smartcard structure), in order to authenticate against
> standard SSH2 public-key hosts. No patching of remote hosts or amending
> authorized_keys files.
> 
> I agree with your reasoning for x509 over raw RSA support. But I think a
> replacement of the existing OpenSC support would need to still handle
> raw RSA. It would be invaluable for people with existing SSH2 PKI
> environments.
> 
> Regards,
> Dan
> 
Hello,

Attached is an update to the PKCS#11 patch. It can now be 
applied as standalone without X.509 patch, but is X.509 
patch aware.

A valid X.509 certificate must still exist on the token, but 
without X.509 support it is exported as regular RSA key.

There is a nice utility Timo Felbinger wrote 
(http://www.timof.qipc.org/x509toOpenSSH.c) that extracts 
ssh public key from X.509 certificate.

If you like X.509 support apply the X.509 patch *AFTER* the 
PKCS#11 patch. There are minor rejects that can be easily 
corrected by:
$ autoreconf -i -v

The new patch also supports self-signed certificates. If it 
finds one it treats it as RSA key and not as X.509 RSA key, 
Roumen, I think this should be the default behavior of the 
X.509 patch.

Waiting to receive many more comments...

Best Regards,
Alon Bar-Lev.

Dan C wrote:
> Alon, that's great - thank you for the update. It works perfectly in
> keeping with the old OpenSC support, but with the added flexibility of
> being able to use _any_ available PKCS#11 provider. A good improvement I
> feel.
> 
> My only remaining thoughts echo that of Andreas's, in that it would be
> useful to have "direct" ssh(1) support. For both the ease of
being able
> to choose ie. "ssh -I0 <host>" when you wish, as well as
being able to
> hardset options to use card auth for specified hosts in ssh_config(5).
> 
> Please feel free to pass my comments on to the list/Roumen/Andreas and
> by all means throw any further testing my way.
> 
> Regards,
> Dan
Hello Dan,

I am glad that all works!

I agree that there should be a simple way to use ssh with 
smartcard support... But I don't like current implementation 
in which the code is written twice, once for the agent and 
second for the ssh.

I think that ssh should always use the agent, and if not 
available execute it (Or convert the agent to a library). 
Then ssh can read the config file and add identities as if 
the agent is external. This way the private key handling 
will be implemented in one place....

When I get some kind of positive response from the openssh 
developers, I will discuses what the user interface of the 
PKCS#11 support should be and implement a more friendly 
interface.

Best Regards,
Alon Bar-Lev.