[Pki-devel] patches for review - ECC key archival / recovery feature implementation - Bug 745278 - [RFE] ECC encryption keys cannot be archived

Tue Mar 20 00:41:07 UTC 2012

Took a look at this patch for ECC support.

Comments and questions below.

1.

KeyRecord.java line 273

    public MetaInfo getMetaInfo() throws EBaseException {

        return mMetaInfo;

    }

Why does this throw the EBaseException?
All it does is return a value.

2.

CryptUtil.java line 186

        if (sensitive == 1 )

            keygen.sensitivePairs(true);

        else if (sensitive == 0)

            keygen.sensitivePairs(false);

        if (extractable == 1 )

            keygen.extractablePairs(true);

        else if (extractable == 0)

            keygen.extractablePairs(false);

        keygen.initialize(keysize);

The values extractable and sensitive are sent in as integer?
Should they be a boolean? I see the function calls have -1,-1
for those two final params. How does not setting sensitive and
extractable different from setting them to false?

I can see this being normal.

3.

RecoveryService.java  line 363

       if (!isRSA) {
            CMS.debug("RecoverService: recoverKey: key to recover is RSA");
       } else {

            CMS.debug("RecoverService: recoverKey: key to recover is not RSA");
       }

Would it not be simpler to print out the value of isRSA?

4.

EnrollmentService.java line 451

                metaInfo.set("EllipticCurve", oidDescription);

I think in KeyRecord.java you have a static public member variable for "EllipticCurve"
Probably should use that here for clarity.

5.

ASN1Util.c line 58

SECItem *oid;

Initialize to null?

6.

ASN1Util.c line 78

        oidTag = SECOID_FindOIDTag(oid);

        if (oidTag == SEC_OID_UNKNOWN) {

            JSS_throwMsg(env, INVALID_PARAMETER_EXCEPTION,

                "JSS getTagDescriptionByOid: OID UNKNOWN");

            goto finish;

        }

What if oidTag is null? Perhaps the call can never return this but some well known constant?

7.

ASN1Uti.java line 126

Question, it looks like the routine takes in an entire public key blob as input. 
Would there be some simpler input that could end up giving us the same answer? I do not know.

8.

Also, the routine throws a InvalidBERException exception or some such.

There are a bunch of calls to methods such as : Arrays.copyOfRange

That method appears to throw the following exceptions:

    ArrayIndexOutOfBoundsException -
    IllegalArgumentException -
    NullPointerException -

Instead of catching only an "Exception" and forcing a "InvalidBERException", would it make sense to
declare the function to also throw those exceptions?

8.

Should we check for  X509PubKeyBytes input parameter for null?

9.

File displayBySerial.template line 103

if ((result.header.keyLength == null) || (result.header.keyLength <= 0))  {

It looks like we use the above check to assume an ECC key. Is this the best way to make
this determination? Is there any info that actually tells us we have an ECC key instead of this
lack of information?

Actually it looks like this check is used everywhere in this part of the patch when a decision is needed.

----- Original Message -----
From: "Christina" <cfu at redhat.com>
To: pki-devel at redhat.com
Sent: Wednesday, March 14, 2012 2:49:37 PM
Subject: [Pki-devel] patches for review - ECC key archival / recovery feature implementation - Bug 745278 - [RFE] ECC encryption keys cannot be archived

This is the ECC phase 2 implementation (ECC key archival / recovery feature) in the JSS and DRM (KRA) 

Bug: Bug 745278 - [RFE] ECC encryption keys cannot be archived 

Please review the following patches (see "BEFORE you review" at later part of this email): 

* https://bugzilla.redhat.com/attachment.cgi?id=570109&action=diff&context=patch&collapsed=&headers=1&format=raw 
* https://bugzilla.redhat.com/attachment.cgi?id=570110&action=diff&context=patch&collapsed=&headers=1&format=raw 
* https://bugzilla.redhat.com/attachment.cgi?id=570112&action=diff&context=patch&collapsed=&headers=1&format=raw 

thanks, 
Christina 
============== 
BEFORE you review: 

For the ECC plan and design for the different phases, please refer to http://pki.fedoraproject.org/wiki/ECC_in_Dogtag 

Note: the designs beyond phase 2 were more like a brain dump. Although I said "Do Not Review," you are free to take a peak at what's intended down the road. I will go back and take a closer look and refine/adjust the designs when I begin implementation for each new phase. 

This patch contains code for the following packages: 
JSS, pki-kra, pki-common, pki-util, and pki-kra-ui 

What you need to know: 

* Problem 1 - nethsm issue: 
On the server side, if you turn on FIPS mode, in addition to nethsm, you need to attach certicom as well to have ECC SSL working on the server side. This problem has already been reported to Thales last year and they said they'd look into putting the item on their next release. Recently through a different contact, we learned there might be a way to "turn it on" (still waiting for their further instruction) 

* Problem 2- Certicom issue: 
This is a show-stopper. Initially, on the client side, I used Kai's special version of Xulrunner/Firefox, attached to Certicom token, so that the CRMF requests can be generated with key archival option. However, I encountered (or, re-encountered) an issue with certicom token. Certicom generates ECC keys with the wrong format (not PKCS7 conforming), which makes ECC key archival impossible on the server side if you use non-certicom token with DRM (but we expect an HSM in most product deployment). I have contacted Certicom for this issue, and they confirmed that they indeed have such issue. We are waiting for their fix. 

But then you might ask, "I thought I saw some ECC enrollment profiles/javascripts being checked in? How were the tests done?" The tests for those profiles were done against this ECC key archival/recovery DRM prototype I implemented last year (needs to be turned on manually in 8.1), where I "cheated" (yeah, that's why it's called a prototype) by decrypting the private key in the CRMF on DRM, and then manipulating the byte array to strip off the offending bytes before archival. 
In the real, non-prototype implementation, which is what's in this patch, for security reasons, private keys are unwrapped directly onto the token during key archival, so there is no way to manipulate the keys in memory and bypass the Certicom issue. 

A word about Kai's special version of Xulrunner/Firefox. It is not yet publicly available. 

* Problem 3- Firefox with nethsm issue: 
Another option was to connect Kai's special version firefox with an HSM to test my DRM/JSS code. However, for whatever reason, I could not get SSL going between such Firefox and ECC CA ( I did not try very hard though, as I have one other option -- writing my own ECC CRMF generation tool. I might come back to try the nethsm Firefox idea later) 

My solution (how I work on this official implementation): 
* I hacked up a ECC CRMF tool by taking the CRMFPopClient (existing in current releases), gutting out the RSA part of the code, and replacing it with ECC code. I call it CRMFPopClientEC. Two types of ECC key pairs could be generated: ECDSA or ECDH (That's another benefit of writing my own tool -- I don't know if you can select which type to generate in the Javascript... maybe you can, I just don't know). I'm in no way condoning archival of signing keys!! This is just a test tool. 
This tool takes a curve name as option (along with others), generates an ECC key pair, crafts up an CRMF request with key archival option, and sends request directly to the specified CA. You will see a "Deferred" message in the HTML response (see attachment for example) 
Once CA agent approves the request, the archival request goes to DRM and the user private key is archived. 
For recovery, DRM agent selects key recovery, etc, and you get your pkcs12. 

I did some sanity test with the pkcs12 recovered: 
* Import the recovered pkcs12 into a certicom library: 
pk12util -d . -h "Certicom FIPS Cert/Key Services" -i userEC.p12 

I also tested by retrieving a p12, importing it into a browser, and adding the user as an agent and the user could act as agent via ssl client auth to the CA. 

Finally, much of the RSA-centric code had been cleared out of the way at the time when I worked on the DRM ECC prototype, so you don't see much of that in this round. 

About SELinux: 
I have a set of rules generated on my system to run in enforcing mode. I do a writeup. 

For QE: 
How to set up the servers? The internal wiki is at https://wiki.idm.lab.bos.redhat.com/export/idmwiki/Working_with_ECC . I might have given someone a copy of how to set up ECC CA to publish on fedora.org when I worked on phase1 back a couple years ago. I will put an updated copy to cover both CA and DRM when I am checking in to dogtag. 

How do you test? Well, unless you want to use my CRMFPopClientEC tool hooked up with a nethsm (like I did), or write your own tool, you can't really test it until Certicom fixes their issue. (BTW CRMFPopClientEC can also be changed to work with ceriticom, although you would run into the same issue I mentioned above) 
It is not on my schedule to work on this tool; It is certainly not in production quality to be released as a regular tool. However, if you are interested in it, if we get enough request maybe we can think about doing something with it. 
Other test suggestion: I did not try to clone the ECC DRM. It's a good idea to test it out. 

For techpub: 
TBD 

_______________________________________________
Pki-devel mailing list
Pki-devel at redhat.com
https://www.redhat.com/mailman/listinfo/pki-devel