| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
For the ECC plan and the different phases, please refer to
http://pki.fedoraproject.org/wiki/ECC_in_Dogtag
Design for each phase is on the same wiki page.
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.
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 for deployment. 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 hoping they will fix it.
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 (due out in Firefox 10.0.4 on RHEL 5.8).
* 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.
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)
|
