[keycloak-dev] PAM integration with FreeIPA

[John Dennis]

On 06/24/2016 09:07 AM, Bruno Oliveira wrote:
> On 2016-06-23, Bill Burke wrote:
>>
>>
>> On 6/23/16 2:56 PM, Bruno Oliveira wrote:
>>> On 2016-06-23, Bill Burke wrote:
>>>>
>>>>
>>>> On 6/23/16 12:25 PM, John Dennis wrote:
>>>>> On 06/23/2016 10:00 AM, Bruno Oliveira wrote:
>>>>>> Good morning,
>>>>>>
>>>>>> One of the use case scenarios described for FreeIPA, is the integration via PAM
>>>>>> and SSSD, which "automagically" handles the authentication against the IdM.
>>>>>>
>>>>>> This first step requires pretty much an IPA setup, but
>>>>>> works with libpam4j[1]. Now, thinking about Keycloak, I
>>>>>> would like to have an Authenticator for PAM[2], which is pretty much our
>>>>>> UsernamePasswordForm + PAM. Does it make sense?
>>>>>>
>>>>>> Current flow:
>>>>>>
>>>>>> * User logs into Web application with username/password
>>>>>> * PAM authenticator collects data and authenticate against PAM
>>>>>> * SSSD authenticates against IdM
>>>>>> * Authentication is complete
>>>>>>
>>>>>> After the last step, should we propagate that user to our database?
>>>>>> Maybe, like Marek already mentioned, have a SSSDFederationProvider?
>>>>>>
>>>>>> [1] -
>>>>>> http://search.maven.org/#artifactdetails%7Corg.abstractj%7Clibpam4j%7C1.9.0%7Cjar
>>>>>> [2] - https://keycloak.gitbooks.io/server-developer-guide/content/topics/auth-spi.html
>>>>>
>>>>> Simo brought up a concern after forwarding this to our internal identity
>>>>> team list. His comment is:
>>>>>
>>>>>  >
>>>>>  > Current flow:
>>>>>  >
>>>>>  > * User logs into Web application with username/password
>>>>>  > * PAM authenticator collects data and authenticate against PAM
>>>>>
>>>>> I am worried about how these 2 steps are expressed, it seem to imply PAM
>>>>> is used only as a username/password verifier.
>>>>> There is no mention/awarness of PAM conversations where we can prompt
>>>>> for things like second factors or password changes.
>>>>>
>>>>
>>>> Ok, I've spent maybe 20 seconds googling into what PAM conversations are
>>>> "PAM example conversation code".   You'll have to explain to me why PAM
>>>> conversations have any relevance to web login.  Just looking at this
>>>> example:
>>>>
>>>> https://www.freebsd.org/doc/en_US.ISO8859-1/articles/pam/pam-sample-conv.html
>>>>
>>>> It looks as if PAM conversations are targeted to simple text logins
>>>> (i.e. SSH, telnet, etc.).  Pushing and pulling text to and from stdin
>>>> and stdout.  What does that have to do with web login?
>>>
>>> Your question is totally fair. And the reason why we have to integrate
>>> with PAM is pretty much because there's no DBus interface for SSSD
>>> to provide username/password. Otherwise we would just communicate
>>> directly with DBus and call it a day.
>>>
>>
>> This is solely to allow keycloak to update passwords?  Not really
>> understanding here.
>
> Not really Bill, to give you more context. Login through PAM is just one
> of the scenarios described by Dmitri at slide #19[1].
>
> * User starts browser and connects to a resource
> * Resource redirects to Keycloak
> * User is presented with a login form
> * User fills username and password
> * User data is collected and passed to SSSD over D-Bus
>
> Here, we can't provide username/password to SSSD, because we don't have
> a DBus interface for it. So instead, we make use of PAM to make it happen.
>
> * SSSD authenticates against AD
> * Authentication complete (against FreeIPA)
>
> This is where I need some help to define what would be the best next
> step for us.
>
> * Assertion/token is issued
> * User is redirected to the resource
>
> In this scenario nothing is stored/updated on Keycloak.
>
>>
>>> The goal is pretty much to be used for Basic Authentication.
>>>
>>>>
>>>> As for PAM itself, it looks like it is a library.  (again a 20 second
>>>
>>> It's pretty much a low level authentication module to support multiple
>>> schemes like: login, ftp, ssh, telnet...(you certainly found it already)
>>>
>>>> Google search).  What I don't know is where PAM ends and SSSD takes
>>>> over.  So its hard to give you advice.
>>>
>>> This is how it happens from my understanding:
>>>
>>> 1. We start the PAM conversation from our client application (a IPA client machine),
>>> pam_sss is contacted (SSSD)
>>> 2. SSSD's PAM responder receives the authentication request and forwards
>>> it to FreeIPA server
>>> 3. FreeIPA server process the request and returns the result back to PAM
>>> responder.
>>>
>>> The data flow is better described here (https://fedorahosted.org/sssd/wiki/InternalsDocs#a3.2.2.2.DataFlowPAMResponder).
>>>
>>
>> It looks like a conversation requires some sort of session object or session
>> connection.  Remember, a web login can span multiple requests and could
>> possibly be serviced on different machines.  Sounds like any integration
>> with PAM is going to be quite limited.  Maybe that's what you are getting
>> at?
>
> I fully understand that, certainly something that requires more testing
> to see how SSSD will behave with PAM.
>
>>
>> Or are you just talking about writing a client adapter and this has nothing
>> to do with the Keycloak auth server?
>
> Good question. My initial naive idea was to have an authenticator SPI
> for PAM and benefit from the work already done by Marek with LDAP and
> Kerberos. Plus, have a federation SPI to retrieve user's data from SSSD
> and propagate it to Keycloak.
>
>>
>> Also, where does the identity data come into play (aka LDAP info)?  Is this
>> also a part of the PAM/SSSD flow?
>
> At the flow described here#17[2]:
>
> * User starts browser and connects to a resource
> * Resource redirects to Keycloak
> * User is presented with a login form
> * User fills username and password
> * User data is collected and passed to SSSD over D-Bus
> * SSSD authenticates against LDAP server
> * Authentication complete
> * Assertion/token is issued
> * User is redirected to the resource
>
>>
>> Bill
>
> [1] - https://docs.google.com/presentation/d/1-WvQTQ1M0Q9kfRl3d7FVWFn9GLL7vn8sAQmXGv0SVcs/edit#slide=id.g113bf6b186_1_130
> [2] - https://docs.google.com/presentation/d/1-WvQTQ1M0Q9kfRl3d7FVWFn9GLL7vn8sAQmXGv0SVcs/edit#slide=id.g113bf6b186_1_107

Let me try to clarify a few things.

PAM is designed as a "conversation", there are a few analogues you could 
compare it to:

* a series of requests/responses

* challenge/response authentication (e.g. CRAM)

PAM has something equivalent to a session where state is stored during 
the "conversation". When you use PAM you establish a context (session) 
and iterate. In each iteration the PAM library will ask you for 
something and you reply. The iteration stops when the library signals 
completion.

For simple password auth the iteration is very short. But depending on 
how the PAM service is configured you could be prompted for other 
things. I suspect with Web forms they way you handle this is via 
redirects until such time as the PAM conversation completes.

So my suggestion would be to design this where there is a simple web 
form prompting for username/password but allow for the fact you may have 
to redirect to another page.

Currently SSSD does not allow for authentication conversations which is 
one reason PAM is being promoted as an interim solution. The SSSD team 
does not want to add equivalent DBus authentication without carefully 
designing it. The concern is in a rush to provide authentication with 
SSSD DBus it would end up as just another broken variant of PAM (yes, 
PAM has warts, it was designed 20 years ago for Solaris and it's showing 
it's age and weaknesses, but none the less it remains the standard 
authentication interface on UNIX like systems).

Does that help?


-- 
John