RE: [saml-dev] Authentication on IDP.

From
Giuseppe Sarno <>
Date
2005-11-04T13:45:56+00:00
ID
Thread
RE: [saml-dev] Authentication on IDP.
Title: Message

Hi 
many thanks,

Actually now that you mentioned the word "stronger"

I've 
seen few Identity enabled servers like the IBM webseal that have a Level of 
authentication.

 

Where 
Level 1 is better that level 0 and so on without actually defining what Level 
0/1 etc means.

 

Now 
SAML seems not defining levels but rather types or classes of authentication 
mech.

SAML, 
though,  allows cases where an SP can use the word or "better","stronger", 
etc. but if I have understood right 

the 
fact that a class is better or stronger of another is not part of the 
spec and it's implementation specific at the IDP.

 

Is 
this right ?

 

Thanks.

Giuseppe. 

  

  
-----Original Message-----
From: Conor P. Cahill 
  [mailto:] 
Sent: 04 November 2005 
  13:07
To: Sarno, Giuseppe [MOP:GM15:EXCH]
Cc: 
  
Subject: RE: [saml-dev] Authentication 
  on IDP.

Sorry for the delay in response... was out of touch yesterday (and 
  will be again later today).

Giuseppe Sarno wrote on 
  11/2/2005, 9:44 AM: 
  

  
    
      
        
          
do we have 1 
          certificate for all the request at the IDP ? IMPORTANT: but then how 
          the IDP (and in the end the SP) distinguish which user is that request 
          from? 

    
User A access SPA.

    
SPA redirect to IDP asking for a Certificate based 
    authentication.

    
IDP perform this but (unless I'm wrong) the certificate wouldn't 
    contain any info specific to the user but mainly keys 
    info.

    
in this case the IDP has to also identify the user in order to reply 
    with a subject to the SP. I guess it would perform a 
    Userid/password (basic auth)

    
authentication.

    
I guess this is reasonable to expect from the 
    IDP.
This is out of scope for SAML, but I'm 
  guessing that most IdPs will either a) have a certificate registered for a 
  particular user  or b) they will have a requirement that user identity 
  information be present in the signed certificate (this is  how many web 
  sites do client-auth ssl).  

The certificate is usually enough to 
  identify the user (assuming it was an individually issued 
  certificate).

Of course, there is nothing that stops an IdP from doing 
  exactly what you are descrbing (in fact, I *think*many VPN clients do this 
  kind of thing (where the client authenticates to the server using a client 
  cert that is the same for many users and then the user authenticates ontop of 
  that communciations channel)).

  
    
Now What is the Auth context he is going to send back 
    ?  

    
    Certificate based authentication 
    ?

    
    User Id/ password 
    ?

    
    or both 
    ?
I'm guessing that if the IdP were to do what 
  you are describing, they would define a new class that was along the lines of 
  ...PasswordClientTLS (what I would consider to be "stronger" than 
  ...PasswordProtectedTransport).

Although, like I said earlier, I would expect that the 
  certificate alone would be good 
enough.

Conor