Multiple-Mechanism Sample Code?

[Henry B. Hotz]

On Dec 18, 2006, at 2:12 PM, Alexey Melnikov wrote:

> Henry B. Hotz wrote:
>
>> The published sample code seems to only try the first mechanism  
>> and  then quit.  I'm told the "correct" way to do SASL is to try  
>> all the  mechanisms (or at least all the ones supported) and don't  
>> quit until  you've tried them all.  Is there any example code that  
>> illustrates this?
>
> (I wanted to point you to Cyrus imtest, but it doesn't do that).

It's worse than that.  |-(

I kept looking after I posted that note, and I haven't even found any  
suggestion that it was feasible beyond Simon Wilkinson's comment at  
the UMich workshop.

In a(n admitedly limited) test calling sasl_client_start multiple  
times with the same mechanism list always results in the same  
mechanism being chosen.  This seems to mean that you need to  
duplicate the list parsing code in your app and keep track of the  
untried mechanisms yourself.  It also means I need to play games in  
order to avoid prompting the user for a username once for each  
mechanism that I try.

Now that I'm trying to actually use the API there are a number of  
other things that seem awkward as well, but that's OT.

> In general, I think a well written SASL client should behave as  
> follows:
>
> It should sort SASL mechanisms that both client and server support  
> by their "strength" or features recognized by the client. For SASL  
> mechanisms with equal strength the order used by the server can be  
> used.

How *does* the server order them?

Putting "GSSAPI ANONYMOUS" in the config file results in the same  
list.  Putting "GSSAPI PLAIN" in the config file results in "PLAIN  
GSSAPI" from sasl_listmech().  Whyzzat?

I don't think I'm supposed to care about the ordering.  The security  
considerations section of the SASL RFC says the mechanism is  
negotiated in the clear, and you should expect that a 3rd party may  
modify it.  That means that *any* mechanism accepted by the server or  
client under any circumstances had better be "good enough".  That  
said I would much prefer GSSAPI/Krb5 to a mechanism that requires me  
to type in stuff.

> The client starts iterating through the ordered list, starting from  
> the strongest mechanism. It tries the mechanism. If authentication  
> succeeds - success. If not, the client may retry the mechanism  
> (e.g. if the server returned an indication that the password is  
> incorrect) several times, say 3 times. After that the client should  
> move on to the next strongest SASL mechanism and so on.

Since the password prompt is very early after you start the client I  
don't think it's a big deal to just start the client over again.   
(Usually, not always!)

> There are of course some complications. Some SASL mechanisms that  
> can potentially be stronger can end up being weaker, because of the  
> options that the server supports.

------------------------------------------------------------------------
The opinions expressed in this message are mine,
not those of Caltech, JPL, NASA, or the US Government.
Henry.B.Hotz at jpl.nasa.gov, or hbhotz at oxy.edu