4:44 a.m.
Thanks for the clarification Bill.
So in your case we know in advance that there is a single mechanism
which should be the only mechanism sending challenges and at least one
more mechanism that can perform the authentication without the need to
send challenges.
I am just wondering for this scenario if we can handle this better
during registration of the mechanisms rather than at runtime processing
a request.
Regards,
Darran Lofthouse.
On 20/12/13 00:12, Bill Burke wrote:
On 12/19/2013 12:13 PM, Darran Lofthouse wrote:
> Coming back to this topic ;-)
>
> The reasons I believe it is the authenticate method that should signal a
> mechanism wants exclusivity in sending a challenge are: -
>
> 1 - This method already returns one of three values to indicate it's
> authentication state, adding one more value to indicate that a mechanism
> has started authentication but wants exclusivity would fit with this.
>
> 2 - The mechanism doesn't need to be first just to be able to request
> exclusivity - also not going to cause a complication if another
> mechanism had added it's challenge as that would not be able to happen.
>
> 3 - In general we should try and avoid writing mechanism that only work
> alone (depending on various constraints), having a sendChallenge that
> always requested exclusivity would go against this - whereas a mechanism
> detecting it wants exclusivity in the authenticate stage would be valid
> as remaining mechanisms would still be usable for the times it does not
> request exclusivity.
>
#3 just doesn't work for me.
I have two mechanisms for keycloak:
A) Bearer token mechanism that uses the Authorization header and
challenges with a 401 and WWW-Authenticate
B) OAuth browser redirection mechanism, that challenges with a 302 and a
Location header. This is a redirect to the authentication server where
it will display a login form.
Mechanism (B) will not ever request exclusivity during the authenticate
phase because keycloak also allows for (A) in the same deployment.
Mechanism (B) will not ever be able to challenge if it can't set a 302
response code. I worry that if other mechanisms add WWW-Authenticate
headers to a 302 response, it may confuse the user-agent, but this worry
might be bogus.
So, I don't agree with #3's statement that "having sendChallenge that
always requested exclusivity would go against this". Mechanism (B)
works with a other mechanisms as long as the user-agent is doing
preemptive authentication.
Whether Keycloak is successful or not is besides the point. The use
case where an auth mechanism needs to redirect to another machine to
display a login form is a valid one.
Bill