Re: FW: Redirection (was UNH Plugfest 5)

[Paul Koning <pkoning@equallogic.com>]

>>>>> "Julian" == Julian Satran <Julian_Satran@il.ibm.com> writes:

 Julian> David, The only way to do it cleany the way you want it is to
 Julian> allow the redirect response (0101 and 0102) only in
 Julian> operational parameter stage.  But that seems rather
 Julian> excessive. If we want to mandate a single way of handling I
 Julian> would suggest stating that 0101 and 0102 SHOULD be accepted
 Julian> even during authentication (Paul's POV). Again I don't thing
 Julian> it adds anything as local policy may prevent an initiator
 Julian> from considering those values.

"SHOULD" is helpful in that at least it gives a recommendation.  But
it is not good enough.

We want to build targets that interoperate with all initiators.  Right
now, the spec simply does not permit us to achieve this goal.  

I've expressed a preference in how things would work, but it doesn't
matter a whole lot which way things go.  

Right now, we have an implementation that will issue a redirect before
completing the full authentication handshake.  Most initiators accept
this, but some do not.

We're perfectly willing to change it so the target does complete the
whole authentication handshake, and only then sends the redirect.  But
we're afraid to do so because the spec does not require initiators to
accept that either! 

So we're faced with a known interop problem, and if we change the
behavior to the other possible way we are at risk of running into
other initiators that don't like doing things THAT way.

So pick one, but it MUST be a MUST.

1. An initiator MUST accept a redirect from a target that has
   completed the authentication handshake; it MAY (or SHOULD) accept
   it from a target that has not yet completed the handshake.

or

2. An initiator MUST accept a redirect from a target that has not yet
   completed the authentication handshake; it SHOULD (or MAY) accept
   it from a target that has completed the handshake.

Well, of course there's a third alternative, which is to require that
both alternatives MUST be accepted.  I do not propose that but would
not object to it.

    paul