ietf-smtp
[Top] [All Lists]

Re: Recap: Re: rfc2821bis-03 Issue 32: "MUST take responsibility"

2007-05-01 09:24:22

Hector Santos wrote:
 
    When a successful transaction occurs at a SMTP-server (a message is
    accepted by the server), the protocol requires that a server MUST
    begin the process of delivering the message or begin the process
    of sending a failure notification as described in section 6.1
    "Reliable Delivery and Replies by Email."

That's likely not what you mean.  An SMTP-server accepting a mail
for the purpose of sending a failure report is broken, it should
have rejected it.  With that I get:  "MUST begin the process of
delivering the message and, if that fails, send a non-delivery
report as described" etc.  This "process of delivering" sounds a
bit awkward, but I've no better idea at the moment.

    Please keep in mind that local site policies is the exception
    to the SMTP protocol technical server requirement for mail
    delivery/failure notification.  Please refer to section 7.9
    regarding the scope of operation of SMTP Servers.

This "local site policy" business is very unclear.  If your local
site consists of an MX doing its duty and a final delivery mailer
happily rejecting anything it doesn't like the setup is dubious,
even if "accept + bounce" happens to be your "local site policy".

Think in terms of the "new developer" when reads this new standard
and begins to code his SMTP program, you want him to design the
model based on this technical Delivery/Notification principle that
is essentially burned into our society.

Especially for this new developer it's important to understand that
his code might be not used in isolation for a "one server does it
all, talk to stangers up to final delivery" scenario.  The duties
are different at the border or behind it.  Once a mail has been
accepted the top priority is "try to deliver", reject or bounce is
no issue anymore, it's too late, both result in a bounce.

At the border the top priority is to decide "should I accept this",
ideally it's the point of no return.  Some parts of the code for a 
border MTA aren't used behind it, you're not doing "CBV" etc. if
you get mail from your own MX.  The opposite also holds, the code
for final delivery isn't needed for an MX not doing final delivery.

    7.9.  Scope of Operation of SMTP Servers

7.9 as is is much clearer, in essence it says that there SHOULD be
code to get the crucial decision at the border right.  That's no
foggy "site policy", it's a consequence of the SMTP architecture.

Frank


<Prev in Thread] Current Thread [Next in Thread>