Re: [exim] Filtering spam bounces

On Mon, 15 May 2006, John W. Baxter wrote:

> On 5/15/06 3:12 AM, "Alastair Campbell" <ac@???> wrote:
>
> > I'm afraid it's too late, I've used it over the last 5 years, and
> > didn't keep track of all the addresses used.
>
> As part of my long-running campaign to get rid of the catchall in my
> domain, I have now changed the settings so that catchall mail goes
> to the cleverly named "invalid" mailbox (which I read once or twice
> a week in an MUA devoted to the purpose).

Seems to me there's two issues here. One is mail (maybe bona fide) to
the catchall addresses which A.C was *previously* using (but
presumably is no longer using). The other is bounces (mostly the
result of misguided reactions to spam with faked sender addresses).

Since these legacy addresses are presumably no longer being used for
sending out mail, I don't see why A.C can't immediately start
rejecting bounces to those addresses. He's hardly going to get a
genuine bounce relating to a mail which e.g he sent last year, is he?

It *might* have to be done with a little care, depending on whether he
wants to respond affirmatively to callouts for those obsolescent
addresses.

My gut reaction is that I'd prefer to repudiate obsolescent addresses
being tested by callouts, even if I was willing to accept mail (which
could well be bona fide) with a non-null envelope sender addressed to
that obsolescent address.

This would also mean that third parties who validate dubious mail by
attempting a callout on the purported envelope sender would be able to
reject faked senders on the spot, which is good (I'm no fan of
*blanket* callouts to validate sender addresses, but when things look
dubious I still reckon they can be a handy extra string to the bow).

But maybe there's some reason for still declaring these addresses
valid in response to callout, in which case you'd need to return
affirmative responses to such "bounces" at the MAIL FROM and RCPT TO
phases (you can't, after all, in these phases tell the difference
between a callout and a "real bounce"), but you'd then return 5xx at
the DATA phase.

regards