Re: [Exim] local_scan(), content-scanning, multiple recipien…

Top Page
Delete this message
Reply to this message
Author: Chris Edwards
Date:  
To: exim-users
Subject: Re: [Exim] local_scan(), content-scanning, multiple recipients and bounces
On Wed, 5 Feb 2003, Ollie Cook wrote:

| Hi,

|
| I'm delevoping a content scanner (spam and antivirus), where each individual
| recipient at our site may have different settings for
| accepting/rejecting/marking mails that match, using Exim's local_scan()
| function but am having trouble decided how to treat messages with multiple
| recipients.


We've been pondering this too...


| For example, if recipient A has spam rejection enabled and recipient B does not
| I will need to accept the message and deliver it to B only (by manipulating the
| recipients vector), but also generate a bounce message for the sender,
| announcing that recipient A did not receive the message. (The bounce will
| probably go nowhere, but we have to have tried, in case the 'spam' was in fact
| legitimate.)

|
| Since I would not be able to give a 5xx permanent fail (which would suggest
| both recipients failed), I would have to accept the mail with 2xx and then
| subsequently generate a bounce message.

|
| I'd rather be able to reject messages based on content at SMTP time, so it's
| the remote site's responsibility to handle bounces (given that the majority of
| spam bounces will just sit on my queue for a while), but I suspect that's not
| possible.


Correct :-(


| I was really just wondering (after all that), if this is what other
| people do in this situation, or if other schemes have been devised?


We thought about the sort of thing you suggest, but it seemed too much
hassle...

In the above example where A has spam rejection enabled and B doesn't,
we'll just we accept the message and deliver to both A+B. Not perfect,
but in practice its impossible to eliminate 100% of spam. We simply aim
to remove "most" spam.

Our plan is to make spam content filtering opt-*out*. With any luck
relatively few folk will opt out and the scenario in question will be
rare...

chris

--
Chris Edwards, Glasgow University Computing Service