> I think Roland answered most of this already - except to say that the way to
> exclude messages from refiltering was to set an X-Already-Seen-By-My-Filter
> header so that you could skip re-running if you saw that header. Suggested
> injection method was I belive by sendmail call.
> 

The X-Already-Seen-By-My-Filter will stay in the mail message,
thus is not secret. So this works for filters that allow the
sender to skip them. (I don't happen to have an example of such
a thing, but I believe there are cases...)

Julian Mehnle wrote:
> 
> That's a bad idea, since any spammer or virus could send messages with
> that header field already included, and thus bypass filtering.

That's basically correct.

"Mitch (WebCob)" wrote:
> 
> 1) there is no reason this would remain constant

If the filter digitally signes the already filtered content, then that works.
Otherwise the filter has no method to distinguish spoofed headers, except
perhaps looking up a database based on some key...

> 2) as this value is only used locally, it wouldn't be that quick to leak
> (relative to other leaks of addresses

The header would be readable to all users. Unless the filter overwrites
its content on-file: E.g. reinject with X-Already-Seen-By-My-Filter: 1234
but upon finding it overwrite on-file as X-Already-Seen-By-My-Filter: ****

> 3) it would be very easy to rotate this regularly to keep it from being guessed

I don't think you can make assumption as to when the reinjected message will
come around. Can that be a couple of days? So old values must remain valid.

> 4) there would be a trail of received headers that could be used to validate it

But it can be spoofed...
 
> Don't think it's the best option, but this was the best idea when this was last
> discussed.

It seemed to me as well, but when I tried to write it it made no sense any more.

> I don't know WHY we can't change the architecture to allow filters
> to modify the messages... perhaps someone with more C knowlege than I can look at it?

The explanation I recall is that filtering happens while courier is in the middle
of something and it keeps offsets to various mime entities. If that's true the trick
in (2) above might work.

> [...] compare with a patch to courier that would allow modification of
> message files during global filtering. Although such an option may lower the
> efficiency of delivery perhaps if submitted as a compile time option it
> could be acceptable for inclusion in the distribution.

I think the architectural change you mention could be done without having to
re-parse each message after any filtering. E.g. there could be an api that
manages the single mime entities and that is used by courier an filters alike.
But I bet this is not one of Sam's priorities.

Ale


-------------------------------------------------------
SF.Net is sponsored by: Speed Start Your Linux Apps Now.
Build and deploy apps & Web services for Linux with
a free DVD software kit from IBM. Click Now!
http://ads.osdn.com/?ad_id=1356&alloc_id=3438&op=click
_______________________________________________
courier-users mailing list
[EMAIL PROTECTED]
Unsubscribe: https://lists.sourceforge.net/lists/listinfo/courier-users

Reply via email to