Jeff Chan wrote on Fri, 10 Sep 2004 07:43:39 -0700:

> If you're talking about adding resolved IP addresses to SURBLs,
> no we're not going to do that.   :-(
> 
> What I'm talking about is an internal process where we keep track
> of resolved IP addresses and use that to add new domains to
> SURBLs sooner if they resolve to a similar IP range (probably
> /24s).  We would use the resolved IP addresses to add domains
> to sc.surbl.org and possibly other lists sooner.  Most would
> probably get added on the first report.  :-)
>

Haven't been reading the list for a few days I got the chance of reading 
this huuuge thread at once. What came to my mind is the following.

1. What do you do when spammers start using IPs instead of host names? 
Think ipv6, they are as good as throwaway domains. And it's in Asia where 
they are getting in big use real soon. Sooner or later you *will* need to 
handle IPs like domain names.

2. It's being said that there's a high chance of collateral damage because 
of virtual hosting. Is it? If you simply go to the sites in Chris' list by 
IP instead of hostname you find them showing a spammer page. I'd say 
there's a high probability if the default domain on that IP is a spammer 
domain all the rest will be as well. Also, how great is the chance that 
innocent by-standers on these machines send out email with links to their 
sites in them? If they are not marketing emails the chance is quite low 
and a link found in the signature the most possible appearance only. It 
should be possible to come up with a way to differentiate between a 
business card-like signature including a domain name and a spammer link. 
Furthermore, spammers need a working host which doesn't fail because it 
gets too many hits. A virtual host doesn't provide that. Also, a virtual 
host is much more risky to get shutdown quickly by an ISP than a whole 
box. Spammers probably use whole boxes most of the time, not normal 
hosting webspace.
So, this situation is quite different from mail collateral damage and the 
chance of it is much less. I'm sure there would be many people willing to 
accept that low risk.

3. On the other hand blacklisting whole /24s as you propose *is* bound to 
a much higher collateral damage I think. Same goes for checking the 
nameservers. If I understood correctly what Justin said SA looks up the 
nameservers for a domain in spamhaus with the URIDNSBL rule for spamhaus. 
Isn't that much more prone to false positives?

4. I think the way this should/could work is like following:
- you need to have a database of hostname - IP correlations. Not just 
domains, it needs to be hostnames, so you can handle 1.domain.com pointing 
at a different IP than 2.domain.com. All IPs resolved this way from 
spamvertised hostnames go in this list. And all IPs which are used 
directly in spam (see 1.).
- after SURBL lookups are done and the domain (as I understand SURBL 
lookups are done by domain) isn't found, do a lookup on the resolving IP 
and look it up in that separate SURBL list.
- if it is found there's a chance that it's a spam hosting IP. Since SA 
uses scores we (the end-user ISP) have methods of decreasing false 
positives. Also, if this rule is scored 0 by default just those people who 
are confident in this method could enable it.
- report the hostname associated to that IP and which (the hostname) isn't 
in SURBL yet automatically to SURBL, so you can put it on a "normal" SURBL 
list after you get more confirming reports.

5. What about reporting back? This could build up confidentiality in rules 
and BLs. F.i. report back the spam score of that message which is 
associated to this host name/IP right after SA finishes with scoring. You 
can build a statistic which shows the resultant spam score distribution of 
messages which contain that host name/IP. If there's too many hammy scores 
you throw it out. This could be similarly used for rules and may be an 
alternative way to mail corpora runs of getting to know how well the rules 
behave. Of course, this can't be on by default and needs to be well 
thought thru to get good results which *mean* something. Also, if someone 
wanted to enable this he had to register first, so you don't get wrong 
scores from spammers. Just an idea.


Kai

-- 

Kai Schätzl, Berlin, Germany
Get your web at Conactive Internet Services: http://www.conactive.com
IE-Center: http://ie5.de & http://msie.winware.org



Reply via email to