> All the major ISP and email providers (such as gmail) use "mail farms" > to handle outbound SMTP connexions: mails in the output queue may be > processed (and retried) by whichever server in the farm. gmail is the only such farm I've encountered to both rotate servers on sending attempts AND exceed the number of servers to cause a problem. Now that I've been alerted to lazyaw, this is less of an issue (since popular sites send mail to/from lots of people, and other "mail farms" won't be as extensive as gmail, which I've already whitelisted). Scary "fun fact" - gmail lists almost 140,000 IPs in their SPF record. > So workarounds *must* be taken in greylist.conf: > - define "lazyaw" and "subnetmatch" to be more tolerant about incoming > submissions (sounds quite dirty to me); I didn't know about lazyaw. That almost solves my issue, though it breaks with free mail, universities, ISPs, and other giants (which is of course why I suggested using the sender address if the domain has 4+ MX records). To formalize: I propose an optional argument to lazyaw specifying the number of MX records a domain should have in order to make use of the sender address (so my earlier proposal would be implemented with "lazyaw 4"). Not specifying a number would make the default very high, depending on what Emmanuel et al think is a good default high threshold (maybe 5?); hotmail has only four (each of which maps to three IPs), and other big players seem to have 5+. 255 should be enough to disable the threshold. "subnetmatch" would end up whitelisting compromised DHCP subnets (which is likely common thanks to spammer viruses and thus very "dirty" indeed). My SPF suggestion should alleviate the need for subnetmatch on domains that properly use SPF. I'll call this feature proposal "awbyspf" and define it as "autowhitelist all of a domain's SPF-specified IPs when one passes." See also http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=445841 > - locally whitelist by ACL (IP or domain match) all known server farms > (a pain to maintain, but quickly responsive in case of problem discovery); This is in place, but I again agree that this is a pain to maintain and not an acceptable long-term strategy. > - make use of centralized DNS whitelists such as list.dnswl.org by a > DNSRBL ACL clause; hrm. that's an idea (since i've already compiled dnsrbl support in). http://www.dnswl.org/tech doesn't list configuration help for using dnswl with milter-greylist. it appears nontrivial due to their different response codes and milter-greylist's apparent need to know the response ahead of time. (Or perhaps I don't understand the option.) Advice? > - whitelist on behalf of sender's SPF record. I can't see using milter-greylist without the "nospf" option as responsible. If you're suggesting manual insertion of known problematic domains based on the mail servers listed in their SPF records, yes, I do that manually. If there were an option for this, that would be nice (it would be a trivial add-in if my "light lazy" feature is added). Ondrej Valousek wrote: >> racl whitelist spf pass It took me a while to figure out what "racl" was (I'm using milter-greylist 3.0 since 4.x isn't in debian unstable yet). It turns out that "racl" is a synonym for "acl," so your line appears to simply undo the "nospf" option that I consider a must-have. My "awbyspf" proposal autowhitelists all SPF records for a domain once one of them becomes autowhitelisted. milter-greylisting's built-in SPF capabilities only implement the first half of that (all SPF servers for all domains are always whitelisted).
Message
Re: [milter-greylist] greylisting delay sometimes in hours instead of minutes?
2008-03-11 by Adam Katz
Attachments
- No local attachments were found for this message.