[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Replacement for dspam

Am Freitag, 18. Juli 2014, 20:26:57 schrieb Erwan David:
> Le 18/07/2014 19:56, Martin Steigerwald a écrit :
> > Am Freitag, 18. Juli 2014, 18:03:20 schrieb Erwan David:
> >> Le 18/07/2014 14:10, Martin Steigerwald a écrit :
> >>> Am Freitag, 18. Juli 2014, 13:55:43 schrieb Erwan David:
> >>>> On Fri, Jul 18, 2014 at 01:45:55PM CEST, Martin Steigerwald
> >>> 
> >>> <Martin@lichtvoll.de> said:
> >>>>> Hello!
> >>>>> 
> >>>>> Am Freitag, 18. Juli 2014, 12:10:44 schrieb Jonathan Dowland:
> >>>>>> On Thu, Jul 17, 2014 at 11:26:56PM +0200, Jochen Spieker wrote:
> >>>>>>> Ok, I already fail at the first step: call crm114 from Postfix so
> >>>>>>> that
> >>>>>>> it uses user-specific spam/ham CSS files. My web search suggests
> >>>>>>> nobody
> >>>>>>> has ever done that before. I don't use procmail but Dovecot's LDA
> >>>>>>> for
> >>>>>>> mail sorting which doesn't allow to call external programs. There is
> >>>>>>> sieve_extprograms but that looks a little weird to me.
> >>>>>> 
> >>>>>> I've never tried that, I'd be interested to see it working. I use
> >>>>>> procmail
> >>>>>> for local users who have their own CSS files. I did once look at
> >>>>>> http://johannes.sipsolutions.net/Projects/crm114-spamd which provides
> >>>>>> a
> >>>>>> spamd-like interface to crm114, but I didn't go live with it. For
> >>>>>> SMTP-time
> >>>>>> rejection you will need CSS files shared across all recipients.
> >>>>> 
> >>>>> For a long time I used CRM114 integrated in KMail with a POP3 account
> >>>>> and
> >>>>> I
> >>>>> can say that CRM114 is exceptionally good at detecting spam. To the
> >>>>> extent
> >>>>> that spam just has been a non-issue for me. I am not using this at the
> >>>>> moment cause I had severe mail loss with Akonadi based KMail and
> >>>>> CRM114
> >>>>> filter rules. This may have been fixed meanwhile but I didn´t try
> >>>>> again
> >>>>> and having tried once. So I just weed out spam manually which works
> >>>>> okay
> >>>>> cause policyd-weight catches most of them on the server already.
> >>>>> 
> >>>>> I am currently in the process of setting up my Postfix + Dovecot
> >>>>> installation with IMAP additionally to POP3.
> >>>>> 
> >>>>> I already have IMAP working, route all mail to me to POP3 + IMAP
> >>>>> account
> >>>>> (with different users) and got managesieve working so I can edit
> >>>>> filter
> >>>>> rules on IMAP server with newest KMail.
> >>>>> 
> >>>>> Next step is to install dovecot-antispam and integrate it with CRM114
> >>>>> on
> >>>>> the server. Either with a copy of my existing CRM114 installation on
> >>>>> the
> >>>>> laptop or with a new setup.
> >>>>> 
> >>>>> I didn´t complete this yet. I think it can work quite well.
> >>>> 
> >>>> Did you find the tutorial foir integrating postfix and crm114 ?
> >>>> 
> >>>> On crm114 it is mentioned with a link to the "add-ons" page of postfix
> >>>> doc, whioch in turn has a link to crm114 main page...
> >>>> 
> >>>> I could not find whether crm114 is used with lmtp (as I used dspam),
> >>>> or as a milter, or another way.
> >>> 
> >>> Since CRM114 requires learning I´d use dovecot-antispam plugin (debian
> >>> package is available). It automatically learns from anything one puts
> >>> into spam folder and sports an unsure folder as well.
> >>> 
> >>> But as I didn´t do it yet, I don´t know the details.
> >> 
> >> That's the learning part, however there is also the distribution part :
> >> is the email tagged ? How is the email transmitted by postfix to crm114,
> >> and how is it transmitted to dovecot LDA ?
> >> crm114 page does not even address those basic email settings, I did not
> >> even see a command name on it, or a daemon or any thing...
> >> 
> >> What is the storage backend for the learned data ?
> > 
> > crm114 is not just for mail. Its for any kind of textual data that one can
> > through regular expressions at.
> Yes, that's what I understood, but for me I only need email bayesian
> filtering.
> > I suggest reading through docs of dovecot-antispam.
> I already use it. dovecot-antispam does NOT filter email. I use postfix
> -> dspam -> dovecot LDA all communicating through lmtp. Users have a web
> interface to check filtered spam and can ask for delivery of false
> positive. Dovecot anti-spam is used to retrain *after* delivery, I want
> on server classification before delivery.

Hmmm. Actually I think I won´t care as long as it sorts spam into spam folder 
and unsure into unsure folder and I can put unsure into a good folder so it 
gets filtered with managesieve again.

Dunno whether thats possible. Moving to IMAP is low priority here. And a lot 
of work. I almost got most mailing lists sorted by sieve, but its still some 
work left, especially with the antispam stuff.

> >  I use dovecot lda within
> > 
> > postfix already for sieve.
> And what do you use in sieve to put spam somewhere specific ? How do you
> allow users to retrieve false positive ?

As I wrote, I *didn´t* yet setup any spam filtering except policyd-weight which 
denies on SMTP level.

> > I didn´t do it, but from a short contact with the author of the plugin, I
> > got the impression that would be all I need (except Dovecot and CRM114
> > that is).
> > 
> > CRM114 uses two 12 MiB files which it mmap()s for best performance. At
> > least thats what my setup uses. I described basic setup for mail in Linux
> > User article (in german) and I think this was one of those which got
> > translated into english for Linux Magazine. Although it described
> > integrating CRM114 with KMail the basic setup of CRM114 should be the
> > same or similar. Both articles should be online freely since they are
> > older than a year.
> This does not seem to scale well with different spam/ham token set per user.

Huh? Just copy the CRM114 folder to another home and empty the spam/ham token 

But as said, the CRM114 setup part should be no different with IMAP. I didn´t 
say you have to use POP3.

> > And I think I even based this article on some doc on the mailreaver.crm
> > stuff. I think it is described in crm114 package somewhere.
> I usually prefer reading doc before installing, but I made an exception
> : it seems to me useful only for a full geek user, each user needs to
> learn a complicated language,  need access to server file system : it
> canit work for me, where only access from users are managesieve (and
> often only through roundcube sieve frontend) and imap.

Learn a complicated language? For CRM114? No.

I just trained it by sorting unsure mails as good or spam. And it learned. For 
setup all I did was filling in some configuration option in mailreaver.crm.

For about one or two days back then. And then my spam problem was gone for 
good. At that time even before I put policyd-weight onto the server. I just 
put policyd-weight there as I did not even want to download the most obvious 
spam via POP3 anymore.

> And I want per user spam/ham token sets, generic sets just do not work

Anyway, if you find something better, I am all ears.

I am pretty sure with my spam filtering setup for IMAP I will have a go at 
CRM114, cause I know how well it works.

I never got dspam working in a test setup back then. I think I even opened a 
bug report at that time. Anyway, maybe I missed something obvious, but as 
CRM114 worked for me way better than any other attempt with a learning spam 
filter I kept to it until that bug in KMail caused data loss.

Right now, spam filtering is not that urgent. My policyd-weight setup keeps 
away most of the spam and while its annoying to weed out the rest of it 
manually, its better than risking data loss again.

But I want to have it at some time again. So feel free to share when you find a 
nice solution.

Martin 'Helios' Steigerwald - http://www.Lichtvoll.de
GPG: 03B0 0D6C 0040 0710 4AFA  B82F 991B EAAC A599 84C7

Reply to: