You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@spamassassin.apache.org by Stephen Carville <st...@gmail.com> on 2010/03/10 17:43:59 UTC
Inconsistent Application of Rules?
I am using spamassassin v 3.2.5 invoked via spamc from Postfix v2.3.3.
My local.conf file is pretty basic. The only change is to shut off
the FH_DATE_PAST_20XX rules which triggers for every message from Jan
1, 2010 on
required_hits 5
report_safe 0
rewrite_header Subject [SPAM]
skip_rbl_checks 0
score FH_DATE_PAST_20XX 0
I've been seeing several emails lately that are being scored low that,
from what I know of the SA rules should be scored higher. A recent
example was a typical spam message:
--- Begin ---
March 10, 2010
12:59:09
---> Enter shop here [URL deleted because it causes my message to be rejected]
<bunch of random sentence frasgments deleted>
--- End ---
The original incoming message arrived in my inbox with the following score:
x-spam-status: No, score=2.1 required=5.0
tests=BAYES_50,RDNS_NONE,URIBL_BLACK autolearn=no version=3.2.5
x-spam-checker-version: SpamAssassin 3.2.5 (2008-06-10) on cadmzmx01.lereta.com
x-spam-level: **
Out of curiosity I copied the original message body and sent it to
myself from a test address. The headers this time looked like:
x-spam-status: No, score=4.4 required=5.0 tests=AWL,BAYES_00,
FROM_STARTS_WITH_NUMS,RCVD_IN_DNSWL_LOW,URIBL_AB_SURBL,URIBL_JP_SURBL,
URIBL_OB_SURBL,URIBL_SC_SURBL,URIBL_WS_SURBL autolearn=no
version=3.2.5
x-spam-checker-version: SpamAssassin 3.2.5 (2008-06-10) on cadmzmx01.lereta.com
x-spam-level: ****
The second message invoked a larger number of body check rules than
the first but I don't understand why. Is that normal or do I have
something configured incorrectly?
(I tried sending the complete headers but that message was bounced as spam!)
--
Stephen Carville
Re: Inconsistent Application of Rules?
Posted by John Hardin <jh...@impsec.org>.
On Wed, 10 Mar 2010, Stephen Carville wrote:
> On Wed, Mar 10, 2010 at 9:14 AM, John Hardin <jh...@impsec.org> wrote:
>
>> It looks like a simple matter of a very short spam with a URI that
>> wasn't broadly recognized as bad the first time you saw it. Train your
>> bayes with it, and consider adding greylisting to give the URIBLs a
>> chance to get updated with new spam domains.
>
> I am haven't turned on Baysean filtering yet. I still need to test
> for the false positive rate.
Turn it on and set all the BAYES_* scores to 0.01, then look for
BAYES_[6789] in ham. You can do that in your mail log on the server
without violating anybody's mailbox privacy.
The reliability of Bayes is based entirely on how well you train it.
And again, you might consider turning on greylisting for sites you don't
regularly correspond with, to allow URIBLs time to list new spam domains.
--
John Hardin KA7OHZ http://www.impsec.org/~jhardin/
jhardin@impsec.org FALaholic #11174 pgpk -a jhardin@impsec.org
key: 0xB8732E79 -- 2D8C 34F4 6411 F507 136C AF76 D822 E6E6 B873 2E79
-----------------------------------------------------------------------
Failure to plan ahead on someone else's part does not constitute
an emergency on my part. -- David W. Barts in a.s.r
-----------------------------------------------------------------------
3 days until Daylight Saving Time begins in U.S. - Spring Forward
Re: Inconsistent Application of Rules?
Posted by Stephen Carville <st...@gmail.com>.
On Wed, Mar 10, 2010 at 9:14 AM, John Hardin <jh...@impsec.org> wrote:
> On Wed, 10 Mar 2010, Stephen Carville wrote:
>
>> ---> Enter shop here [URL deleted because it causes my message to be
>> rejected]
>
> Please publish the entire message including all headers to someplace like
> pastebin, and send us the link.
http://www.heronforge.net/files/spam.message.01.txt
http://www.heronforge.net/files/spam.message.02.txt
> The RCVD_IN checks being different aren't surprising as the test message
> came from a different source. The URIBL hits may be due to timing - i.e. the
> URI was added sometime between the original receipt of the message and your
> later test. Whether the original sender should have hit DNSBLs we can't tell
> until you post the original headers.
That would explain the apparent anomaly.
> It looks like a simple matter of a very short spam with a URI that wasn't
> broadly recognized as bad the first time you saw it. Train your bayes with
> it, and consider adding greylisting to give the URIBLs a chance to get
> updated with new spam domains.
I am haven't turned on Baysean filtering yet. I still need to test
for the false positive rate. Until a week ago my employer used
Forefront for spam filtering and about 10% of the messages it caught
were legitimate and users were still getting an estimated 72
undetected spam messages per user per day. That made management
nervous about filtering so I'm taking baby steps at turning on some
tests. So far the UCE controls in Postfix combined with
Spamassassin's default settings have reduced the undetected spam to an
estimated 2 - 3 per user per day.
--
Stephen Carville
Re: Inconsistent Application of Rules?
Posted by John Hardin <jh...@impsec.org>.
On Wed, 10 Mar 2010, Stephen Carville wrote:
> ---> Enter shop here [URL deleted because it causes my message to be rejected]
Please publish the entire message including all headers to someplace like
pastebin, and send us the link.
> tests=BAYES_50,RDNS_NONE,URIBL_BLACK
>
> Out of curiosity I copied the original message body and sent it to
> myself from a test address. The headers this time looked like:
>
> x-spam-status: No, score=4.4 required=5.0 tests=AWL,BAYES_00,
> FROM_STARTS_WITH_NUMS,RCVD_IN_DNSWL_LOW,URIBL_AB_SURBL,URIBL_JP_SURBL,
> URIBL_OB_SURBL,URIBL_SC_SURBL,URIBL_WS_SURBL autolearn=no
> version=3.2.5
> x-spam-checker-version: SpamAssassin 3.2.5 (2008-06-10) on cadmzmx01.lereta.com
> x-spam-level: ****
>
> The second message invoked a larger number of body check rules than
> the first but I don't understand why. Is that normal or do I have
> something configured incorrectly?
The RCVD_IN checks being different aren't surprising as the test message
came from a different source. The URIBL hits may be due to timing - i.e.
the URI was added sometime between the original receipt of the message and
your later test. Whether the original sender should have hit DNSBLs we
can't tell until you post the original headers.
It looks like a simple matter of a very short spam with a URI that wasn't
broadly recognized as bad the first time you saw it. Train your bayes with
it, and consider adding greylisting to give the URIBLs a chance to get
updated with new spam domains.
--
John Hardin KA7OHZ http://www.impsec.org/~jhardin/
jhardin@impsec.org FALaholic #11174 pgpk -a jhardin@impsec.org
key: 0xB8732E79 -- 2D8C 34F4 6411 F507 136C AF76 D822 E6E6 B873 2E79
-----------------------------------------------------------------------
Failure to plan ahead on someone else's part does not constitute
an emergency on my part. -- David W. Barts in a.s.r
-----------------------------------------------------------------------
4 days until Daylight Saving Time begins in U.S. - Spring Forward
Re: [sa] Inconsistent Application of Rules?
Posted by Charles Gregory <cg...@hwcn.org>.
On Wed, 10 Mar 2010, Stephen Carville wrote:
> I've been seeing several emails lately that are being scored low that,
> from what I know of the SA rules should be scored higher. A recent
> example was a typical spam message:
> FROM_STARTS_WITH_NUMS,RCVD_IN_DNSWL_LOW,URIBL_AB_SURBL,URIBL_JP_SURBL,
> URIBL_OB_SURBL,URIBL_SC_SURBL,URIBL_WS_SURBL autolearn=no
> The second message invoked a larger number of body check rules than
> the first but I don't understand why. Is that normal or do I have
> something configured incorrectly?
The extra rules are all 'SURBL' blocklist tests which check the embedded
URI against internet blocklists. It is not uncommon for the first few
spams using a new URI to get through before the blocklists are updated.
By the time you reran your tests, they had been updated, and so it scored
higher....
- C