Roller Network Help - SpamAssassin Content Filter
Roller Network utilizes the popular and powerful SpamAssassin content scoring filter from the Apache SpamAssassin Project. When enabled, SpamAssassin is the final stage in our mail filters with the task of inspecting the message headers and body and applying rules-based scoring.
Configuration
The SpamAssassin content filter may be enabled or disabled on a per-domain basis. When enabled, a message is scanned when at the SMTP "end-of-DATA" stage. Depending on the configuration, the message may be rejected at SMTP time or permitted with several different reporting options.
In SpamAssassin, various rules are triggered during scanning that add up to a total score. This score is the minimum score at which the SpamAssassin filter will operate; anything scoring lower will not be considered spam, while any score greater than or equal is considered spam. This minimum score is configured as the Required SpamAssassin Score value. The default score is 5.
If a message is scored as spam there are two handling options: Reject the message (default) or Allow the message. In either case, a report of matching rules and total score value is logged in the mail log. When you choose to reject the message, an SMTP error is returned at end-of-DATA. If you choose to allow the message, there are three additional options available to customize the handling process.
The first option is to choose the spam report mode. There are three options: attach the original message, add X-Spam headers only, or attach the original message as text/plain. The default option is to attach the original message (as message/rfc822) to a SpamAssassin report that includes a detailed explanation of why the message was scored as spam. A safer option is to choose to attach the original message as text/plain which will "disarm" the original message and prevent mail clients that automatically run attachments from doing so. The third option, add X-Spam headers only, will exclude the report and only add X-Spam-* headers.
The subject tagging option allows you to choose between Prepend tag value to Subject (default) which will prepend the configured subject tag value (default ** SPAM) to the subject of any message that scores as spam, or do not tag Subject which will not make any changes to the message subject regardless of spam score. The subject tag value may be customized with a limited set of ASCII values.
The third option Maximum Allowed Score acts as a absolute maximum score. If a message scores at this maximum or higher, it will be rejected at end-of-DATA regardless of the other settings. This option allows you to permit messages that exceed the minimum required score and reject exceptionally high scores. The default value of this option is 20. These three classes of scoring with SpamAssassin can be seen as: not spam, might be spam, and definitely spam.
Bayesian Filtering and Learning
SpamAssassin offers optional Bayesian filtering and learning. By default, these options are disabled. The learning function and application of the filtering rules may be enabled or disabled independently. Using Bayesian classification, SpamAssassin is able to calculate a probability that a message is or is not spam. Our Bayesian database is distributed, allowing all of our mail servers learn and classify using a safe-queuing method to send updates to the master copy. Bayesian databases are unique to each mail domain.
To enable the Bayesian scoring function, set the Bayesian Filtering Rules option to "Enabled". Likewise, set Bayesian Learning to "Enabled" to start collecting messages for classification as spam or non-spam. If this is your first time using the Bayesian filter, it will not apply scoring until at least 200 spam and non-spam messages have been learned. The current status of your learning database is shown in the Learning Status field. Bayesian databases are unique to each domain name.
During the learning process (whenever the Bayesian Learning option is enabled), SpamAssassin will calculate a learning score that is used to determine if a message should be fed into the learner as spam or non-spam (also known as "ham"). You can adjust the thresholds the auto-learner uses for this classification through the Learning Thresholds options. The minimum setting is the threshold below which a message needs to score in order to be automatically learned as a non-spam message. Likewise, the maximum setting is the threshold above which a message needs to score in order to be automatically learned as spam. (Internally, SpamAssassin requires 3 header points and 3 body points, so the minimum effective value for maximum is 6.)
In some instances it is undesirable for SpamAssassin to learn from certain sources, such as a mailing list where spam samples are shared. In order to accommodate this, we provide settings for Bayesian Learning Exceptions. To modify your exception list, click on the "exceptions" link next to the Bayesian Learning option. You may add exceptions for sender domains, sender addresses, recipient domains, and recipient addresses. If the auto-learner encounters one of these during the learning processes, the message will be ignored. Don't add your own mail domain to the exception list; doing so will cause learning to ignore all messages.
Custom Scoring
Every SpamAssassin rule has a different scoring value. The value may be positive or negative to affect the outcome of the total score, or 0 (zero) to disable a particular rule. You may customize these scores with our custom scoring configuration.
To customize the score from a rule, either select it from the drop down or enter the specific rule name (for example, "MISSING_SUBJECT"). Scores may be entered as positive, negative, or 0 to disable the chosen rule. Custom scoring will be displayed in a table below. You may delete custom scores at any time.
Roller Network Globals
The following global configuration rules are currently in effect:
Configuration Examples
SpamAssassin content filter configuration example:
Bayesian learning exceptions configuration example:
