Filters

In this section there is a list of available filters and detailed description of each filtering algorithm.

Same long message body

When there is a SPAM being sent using XMPP server in most cases number of messages with longer body size increases and in most cases every SPAM message contains same body part. This filter is identified by following id message-same-long-body.

Detection is based on:

  • message body being longer that particular value
  • multiple messages being sent with same long body

Below is list of possible settings which may be modified to adjust this filter behaviour.

Message body length

SPAM messages are usually longer messages (over 100 chars). To reduce overhead of filtering and memory required for filtering we check length of message body and process it further only if message exceeds declared message body length (default: 100 chars).

You can also check messages with smaller body (ie. only 50 chars) by setting body-size property to 50.

Setting filter to check message with body bigger that 50 chars. 

'sess-man' () {
    'spam-filter' () {
        'message-same-long-body' () {
            'body-size' = 50
        }
    }
}

Number of allowed message with same body

In most cases message with same body is sent to multiple users. Filter will count messages with same body (which is bigger that declared message body length) and if it exceeds message number limit then any further message with same body will be detected and marked as SPAM. By default we allow 20 messages with same body to be processed by SessionManager. If you wish to change this limit set number-limit to appropriate value.

Setting number of allowed message to 10. 

'sess-man' () {
    'spam-filter' () {
        'message-same-long-body' () {
            'number-limit' = 10
        }
    }
}

Size of counters cache

We process every message and for every body of message which body length exceeds body length limit we needs to keep counter. This counters are kept in cache which size is configurable and by default equals 10000. To change size of counters cache assign proper value to counter-size-limit.

Increasing cache size to 1000000. 

'sess-man' () {
    'spam-filter' () {
        'message-same-long-body' () {
            'counter-size-limit' = 1000000
        }
    }
}

Error message and missing <error/> child

Some of SPAM messages are sent as stanzas which are invalid if we compare them with XMPP specification, ie. <message/> stanza with type attribute set to error are sent without child element <error/> which is required for all packets of type error. This filter detects this kind of messages and marks them as SPAM.

This filter is identified by following id message-error-ensure-error-child.

Groupchat messages sent to bare JID

In some cases SPAM messages are being sent as groupchat messages (messages with type attribute set to groupchat). With this type of messages we cannot use filtering based on number of message sent with same body as in case of MUC messages we must accept a lot of messages with same body, because there may be many users which are participants of same MUC room and should receive same message.

To address this issue we decided to drop all groupchat messages which are sent to our server XMPP users with to attribute set to bare jid, as real MUC component is aware of user resources which joined particular room and will send messages only to this particular resource by addressing message with full jid. This filter is identified by following id muc-message-ensure-to-full-jid.

Known spammers

To deal with spam it is required to filter every messages to verify if it is spam or not. Usually spammers are using same accounts to send bigger number of messages. This filter takes it as an advantage of this to reduce time required for filtering spam messages as when any other filter marks message as spam this filter will be notified and will mark senders jid as a spammer. This will result in a ban for any packet exchange with this user for configured ban time.

If user will send a burst of spam messages then he will be banned for configured ban time for every spam message, ie. if user would send 20 messages and ban time will be set to 15 minutes then users will be banned for 300 minutes (5 hours).

This filter is identified by following id known-spammers.

<ban-time> <title>Ban time</title>

Time in minutes for which user marked as spammer will not be able to exchange packets with any other users. By default this value is set to 15 minutes and if you would like to increase it to 30 minutes just add following line to etc/init.properties file:

'sess-man' () {
    'spam-filter' () {
        'known-spammers' () {
            ban-time = 30
        }
    }
}
</ban-time>

Cache time

Time in minutes for which user will be remembered as a spammer. It will be able to exchange messages with other users (after ban time passes), but if the situation repeats within this time and our algorithm will be sure that user is a spammer - it may disable local user account.

'sess-man' () {
    'spam-filter' () {
        'known-spammers' () {
            cache-time = 10080
        }
    }
}

Disabling account

If filter, depending on other filter reports, will establish that user is for sure a spammer it may not only ban user for some time, but it may disable that user account. This is done by default, if you wish to disable account deactivation add following line to etc/init.properties file:

'sess-man' () {
    'spam-filter' () {
        'known-spammers' () {
            disable-account = false
        }
    }
}

Print list of detected spammers

It is possible to request filter to print full list of known spammer which are currently banned every minute. To do so, you need to set print-spammers property to true.

'sess-man' () {
    'spam-filter' () {
        'known-spammers' () {
            print-spammers = true
        }
    }
}

Frequency of printing list of spammers

By default, list of detected spammers is printed to logs every day. If you wish you can adjust this value to 1 hour, then add following entry to etc/init.properties file:

'sess-man' () {
    'spam-filter' () {
        'known-spammers' () {
            print-spammers-frequency = 60
        }
    }
}

Presence subscription filter

When there is a presence-based SPAM being sent using XMPP server in most cases there is a lot of presence of type subscribe being sent from the single JID. This behavior is annoying and has negative impact on the XMPP server as according to the XMPP specification each presence of type subscribe sent from JID which is not in the users roster causes adding this JID to the user’s roster until user declines subscription request.

Detection is based on counting subscription request being sent from the same bare JID within a period of time.

Below is list of possible settings which may be modified to adjust this filter behaviour.

Number of allowed subscription requests per minute

Be default filter allows 5 subscription requests to be sent from the single JID per minute. If some client will send more than 5 subscription requests it will be marked as a spammer.

Setting filter to allow 7 subscription requests per minute. 

'sess-man' () {
    'spam-filter' () {
        'presence-subcribe' () {
            'limit-per-minute' = 7
        }
    }
}