# IP Score

IP Score stores records in Redis - see here for information about configuring Redis.

Since Rspamd 1.3.4, IP Score requires lookup information from ASN module.

### Theory of operation

IP Score tracks the number of messages received from a given IP/subnet/ASN/country and records this alongside a total score. The scores which are added to these total scores are calculated as follows:

ip_score = action_multiplier * tanh (e * (metric_score/score_divisor))


e is the mathematical constant: 2.718. tanh is the hyperbolic tangent function. metric_score is the score Rspamd assigned the message. action_multiplier is the multiplier configured for the metric action, or zero in case action is no action and score is positive. score_divisor is supplied from setting with the same name- if not supplied no division is done (recommended value: 10-100).

Default multipliers are shown below:

actions {
reject = 1.0;
"rewrite subject" = 0.25;
"no action" = 1.0;
}


So with these settings:

• a message with score -0.1 gets ip score: -0.265
• a message with score -1.0 gets ip score: -0.991
• a message with positive score & no action action always gets ip score: 0.00
• a message with add header action & score 7 gets ip score: 0.249
• a message with reject action and score 15 gets ip score: 1.0

For each IP address/ASN/country/subnet Rspamd stores a key in a hash in Redis the value of which is formatted: total ip score|total number of messages received - for each incoming message Rspamd increments the total number of messages by one and adds the new ip score to the total.

Once a predefined number of messages from a given IP address/subnet/ASN/country have been seen (10 in default configuration), Rspamd will begin to add scores to messages, which are calculated as follows:

First Rspamd calculates a subscore for whichever things it has seen enough messages for (IP address/subnet/ASN/country) as follows:

subscore = score_multiplier * tanh(e * total_score / total_messages)
subscore = floor(subscore * 10)


Score multiplier is dependent on the component the subscore is being generated for; default multipliers are shown below:

scores {
asn = 0.5;
country = 0.1;
ipnet = 0.8;
ip = 1.0;
}


Subscores are added to each other to determine a total. If min_score or max_score are defined in configuration these set a lower/upper bound for the total score.

### Configuration

Refer to example configuration below for available settings. To use default settings, just configure Redis either globally or just for ip_score and assign a weight to the IP_SCORE symbol. Module configuration should be added to /etc/rspamd/local.d/ip_score.conf.

# how each action is treated in scoring
actions {
reject = 1.0;
"rewrite subject" = 0.25;
"no action" = 1.0;
}
# how each component is evaluated
scores {
asn = 0.5;
country = 0.1;
ipnet = 0.8;
ip = 1.0;
}
# prefix for asn hashes
asn_prefix = "a:";
# prefix for country hashes
country_prefix = "c:";
# hash table in redis used for storing scores
hash = "ip_score";
# prefix for subnet hashes
ipnet_prefix = "n:";
# minimum number of messages to be scored
lower_bound = 10;
# the metric to score (usually "default")
metric = "default";
# upper and lower bounds at which to cap total score
#max_score = 10;
#min_score = -5;
# Amount to divide subscores by before applying tanh
score_divisor = 10;
# list of servers (or configure redis globally)
#servers = "localhost";
# symbol to be inserted
symbol = "IP_SCORE";


You will also have to register some weight for the symbol in metric. For example you could add the following to /etc/rspamd/local.d/metrics.conf:

symbol "IP_SCORE" {
weight = 2.0;
description = "IP reputation";
}