6.41. Datasets¶
Using the dataset
and datarep
keyword it is possible to match on
large amounts of data against any sticky buffer.
For example, to match against a DNS black list called dns-bl
:
dns.query; dataset:isset,dns-bl;
These keywords are aware of transforms. So to look up a DNS query against a MD5 black list:
dns.query; to_md5; dataset:isset,dns-bl;
6.41.1. Global config (optional)¶
Datasets can optionally be defined in the main config. Sets can also be declared from the rule syntax.
Example of sets for tracking unique values:
datasets:
ua-seen:
type: string
state: ua-seen.lst
dns-sha256-seen:
type: sha256
state: dns-sha256-seen.lst
Rules to go with the above:
It is also possible to optionally define global default memcap and hashsize.
Example:
datasets:
defaults:
memcap: 100mb
hashsize: 2048
ua-seen:
type: string
load: ua-seen.lst
or define memcap and hashsize per dataset.
Example:
datasets:
ua-seen:
type: string
load: ua-seen.lst
memcap: 10mb
hashsize: 1024
Note
The hashsize should be close to the amount of entries in the dataset to avoid collisions. If it's set too low, this could result in rather long startup time.
6.41.2. Rule keywords¶
6.41.2.1. dataset¶
Datasets are binary: something is in the set or it's not.
Syntax:
dataset:<cmd>,<name>,<options>;
dataset:<set|isset|isnotset>,<name> \
[, type <string|md5|sha256|ipv4|ip>, save <file name>, load <file name>, state <file name>, memcap <size>, hashsize <size>];
- type <type>
- the data type: string, md5, sha256, ipv4, ip
- load <file name>
- file name for load the data when Suricata starts up
- state
- sets file name for loading and saving a dataset
- save <file name>
- advanced option to set the file name for saving the in-memory data when Suricata exits.
- memcap <size>
- maximum memory limit for the respective dataset
- hashsize <size>
- allowed size of the hash for the respective dataset
Note
'load' and 'state' or 'save' and 'state' cannot be mixed.
Example rules could look like:
- Detect unique User-Agents:
- Detect unique TLDs:
Following image is a pictorial representation of how the pcrexform
works
on domain names to find TLDs in the dataset dns-tld-seen
:

Notice how it is not possible to do certain operations alone with datasets
(example 2 above), but, it is possible to use a combination of other rule
keywords. Keep in mind the cost of additional keywords though e.g. in the
second example rule above, negative performance impact can be expected due
to pcrexform
.
6.41.2.2. datarep¶
Data Reputation allows matching data against a reputation list.
Syntax:
datarep:<name>,<operator>,<value>, \
[, load <file name>, type <string|md5|sha256|ipv4|ip>, memcap <size>, hashsize <size>];
Example rules could look like:
alert dns any any -> any any (dns.query; to_md5; datarep:dns_md5, >, 200, load dns_md5.rep, type md5, memcap 100mb, hashsize 2048; sid:1;)
alert dns any any -> any any (dns.query; to_sha256; datarep:dns_sha256, >, 200, load dns_sha256.rep, type sha256; sid:2;)
alert dns any any -> any any (dns.query; datarep:dns_string, >, 200, load dns_string.rep, type string; sid:3;)
In these examples the DNS query string is checked against three different reputation lists. A MD5 list, a SHA256 list, and a raw string (buffer) list. The rules will only match if the data is in the list and the reputation value is higher than 200.
6.41.3. Rule Reloads¶
Sets that are defined in the yaml, or sets that only use state or save, are considered dynamic sets. These are not reloaded during rule reloads.
Sets that are defined in rules using only load are considered static tests. These are not expected to change during runtime. During rule reloads these are reloaded from disk. This reload is effective when the complete rule reload process is complete.
6.41.4. Unix Socket¶
6.41.4.1. dataset-add¶
Unix Socket command to add data to a set. On success, the addition becomes active instantly.
Syntax:
dataset-add <set name> <set type> <data>
- set name
- Name of an already defined dataset
- type
- Data type: string, md5, sha256, ipv4, ip
- data
- Data to add in serialized form (base64 for string, hex notation for md5/sha256, string representation for ipv4/ip)
Example adding 'google.com' to set 'myset':
dataset-add myset string Z29vZ2xlLmNvbQ==
6.41.4.2. dataset-remove¶
Unix Socket command to remove data from a set. On success, the removal becomes active instantly.
Syntax:
dataset-remove <set name> <set type> <data>
- set name
- Name of an already defined dataset
- type
- Data type: string, md5, sha256, ipv4, ip
- data
- Data to remove in serialized form (base64 for string, hex notation for md5/sha256, string representation for ipv4/ip)
6.41.4.3. dataset-clear¶
Unix Socket command to remove all data from a set. On success, the removal becomes active instantly.
Syntax:
dataset-clear <set name> <set type>
- set name
- Name of an already defined dataset
- type
- Data type: string, md5, sha256, ipv4, ip
6.41.4.4. dataset-lookup¶
Unix Socket command to test if data is in a set.
Syntax:
dataset-lookup <set name> <set type> <data>
- set name
- Name of an already defined dataset
- type
- Data type: string, md5, sha256, ipv4, ip
- data
- Data to test in serialized form (base64 for string, hex notation for md5/sha256, string notation for ipv4/ip)
Example testing if 'google.com' is in the set 'myset':
dataset-lookup myset string Z29vZ2xlLmNvbQ==
6.41.4.5. dataset-dump¶
Unix socket command to trigger a dump of datasets to disk.
Syntax:
dataset-dump
6.41.5. File formats¶
Datasets use a simple CSV format where data is per line in the file.
6.41.5.1. data types¶
- string
- in the file as base64 encoded string
- md5
- in the file as hex encoded string
- sha256
- in the file as hex encoded string
- ipv4
- in the file as string
- ip
- in the file as string, it can be IPv6 or IPv4 address (standard notation or IPv4 in IPv6 one)
6.41.5.2. dataset¶
Datasets have a simple structure, where there is one piece of data per line in the file.
Syntax:
<data>
e.g. for ua-seen with type string:
TW96aWxsYS80LjAgKGNvbXBhdGlibGU7ICk=
which when piped to base64 -d
reveals its value:
Mozilla/4.0 (compatible; )
6.41.5.3. datarep¶
The datarep format follows the dataset, expect that there are 1 more CSV field:
Syntax:
<data>,<value>