.. _datasets:

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;

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:

.. container:: example-rule

alert dns any any -> any any (msg:"dns list test"; dns.query; to_sha256; dataset:isset,dns-sha256-seen; sid:123; rev:1;)

.. container:: example-rule

alert http any any -> any any (msg: "http user-agent test"; http.user_agent; dataset:set,ua-seen; sid:234; rev:1;)

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.

Rule keywords

dataset


Datasets are binary: something is in the set or it's not.

Syntax::

    dataset:<cmd>,<name>,<options>;

    dataset:<set|unset|isset|isnotset>,<name> \
        [, type <string|md5|sha256|ipv4|ip>, save <file name>, load <file name>, state <file name>, memcap <size>, hashsize <size>
         , format <csv|json|ndjson>, context_key <output_key>, value_key <json_key>, array_key <json_path>,
         remove_key];

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
format <type>
  the format of the file: csv, json. Defaut to csv. See
  :ref:`dataset with json format <datasets_json>` for json
  and ndjson option
context_key <key>
  the key to use for the enrichment of the alert event
  for json format
value_key <key>
  the key to use for the value of the alert
  for json format
array_key <key>
  the key to use for the array of the alert
  for json format
remove_key
  if set, the JSON object pointed by value key will be removed
  from the alert event


.. note:: 'type' is mandatory and needs to be set.

.. note:: 'load' and 'state' or 'save' and 'state' cannot be mixed.

Example rules could look like:

1. Detect unique User-Agents:

.. container:: example-rule

    alert http any any -> any any (msg:"LOCAL HTTP new UA"; http.user_agent; dataset:set,http-ua-seen, type string, state http-ua-seen.csv; sid:8000001; rev:1;)

2. Detect unique TLDs:

.. container:: example-rule

    alert dns $HOME_NET any -> any any (msg:"LOCAL DNS unique TLD"; dns.query; pcrexform:"\\.([^\\.]+)$"; dataset:set,dns-tld-seen, type string, state dns-tld-seen.csv; sid:8000002; rev:1;)

Following image is a pictorial representation of how the ``pcrexform`` works
on domain names to find TLDs in the dataset ``dns-tld-seen``:

.. image:: dataset-examples/detect-unique-tlds.png

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``.

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.

.. _datasets_json:

dataset with JSON


Dataset with JSON allows matching data against a set and output data attached to the matching
value in the event.

There are two formats supported: ``json`` and ``ndjson``. The difference is that
``json`` format is a single JSON object, while ``ndjson`` is handling file with
one JSON object per line. The ``ndjson`` format is useful for large files
as the parsing is done line by line.

Syntax::

    dataset:<cmd>,<name>,<options>;

    dataset:<isset|isnotset>,<name> \
        [, type <string|md5|sha256|ipv4|ip>, load <file name>, format <json|ndjson>, memcap <size>, hashsize <size>, context_key <json_key> \
         , value_key <json_key>, array_key <json_path>];

Example rules could look like::

    alert http any any -> any any (msg:"IP match"; ip.dst; dataset:isset,bad_ips, type ip, load bad_ips.json, format json, context_key bad_ones, value_key ip; sid:8000001;)

In this example, the match will occur if the destination IP is in the set and the
alert will have an ``alert.content.bad_ones`` subobject that will contain the JSON
data associated to the value (``bad_ones`` coming from ``context_key`` option).

When format is ``json`` or ``ndjson``, the ``value_key`` is used to get
the value in the line (``ndjson`` format) or in the array (``json`` format).
At least one single element needs to have the ``value_key`` present in the data file to
have a successful load.
If ``array_key`` is present, Suricata will extract the corresponding subobject that has to be
a JSON array and search for element to add to the set in this array. This is only valid for ``json`` format.

If you don't want to have the ``value_key`` in the alert, you can use the
``remove_key`` option. This will remove the key from the alert event.

See :ref:`Dataset with JSON format <datajson_data>` for more information.

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.


Unix Socket
-----------

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==

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)

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

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==

dataset-dump
~~~~~~~~~~~~

Unix socket command to trigger a dump of datasets to disk.

Syntax::

    dataset-dump

dataset-add-json
~~~~~~~~~~~~~~~~

Unix Socket command to add data to a set. On success, the addition becomes
active instantly.

Syntax::

    dataset-add-json <set name> <set type> <data> <json_info>

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-json myset string Z29vZ2xlLmNvbQ== {"city":"Mountain View"}


File formats
------------

Datasets use a simple CSV format where data is per line in the file.

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)


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; )


datarep
~~~~~~~

The datarep format follows the dataset, except that there are 1 more CSV
field:

Syntax::

    <data>,<value>

.. _datajson_data:

dataset with JSON enrichment

If format json is used in the parameters of a dataset keyword, then the loaded file has to contain a valid JSON object.

If value_key``` option is present then the file has to contain a valid JSON object containing an array where the key equal to value_key`` value is present.

For example, if the file file.json is like the following example (typical of return of REST API call) ::

{
    "time": "2024-12-21",
    "response": {
        "threats":
            [
                {"host": "toto.com", "origin": "japan"},
                {"host": "grenouille.com", "origin": "french"}
            ]
    }
}

then the match to check the list of threats using dataset with JSON can be defined as ::

http.host; dataset:isset,threats,load file.json, context_key threat, value_key host, array_key response.threats;

If the signature matches, it will result in an alert with the following ::

{
    "alert": {
        "context": {
            "threat": {
                "host": "toto.com",
                "origin": "japan"
            }
        }
    }
}

.. _datasets_file_locations:

File Locations

Dataset filenames configured in the suricata.yaml can exist anywhere on your filesytem.

When a dataset filename is specified in rule, the following rules are applied:

For load, the filename is opened relative to the rule file containing the rule. Absolute filenames and parent directory traversals are allowed.
For save and state the filename is relative to $LOCALSTATEDIR/suricata/data. On many installs this will be /var/lib/suricata/data, but run suricata --build-info and check the value of --localstatedir to verify this location onn your installation.
- Absolute filenames, or filenames containing parent directory traversal (..) are not allowed unless the configuration paramater datasets.allow-absolute-filenames is set to true.

.. _datasets_security:

Security

As datasets potentially allow a rule distributor write access to your system with save and state dataset rules, the locations allowed are strict by default, however there are two dataset options to tune the security of rules utilizing dataset filenames::

datasets: rules: # Set to true to allow absolute filenames and filenames that use # ".." components to reference parent directories in rules that specify # their filenames. allow-absolute-filenames: false

  # Allow datasets in rules write access for "save" and
  # "state". This is enabled by default, however write access is
  # limited to the data directory.
  allow-write: true

By setting datasets.rules.allow-write to false, all save and state rules will fail to load. This option is enabled by default to preserve compatiblity with previous 6.0 Suricata releases, however may change in a future major release.

Pre-Suricata 6.0.13 behavior can be restored by setting datasets.rules.allow-absolute-filenames to true, however allowing so will allow any rule to overwrite any file on your system that Suricata has write access to.