Hi Chris and list,
the problems I see with adopting STIX are: 1) it does not quite fit the needs IMHO 2) most users over here (Europe) went a different route. 3) even DHS which was pushing STIX a lot in private mentioned they went MISP 4) (similar to (1)) - IntelMQ is a different beast than a graph of objects all linked to each other. It's for the lower levels of automation. So, again, it does not quite fit the needs 5) Yes, we should be able to import and export STIX (as MISP does), but .. that's a controlled hand over point 6) STIX is too bloated. It allows sooo many different interpretations that it comes to being not clear in what it actually wants to convey. For IntelMQ we need super clear data which can be used to automatically trigger actions (for example on firewalls, in SIEMs/IR etc).
That being said, with the IEP 03 and 04 I see a risk of bloat for IntelMQ as well. More on that in a separate mail.
On 06.04.2021, at 09:47, Chris Horsley chris.horsley@csirtfoundry.com wrote:
Yes, it does require a bit of a different mindset for STIX representation. It's less a single system event = a single serialised record, and more a package of entities which relate to each other in a graph style.
For the example of an open port / exploitable service (using https://docs.oasis-open.org/cti/stix/v2.1/cs01/stix-v2.1-cs01.html for reference):
- Open port / vulnerable service
- Define an `infrastructure` entity to represent the vulnerable host.
- Define a `tool` entity to represent RDP server
- Define a `vulnerability` entity to represent the specific vulnerability (e.g. CVE #)
- Bind them together with references as a directed graph: infrastructure `hosts` tool, tool `has` vulnerability
By design, it's subject to interpretation since STIX is agnostic about the software consuming it.
If the scope of this feature is strictly IntelMQ <-> IntelMQ data exchange for the foreseeable future, you might consider something that more directly serialises IntelMQ records. In that case, there's probably still value looking at the STIX spec for additional concepts and ideas from a thoroughly debated taxonomy.
Best regards,
Chris
On 31/03/2021 9:00 pm, Sebastian Wagner wrote:
Hi Chris,
Thanks for the input.
I looked at STIX/TAXII 2 some while ago and also have been in contact with the creators back then. IIRC and according to what the creators said, STIX/TAXII cannot be used for some/most of the data we are processing in IntelMQ. For example, how do you represent an open port (vulnerable service), an infected device or a malicious website? I don't see any STIX Object listed on the page linked by you, that could match for that kind of data.
kind regards Sebastian
On 3/31/21 5:43 AM, Chris Horsley wrote:
Before going too far down this road, I'd be looking at the suitability or adaptability of STIX / TAXII 2.1 (https://oasis-open.github.io/cti-documentation/stix/intro).
The STIX steering committee has spent years iterating and debating the data model for STIX. They've already done a lot of the hard work on how entities should reference one another, how TLP is implemented, consistent taxonomies, appropriate metadata and so on. There's also Python libs available, so it's more a case of working out how to integrate rather than reinvent.
TAXII provides a HTTP-based transport layer for STIX (or other data formats) which you can operate via push, pull, or otherwise relay via some sort of chained series of TAXII servers.
As a bonus, it would give sharing inter-operation between IntelMQ and other platforms which also implement STIX / TAXII. MISP is one of those (https://www.misp-project.org/2020/06/24/MISP.2.4.128.released.html) so there'd be some good experiences to draw from their developers I feel.
Best regards,
Chris
On 31/03/2021 2:56 am, Sebastian Waldbauer wrote:
Dear IntelMQ Developers and Users,
nowadays security incidents are more important than 10 years ago. As IntelMQ can be used as core element for automated security incident handling, we would like to provide a way to share information with other intelmq instances. This proposal is also an alternative to IEP03 insofar as solving the "multiple values" is possible by using UUIDs so "link" related events in a backwards-compatible manner.
If you're interested, please let us know, so we could organize a hackathon for further discussions about the specification of the meta-information. Previously this idea was discussed in [0] and [1].
[0] https://github.com/certtools/intelmq/blob/version-3.0-ideas/docs/architectur... [1] https://github.com/certtools/intelmq/issues/1521 # IEP04: Internal Data Format: Meta Information and Data Exchange To ease data exchange between two or more IntelMQ instances, adding some meta-information to the events can make this sharing easier in certain regards. "Linking" events could be based on the same theory as `git` using it - with parent hashes ( we would call it UUID ).
### TL;DR Communication between one or more IntelMQ instances & exchange data with a backwards-compatible format. P2P or centralized architecture is a big topic, which has to be discussed after the format is being set.
### Why is metadata important? Short and simple. To avoid race conditions & being able to discard/drop already processed events from other instances.
### Meta information Metadata is used to transfer some general data, which is not likely related to the event itself. It's more or less just an information to keep events clear & sortable.
A message could look like:
{ "meta": { "version": 1, # protocol version, so we are allowed to fallback to old versions too "uuid": { current: "cert_at:aaaa-bbbb-cccc-dddd" # format to be decided parent: "cert_at:xxxx-yyyy-zzzz-ffff" # format to be discussed, if not set -> current is the parent uuid }, "type": "event", "format": "intelmq", # i. e. this field could contain "n6" or "idea", so the receiving component can decode on demand. }, "payload": { # normal intelmq data "source.ip": "127.0.0.1", "source.fqdn": "example.com", "raw": base64-blob } }
Tell us your opinion about adding non-standardized meta-information fields ( i. e. RTIR ticket number, origin, other local contact informationen ... and so on )
#### The UUID For the UUID there are multiple options:
- Generate a random 128 bit UUID
- A list of entities, which dealt with this event already. For example if an event was passed on from cert-at to cert-ee, the field could look like `!cert-at!cert-ee`. A message sending loop can be detected if the own name is already in this field upon reception.
- Using CyCat: `publisher-short-name:project-short-name:UUID`. For example: `cert-at:intelmq:72ddb00c-2d0a-4eea-b7ac-ae122b8e6c3b`, or `cert-pl:n6:f60c9fb9-81f9-4e0b-8a44-ea41326a15b3`. Some more research and discussion is required before the implementation of this option. Have a look at https://www.cycat.org/services/concept/ for more details.
- A hash: A benefit using a hash is that we're able to recalculate them on every intelmq instance.
### Exporting events to other systems In IntelMQ 2.x the events only comprise of the "payload" and no meta information. For local storages like file output or databases, the meta information may not be relevant in some use-cases. So it needs to be possible to export events *without* meta information, which is also the backwards-compatible behaviour.
The "type" field exists in the current format as "__type" in the flat payload structure. In the output bots there's currently a boolean parameter `message_with_type` to include the field `__type` in the "export". For optionally exporting meta-information like uuid or format, a similar logic could be used.
### How can data exchange work? This now depends on how IntelMQ instances can communicate, either Peer-to-peer or via a central data hub. Both of them do have pro's and con's.
#### P2P ( Peer 2 Peer ) Decentralized network
- Less downtimes: A downtime of one instance, does not affect the whole network
- Better privacy: data is not shared to an unrelated instance
- More secure: data can optionally be encrypted (key-exchange between instances?)
- Decentralized and local maintenance
~ Network latency depends on server locations
- Networking issues may occur
How would data exchange looks like between two instances:
- Instance A has events which should be relayed to Instance B & C, because they're not sure who the actually receiver should be
- Instance A ensures all messages have a UUID
- Instance A sends the data to Instance B & Instance C
- Instance B checks the data & they're sure that the data should be for Instance C
- Instance C receives data from Instance A & Instance B
- Instance C checks the UUID, which is the same & drops the package from Instance B
#### (Central) Data hub
- Less maintenance: Is maintained by the hub administrator
- Central data storage (reports can optionally be cached to be downloaded later)
~ Central data analysis (e.g. statistics) is possible ~ Network latency depends on server locations
- point of failure: if network problems occur, no exchange is possible
As already seen above, data exchange here would be less complicated. The sending may look like:
- Instance A has events which should be relayed to Instance B (e.g. different country)
- Instance A ensures all messages have a UUID
- Instance A sends these messages to the data hub
The reception side can look like:
- Instance B connects to central instance
- Instance B queries and downloads all available messages
- Upon reception, all messages are de-duplicated based on the UUID:
a) If the UUID is already known, discard the message b) If the UUID has not been seen before, continue with processing
To sum up, both exchange variants are useful. More research is needed, i. e. a mixed infrastructure with centralized parts but can be decentralized too. However, this shall not be neither the purpose nor the aim of this IEP.
-- // Sebastian Wagner wagner@cert.at
- T: +43 1 5056416 7201
// CERT Austria - https://www.cert.at/
// Eine Initiative der nic.at GmbH - https://www.nic.at/
// Firmenbuchnummer 172568b, LG Salzburg
IntelMQ-dev mailing list https://lists.cert.at/cgi-bin/mailman/listinfo/intelmq-dev https://intelmq.readthedocs.io/