OASIS Mailing List ArchivesView the OASIS mailing list archive below
or browse/search using MarkMail.


Help: OASIS Mailing Lists Help | MarkMail Help

cti-stix message

[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]

Subject: Re: [EXT] [cti-stix] ability to use UUID5 in STIX2 identifier

I've been thinking a lot about this and I think it makes sense.

One of the concerns we had at the time we chose UUID4 is that users of libraries like python-stix would need to remember to set the UUID5 namespace -- or, if they don't and python-stix has some default namespace, different tools using the libraries could have overlapping IDs. This would also apply to users of the new Java libraries that I've seen come out. It might mean these libraries requiring that people set a unique namespace before creating any objects, vs. now where it can just go ahead and create IDs by default. I'd be curious what other people think about this problem and how we can help avoid it becoming an issue (especially given how many people use those libraries).


ïOn 2/4/19, 11:12 AM, "cti-stix@lists.oasis-open.org on behalf of Sergey Polzunov" <cti-stix@lists.oasis-open.org on behalf of sergey@eclecticiq.com> wrote:

    Hey everybody!
    Current STIX2 spec definition of an`identifier` for STIX2 objects is as follows:
    > An identifier universally and uniquely identifies a SDO, SRO, Bundle, or Marking Definition. Identifiers MUST follow the form object-type--UUIDv4, where object-type is the exact value (all type names are lowercase strings, by definition) from the type property of the object being identified or referenced and where the UUIDv4 is an RFC 4122-compliant Version 4 UUID. The UUID MUST be generated according to the algorithm(s) defined in RFC 4122, section 4.4 (Version 4 UUID) [RFC4122].
    â from http://docs.oasis-open.org/cti/stix/v2.0/cs01/part1-stix-core/stix-v2.0-cs01-part1-stix-core.html#_Toc496709265
    I think the requirement to have UUID4 brings more problems than benefits. It makes STIX1->STIX2 transition difficult, hurting existing STIX1 users.
    I will try to show it in these 2 use cases.
    Use case 1
        Imagine that I'm a client of an intelligence provider A. I've been a client for a long time and I have received intelligence in STIX1.2, which I stored in my DB. I fetch new intelligence daily, downloading only fresh data. Often fresh data links to old objects for context.
        Provider A decides to upgrade and switch to STIX2. In addition to an old STIX1.2 feed, provider creates new STIX2 feed with the same data. In STIX2 all objects have new identifiers and Provider A does not bother to supply a mapping of STIX1.2 ids to STIX2 ids. Now I, as a client, have 2 options:
        - clean slate option: drop all old data from this provider and re-fetch everything. That will work if Provider A is the only provider I use or if I never referenced Provider A's data from my own intelligence. Not a great plan.
        - new era option: leave my STIX1.2 data graph in place and start consuming new STIX2 feed from today. This option has one big issue: new STIX2 data will not be connected to STIX1.2 data I already have, because STIX2 ids are all different. If I want to deduce connection, I need to deduplicate the data against my existing STIX1.2 DB. This means my ingestion pipeline must be smart enough to compare STIX1.2 objects to STIX2 objects and be fast enough to do that for every new STIX2 object. This will be difficult to implement and will have a huge performance penalty.
    Use case 2
        Imagine that I'm a NCSC. I receive intelligence from providers, combine it and distribute it to my clients. My providers are still on STIX1.2 but my clients want STIX2, so I must convert STIX1.2 I receive into STIX2. Full STIX1.2 entities I can transform easily but what do I do with IDREFs I have in my STIX1.2 data?
        I can generate new STIX2 id every time I see new STIX1.2 IDREF in incoming data and store STIX1.2->STIX2 mapping somewhere to be used next time I see this IDREF. This is painful and will require additional resources, but it is doable. But it will only work until the moment my providers switch to STIX2 and start sending me full objects for those IDREFs with new random STIX2 identifiers! I can not predict these identifiers and I can't match them with the ones I generated. So my thinking is - what is the point in even bothering with old IDREFs? I will just drop them, sending my clients sometimes disconnected STIX2 entities, hoping that they will figure it out.
    Proposed solutions if UUID5 is allowed in STIX2 identifiers:
    Use case 1 solution
        There can be a guideline that will recommend providers to use old STIX1.2 IDs as input for new STIX2 identifiers. If STIX2 identifiers are predictable I, as a client, can greatly simplify my deduplication logic. I can run DB migration once to calculate STIX2 identifiers for all my STIX1.2 objects and use these on ingestion for deduplication. Appending STIX2 data to my STIX1.2 DB will be much easier.
        I'm also interested in pushing Provider A to adopting this STIX2 identifier generation practice because it will save me money.
    Use case 2 solution
    WIth UUID5 I have a way out: I can generate new STIX2 ids from old STIX1.2 ids! I can parse IDREF value, that looks like `[ns prefix]:[construct type]-[GUID]`, and use provider's namespace / construct type to build new STIX2 identifier. The logic will be like this:
        - full IDREF will be input for UUID5 function
        - for STIX1.2 types that were split (like TTP), I do not know exact STIX2 type Provider would use for old TTP. My solution here would be to play safe and create relations for all possible types: for IDREF to TTP, I will create 4 relations: one to a possible Tool object, one to Malware, one to Attack Pattern and one to Identity. It is an overhead but it is a small price for keeping interconnected intelligence graph.
        Again, when time comes and my providers move to STIX2, I'm interested in pushing them to adopt this id generation schema for old objects, because it will save me, as NCSC, money.
    To reiterate, I would like to propose:
    - a change in STIX2 spec to allow both UUID5 and UUID4 to be used in an identifier of SDO, SRO, MarkingDefinition and Custom Object entities;
    - creating a guideline, complimentary to the spec, that would explain how STIX1.2 ids can be transformed into STIX2 for easier transition.
    UUID5 ids require use of a namespace. UUID5 RFC (https://tools.ietf.org/html/rfc4122#section-4.3) defines some generic namespaces (https://tools.ietf.org/html/rfc4122#appendix-C) but does not prohibits the use of custom ones. I suggest this algorithm:
        - namespace UUID5 is generated by using predefined `NameSpace_URL` namespace and producer's URL;
        - for old objects, GUID part of STIX2 identifier is namespaced UUID5 generated from old STIX1.2 id
        - for new objects, GUID part of STIX2 identifier is either namespaced UUID5 with random UUID4 string, or just random UUID4.
    Example python code for generating UUID5 with custom namespace:
        In [1]: import uuid
           ...: stix12_id = 'eclecticiq:threat-actor-07fa8672-4bca-46e1-a60f-023882b4a473'
           ...: namespace_uuid = uuid.uuid5(uuid.NAMESPACE_URL, 'https://eclecticiq.com/ns')
           ...: stix2_uuid = uuid.uuid5(namespace_uuid, stix12_id)
           ...: stix2_id = 'threat-actor--{}'.format(stix2_uuid)
           ...: print("new STIX2 id: {}".format(stix2_id))
        new STIX2 id: threat-actor--adee573a-12e9-5dd3-958b-0040d32c6b3e
    BONUS: python functions to convert STIX1.2 IDREFs into STIX2 identifiers - https://gist.github.com/traut/fd4b9b8de3c2aa0e161d68c4099656e5
    Thank you,
    Sergey Polzunov

[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]