HI Tim

I will contact you off list and hope we can review this together. I will put my full idea to you and see what you think of the whole plan. In the mean time I have added a few comments below.

cheers
denis

On 16/05/2015 16:46, Tim Bruijnzeels wrote:
Hi Denis and all,

On 15 May 2015, at 18:34, denis <ripedenis@yahoo.co.uk> wrote:

Hi Tim and All

Personalised authorisation is an idea I developed over the last few years. I talked to many people in the community about it at various RIPE meetings and started to build up support for my ideas.

I believe that the WG appreciates your efforts on this, and remembers your presentation at RIPE 68:
https://ripe68.ripe.net/presentations/299-DB_WG_personalised_Auth_RIPE_68.pdf


The basic idea was to allow authorisation tokens in PERSON objects,

Yes, the important point here is that the credentials are on PERSONs, rather than in one anonymous blob that is today's MNTNER.

Agreed, but in your proposal you are missing several key points about how these credentials can be used. I agree on having the option for PERSON objects to be self maintaining, but disagree that you should not allow an organisation to manage the collection of PERSON objects. There are many reasons why an organisation does this. They may want to not allow anyone to use a password and require all staff to use SSO, for example. By maintaining all the PERSON objects they are in control of that. They may want to be able to delete the PERSON object when the person leaves the company. They may want notifications to be centralised. The PERSON objects may only contain corporate information instead of actual personal data to reflect the persons corporate identity.

There was also a major oversight in the original implementation of RPSL. Allowing PERSON object to be directly referenced anywhere has caused so many problems to so many organisation over the last 15 years and continues to do so today. Here is an opportunity to start fixing that as we move forward. By only allowing PERSON objects with "auth:" to be referenced in a ROLE object and then using the ROLE as the maintaining object, not only do we make it more intuitive, but we decouple the direct references to PERSON objects throughout the rest of the database. As time goes on that will prove to be a massive plus.


group these into ROLE objects and use the ROLE 'instead of' a MNTNER. This is much more intuitive and better reflects real life business operations. The MNTNER object is an abstract construct that many people simply don't understand. The long term goal was to (possibly) eventually deprecate MNTNER objects.

There are different opinions on how to refer to authorised persons.

You said it yourself that the first 3 hours of a DB training course are spent explaining how to create a PERSON and MNTNER object. That tells a story. If you ask someone who maintains their data they say "I do" or "We do". "I" is a PERSON object. "We" is a ROLE object. This is how people naturally think. Many people simply don't understand what a MNTNER is. It is not even a pronounceable word to a native English speaker. It is abstract and non intuitive. I agree technically there is no difference between using a MNTNER or ROLE object to maintain data. It is just semantics. But there is a world of difference in perception and intuitiveness. The concept of a group of people maintaining data naturally fits a role. I accept we could achieve the same result if we simply renamed the MNTNER as something like AUTH-ROLE and it becomes a special case of a ROLE object. But long term I still favour combining both MNTNER and IRT with ROLE and make this one of the most powerful objects used to manage your data in the DB. (Then people will start to understand why "abuse-c:" was implemented the way it is.)


The idea to use ROLEs instead of MNTNERs was presented again at RIPE 69, along side with the idea of just allowing to refer to PERSONs from MNTNERs: https://ripe69.ripe.net/presentations/125-ripe69-db-wg-pers-auth.pdf

There was no support from the room, nor in informal discussions with working group members, for the option of using ROLEs instead of MNTNERs. While the basic idea sounds attractive there are a lot of problems on closer inspection. MNTNER objects differ from ROLEs in a number of ways that make this, and the ultimate deprecation of MNTNERs difficult. Slides 13 lists what is missing from ROLEs, and would be needed to use them in an mnt-* context. Slide 14 lists what is missing from MNTNERs that would have to be made up, or made optional possibly with business rules enforcing behaviour (e.g. address may still be needed for a *-c referenced role), if remaining MNTNERs were to be converted into ROLEs. And if the latter isn't done, then we would have to live with mnt-* being allowed to refer to either a MNTNER or a ROLE (with special attributes turned on), for a long time, and this is hardly intuitive.

I think you need to present a full plan rather that just suggest partial ideas. Just asking if we should replace MNTNER with ROLE does not in itself sound like an inspiring move. My proposal to allow this in a parallel track means you can either use the simpler, intuitive method, or stick with the old MNTNER method. I think once people realise the benefits they would move over.


There was however support for the basic concept of extending MNTNERs with personalised organisation in a backward compatible way that requires no action from any of the over 50,000 maintainers in the database. The RIPE NCC was tasked with working out and presenting a new plan based on this resulting in the presentation given at RIPE70: https://ripe70.ripe.net/presentations/165-ripe70-pers-auth.pdf


Trying to feed personalised auth into objects via MNTNERs, even worse through ROLEs and MNTNERs, is not only adding extra, unnecessary, layers of abstraction but making it even less intuitive and totally unrelated to real life situations.

In this proposal MNTNER objects remain the specialised security objects that they are today, including features lacking from the normal contact-oriented ROLE object, but personalised authorisation is added with minimal changes to the schema to allow those users that want to make use of this to do so, without forcing any existing maintainer to be modified.

My proposal also allows anyone to continue to do what they do now and ignore all the changes.


This is low hanging fruit.

Referring from an object to a MNTNER, and from that MNTNER to a number of authorised PERSONs does not add any layers compared to using a ROLE there instead of the MNTNER. And note that I did not favour using ROLEs in between MNTNERs and authorised PERSONs for this very reason, in response to your comment through chat during the WG session that this should be allowed.


My original idea was to simplify the auth model and bring it closer to reality, adding extra, beneficial features, without losing any of the operational features currently available through MNTNERs....but without the need to use MNTNERs. Everything can be done with PERSON and ROLE objects...which people understand.

I honestly believe that the now proposed model does all this, the one thing people need to understand is that a MNTNER despite the name 'maintainer' being singular would be allowed to make explicit references to different PERSONs who do the actual maintaining.

The MNTNER is like a ROLE for security context.

The MNTNER has extra things important to security like:
 - where do the alerts go?
 - where do the notifications go?

And it's lacking other things relevant to people looking to contact a group of people, but irrelevant when authorising a group of people. Such as: address, email, phone and fax.


I had this all worked out in my head how to achieve all this, which is not technically very difficult to implement and not hard to understand and can be done in parallel with current MNTNER operation (so no one has to change if they don't want to). But I never wrote any of this down or presented any detail to anyone. So I would like to present an alternative option to the community, along the lines I was thinking and had discussed briefly with many people. It may take me a week or so to write it all out and present it as a RIPE Labs article.

cheers
Denis Walker
Independent Netizen

On 15/05/2015 10:27, Tim Bruijnzeels wrote:
Dear working group,

Yesterday during the WG session we presented a proposal for implementing personalised authorisation:

As recorded in the first cut of the minutes:
D. Personalised authentication (Tim Bruijnzeels, RIPE NCC)
  (See presentation)
  This will allow one click creation of person objects
  Maintain credentials in one place.
  Allow better auditing.
  Done by extending person object to have multiple optional auth: attribute
  This will ultimately allow existing auth: sso references to be cleaned up
  Last auth: attribute should not be removed from a person object that is used in an authorisation context.

Apart from questions about possible additions below, there seemed to be general approval for the above as an addition to the existing maintainer mechanism.

We would very much like to implement this soon. We are already working on improving the way users can log in and use the web updates, and manage maintainers (and who is authorised for them), so having this would be extremely useful for that effort.

Technically I don't think the above has to depend on further extensions below. Roles can be added at any time that we consensus on them, and showing audit logs is a separate effort - building on this.

  Should this be extended to the role object as well? This would involve additional business rules but is technically possible.

I understand and fully agree that there is a need to maintain a list of authorised persons centrally. But in effect a maintainer can be used for this purpose. Multiple objects can be maintained by the same maintainer, and the list of persons authorised can then be managed on this single maintainer:

obj1    ---\
            ---> mnt1  --->  pers1
obj2    ---/           \-->  pers2
 

In other words, just like role objects can group persons in a 'contact' context, 'maintainers' could group persons in a 'authorisation' context, where also other things such as "upd-to:" etc can find a home.

So, technically I don't think there is a need to have another role object here:

obj1    ---\
            ---> mnt1  --->  role1  ---> pers1
obj2    ---/                        \--> pers2

Conceptually this can work of course, but it adds some complexity, and things to resolve:

a) referencing roles from maintainers, and authorised persons from roles

The proposal was to refer to authorised persons from maintainers like this:   auth:  person-<nichandle>

Can we resolve this by allowing:
  = auth:  role-<nichdl> on maintainers
  = auth:  person-<nichdl> on roles 

But no other auth: flavours for now.

Also note that this person is not necessarily an authorisation *contact* for others. If we follow current practice consistently we would filter this value for security purpose.

b) business rules regarding auth->role

Suggestion:
- A role can only be added to a maintainer as "auth: role-<nichdl>" if it has at least one "auth: person-<nichdl>"
- The last "auth: person" can not be removed from a role if it's referenced anywhere as "auth: role-"
- As before: "auth: person-<nichl>" can only be added if the person has at least one "auth: <something>"
- As before: the last "auth:" can not be removed from a person if it's referenced anywhere as "auth: person-"

  It would be useful to record what credential (maintainer) was used to make a particular change to an object and this change
  would facilitate this. RV was asked to raise this on the mailing list.

Currently we do know internally which maintainer was used to submit a successful update, but not which credential. Technically this could be added of course. And in case of SSO or PGP people can get some idea of which user did the update. But showing which password hash was used for an update may not be best security practice.

With authorisation delegated to persons (possibly through roles) we will be able to give a much more better output. We can refer to the name of the person, rather than a credential that should be private to that person.

Also note that for any of this we will also need to be sure that the user viewing this information is authorised to see this. So what we had in mind here is to show this only on the web interface for logged in users authorised for at least one mnt-by of the object they are looking at.