[Tf-aai] LoA Questionaire

[Martin Matthiesen]

Hi Sander,

Thanks for your comments, the interview is next week and this is an attempt to get answers on a broader basis. It is by the way not a closed one, I am sure I can arrange for one or two to join in.
I'll answer below (knowing that email is not the greatest tool here, let's maybe set up a meeting, if this thread grows quickly). I kept all text for context.

----- Original Message -----
> From: "Sander Maijers" <sander at clarin.eu>
> To: "Martin" <martin.matthiesen at csc.fi>
> Cc: "tf-aai" <tf-aai at lists.clarin.eu>
> Sent: Thursday, 17 September, 2015 11:51:21
> Subject: Re: [Tf-aai] LoA Questionaire

> Hi Martin,
> 
> I found the survey interesting and so I have some comments. Most of my
> comments are inspired by thinking about what the SPF should imply are
> our demands/requirements by giving specific answers. Also, most
> comments hint at possible lack of evidence to support certain answers,

I hope to be able to substantiate below.

> which is I think critical if we want to more or less ‘petition’ for
> any policy changes on the identity federation/IdP side.

In General I feel LoA is on a different planet when we still are fighting to get some/any attributes in the first place. HZSK is still fighting to get an IdP up. But this issue is certainly relevant the further AAI goes along.
 
> Best,
> Sander
> --
> Sent as system administrator and engineer for CLARIN
> Centre Registry & Service Provider Federation @ {centres,infra}.clarin.eu;
> software engineering tools @ {svn,trac}.clarin.eu;
> identity and access management @ {user,idp}.clarin.eu
> usage statistics and service monitoring @ stats.clarin.eu
> 
> Max Planck Institute for Psycholinguistics, software developer
> personal Skype: sander.maijers | work address: Wundtlaan 1, 6525 XD,
> Nijmegen (NL)
> 
> 
> 
> 
> On Wed, Sep 16, 2015 at 3:33 PM, Martin Matthiesen
> <martin.matthiesen at csc.fi> wrote:
>> Hello Taskforce,
>>
>> I was asked by Mikael Linden to participate in an interview to find out Clarin's
>> needs with regard to Level of Assurance. The interview will likely happen next
>> week and I would like to share with you how I am intending to answer. The
>> general line is that LoA is important or at least will be important. As to
>> auditing I see that self-assertion can work a long way, similar to the Data
>> Seal of Approval process.
>>
>> The original can be viewed here:
>> https://wiki.geant.org/display/AARC/Level+of+Assurance+survey+for+SP+communities
>>
>> Cheers,
>> Martin
>>
>>> 2. Questions on the research infrastructures/communities
>>>
>>> Who are your end users (who need to log in to your services):
>>>
>>>     researchers with a Home Organisation (that operates or potentially operates an
>>>     IdP)?
>>
>> yes
>>
>>>     citizen scientists?
>>
>> potentially.
> 
> We have a good amount of private accounts in the LDAP directory for
> the CLARIN IdP. This seems like a normal phenomenon (they just won't
> have the ‘academic’ entitlement)). So for this group the answer can be
> considered ‘yes’, at least, it is not founded in fact to discriminate
> (in the technical sense of the word) them with an ambiguous
> ‘potentially’.

Thanks for that, a clear "yes" it is. I was not quite aware about the user base at the Clarin IdP.

> 
>>
>>>     students with a Home Organisation (that operates or potentially operates an
>>>     IdP)?
>>
>> yes.
>>
>>>     else/what?
>>>
>>> 3.Questions on Identity and Authentication
>>>
>>> User's "network identity" distinguishes him/her from other users of the SP.
>>> 3.1. Identity concept
>>>
>>> How important is it for you that
>>>
>>>     all user identities (accounts in the Home Organisation) belongs to an individual
>>>     person (i.e. there are no shared accounts like "libraryuser1")?
>>
>> important
>>
>>>     and all users are traceable (i.e. the Home Organization knows who they are and
>>>     can reach them)?
>>
>> important
>>
>>>     and the Home Organisation is willing to collaborate with you if you think their
>>>     user misbehaves in your service?
>>
>> important
>>
>>>     that you (as an SP) can block him/her from your service?
>>
>> important
> 
> Is it? How often does abuse happen at SPs, and does it get detected,
> while the account itself would persist (i.e., not be suspended after a
> complaint to the IdP operator)?

I am not aware of any case either. But I do know that abuse can and will happen the more interesting material there is on offer. Case in point:
This material is available for EPA=faculty to all of Clarin (text behind the link is in English): https://www.csc.fi/-/suomi24-keskustelut-avattiin-yhteiskunnalliseen-tutkimuskayttoon and also via simple application in https://lbr.csc.fi. The "Suomi24" discussion portal sells the same data to companies. Already for that single instance I could not personally claim that being able to block misbehaving users from my SP is "not so important". Also, as far as I understand the Data Seal of Approval process, good behaviour of users is required, so SPs must be able to deal with digressions.

> 
>>>
>>>     user identifiers are persistent i.e. a user account is not re-assigned
>>>     (re-cycled) to another person over time?
>>
>> important. Recycling of old IDs should be phased out.
> 
> Does that happen (statistics)?

Recycling of old ePPNs does not happen to my knowledge, but it is possible as far as I know the definitions.

> If SP operators still identify users by
> ePPN, is it a good position to claim recycling should be phased out,
> when it is allowed by the ePPN semantics? Seems like a weak position
> to me.

This is based on my own services which heavily rely on ePPN and I understand Lindat uses ePPN too. I believe that the role of Attibute Authorities (like lbr.csc.fi) will grow in the future, increasing the need for unique ePPNs.
 
>>
>>>     user identifiers are shared by multiple SPs  i.e. if you have 2 SPs, do they
>>>     both receive the same user identifier when the same user logs in to the two
>>>     services?
>>
>> important
>>
>>>
>>> 3.2.Initial proof of identity
>>>
>>> How important is it for you that
>>>
>>>     the Home Organization has a documented identity vetting process (whatever it is)
>>>     in English and you can study it?
>>
>> important
>>
>>>     each Home Organisation has a machine-readable tag that indicates how the
>>>     organization carries out identity proofing and the tag is from a well-defined
>>>     international vocabulary?
>>
>> important
>>
>>>     each user in a Home Organisation has the above tag and different end users in
>>>     the same organization can have different tags (depending how their identity was
>>>     initially proofed)?
>>
>> important
>>
>>>     the identity proofing is done face-to-face based on a government photo-ID or
>>>     equivalent?
>>
>> Highly depends on the service. For sensitive data strong identification is
>> needed, for less sensitive data a simpler procedure might be enough.
>>
>>>
>>> 3.3.On-line authentication
>>>
>>>     Are password-based authentication good enough for you?
>>
>> Again, it depends. For now, yes.
> 
> I think a lot can be said about this. Personally, I think password
> based authentication should be disfavored.

Yes, me too. Here I dared to claim that it is nevertheless good enough for Clarin as a whole for now, if it is good enough for the Universities' own Identity Management.

> Many experts seem to think
> so, if I'm not mistaken both Google and Microsoft have expressed their
> intention to move past password-based authentication. Read e.g.
> http://www.wired.com/2012/11/ff-mat-honan-password-hacker/ .
> 
> This topic seems not only relevant for technical reasons, but whether
> or not password-based authentication is used or some alternative also
> has an impact on usability.
> 
> I believe this is a topic that merits investigation under the banner of CLARIN+.

I agree, are there such plans?
 
>>>     Should passwords have some kind of quality floor? (What kind of quality floor?)
>>
>> The organisation should state whether it requires a quality floor and what that
>> is.
> 
> Don't know if this is an open or closed question, but why not be a bit
> stricter/more explicit in ‘our demands’ on this? If the IdP operators
> would indeed write such policy documents, then what would SPF SP
> operators really do with it? Block specific IdPs? That seems like a
> lot of overhead, and like it's not going to happen. In that case the
> IdP operator's efforts would go to waste.

I would argue that IdPs could state in their metadata: Password complexity rules are in place: yes/no.
"yes" means they have thought about it and a (possibly weak) policy in place. I would stop there and only go further if IdPs that say "yes" have constant problems with identity theft.

> 
>>>     Do you need two factor authentication? (What kind of?) Are you willing to share
>>>     its costs?
>>
>> For now, no. But that might change in the future.
>>
>>> 3.4.Step-up authentication as a service
>>>
>>> Step-up authentication means that the user first authenticates with a password,
>>> and subsequently with a second factor (such as by a one-time password delivered
>>> to his/her cellphone). Step-up authentication could be delivered to research
>>> communities as a service.
>>>
>>> Would you like to make use of step-up authentication
>>>
>>>     if it costs you money?
>>>     if it costs you work (for instance, you need to operate one or several
>>>     registration authorities where your community's users come to show their
>>>     photo-ID and you record their cellphone number)?
>>
>> So far there was no need for 2 factor authentication.
> 
> Not an unreasonable answer I think. But could you support this answer
> (e.g. with evidence)?

To my knowledge no depositor has ever asked for it. I would also argue that Language Resource protection is relatively low in the food chain if you consider that students mananage their studies, library activities, etc, using the same credentials (eg. https://wiki.oulu.fi/display/ok2/Optima+Learning+Environment). But this might very well change if material on offer via AAI becomes more sensitive. I am not aware of such a case, though.

> 
>>> 4. Questions on user attributes
>>>
>>> Besides an identifier, the Home Organisation's Identity Provider is able to
>>> deliver also other attributes of the person that logs in.
>>> 4.1. Freshness of user accounts and attributes
>>>
>>> Many Home Organisations close the user account when an individual departs (e.g.
>>> researcher changes his/her employer). Closing the account closes also federated
>>> access to your SP. However, some organisations keep the accounts open (e.g. to
>>> serve alumni etc).
>>>
>>>     Do you expect that user accounts are closed as a user departs? How promptly?
>>
>> Prompty, roughly within a week.
>>
>>>     Do you expect that user's role attributes (e.g. eduPersonAffiliation="faculty")
>>>     value is updated as an individual departs? How promptly?
>>
>> Prompty, roughly within a week.
>>
>>
>>>
>>> 4.2. Quality/provenance of user data
>>>
>>> In larger universities the IdP/IdP gathers users' attributes from several
>>> registries (payroll system, CRIS system, student registry) with varying data
>>> quality. Some attributes can even be self-asserted by the user him/herself.
>>>
>>>     Is it important for you to know the quality/provenance of the user data on the
>>>     attribute level? What attributes? On what level of granularity?
>>
>> Freely changeable attributes should be marked, if that is possible.
>> Self-asserted attributes are not a problem as long as the self-assertion has
>> been signed by the user.
> 
> I did not understand what signed by the user means in this context.
> Could you explain? (Was this a closed question?)

I meant that a user self-asserts, say his phone number, but signs for the data to be correct during registration at the university. As opposed to changing data at will via a web form.

>>> 4.3. Population and release of attributes
>>>
>>>     What are the key attributes Home Organisations should populate for their end
>>>     users and release to your SP?
>>
>> The recommendations of the Data Protection Code of Conduct should be enough:
>> displayName, cn, mail, eduPersonAffiliation, eduPersonScopedAffiliation,
>> eduPersonPrincipalName, SAML2 Persistent NameID (eduPersonTargetedID),
>> schacHomeOrganization and schacHomeOrganizationType
>>
>>> 5.Questions on audits
>>>
>>>     Is it enough for you that a Home Organisation self-asserts that it complies with
>>>     a certain LoA level?
>>
>> This heavily depends on the content that needs to be protected. For now
>> self-assertion is enough for most cases.
> 
> CLARIN+ should thoroughly inventorize what kind of content protection
> classes are appropriate for the current SPF services and resources. As
> far as know, otherwise this answer has no solid basis.

I based my answer on the fact that LoA rarely comes up in discussion. And I assume that this is due to the fact that (detected) abuse is rare. And after all, most of our user base come from academic institutions where we even now assume that those institutions know what they are doing in terms of user registration. Even without self-assertion.

>>>     Should some external body have some enforcement rights (e.g. Home identity
>>>     federation can remove “compliant” tag from the Home Organisation if there are
>>>     doubts that a Home Organisation fails its LoA level)?
>>
>> That would add trust, yes.
>>
>>>     Are internal audits needed?
>>>     Are external audits needed? Are you willing to share their costs?
>>
>> To be meaningfully able to self-assert LoA over a longer time period
>> organisations will have to perform internal and external audits.
>> The organisation should not only state its LoA level but also how it is
>> maintained. SPs can then filter, if needed.
> 
> Seems reasonable. Obviously it remain a difficult issue though. For
> instance, what is the assurance level of such a LoA assertion itself?
> ;)

Yes you can play this game ad infinitum. And that is my personal reason for not over-emphasizing audits and certificates and the rest of it, even though they might be helpful. The DSA-process is largely peer-reviewed self-assertion and I found it very helpful for us. But take a few ISO 9000/1 examples for your entertainment: http://qualitymanagementsystem.com/total-quality-management/the-greatest-quality-management-system-failures-in-history

> If SPs start filtering on this openly, this kind of far-fetched
> question may become relevant as IdPs may try to counter being filtered
> out. Not that I think the latter is realistic right now.

Yes. And filtering out does not need to be black and white. If I have a user from a lower LoA-IdP I can always try to use alternative ways of making sure he is who he claims to be, if the protected resource so demands.

Cheers,
Martin