[Freeipa-devel] Proposal for #4456 -- Regular users should not be able to add OTP tokens with custom name

[Nathaniel McCallum]

On Tue, 2014-09-16 at 12:52 -0400, Simo Sorce wrote:
> On Tue, 16 Sep 2014 12:33:16 -0400
> Nathaniel McCallum <npmccallum at redhat.com> wrote:
> 
> > On Tue, 2014-09-16 at 12:28 -0400, Simo Sorce wrote:
> > > On Tue, 16 Sep 2014 12:23:51 -0400
> > > Nathaniel McCallum <npmccallum at redhat.com> wrote:
> > > 
> > > > On Tue, 2014-09-16 at 12:22 -0400, Simo Sorce wrote:
> > > > > On Tue, 16 Sep 2014 11:52:43 -0400
> > > > > Nathaniel McCallum <npmccallum at redhat.com> wrote:
> > > > > 
> > > > > > On Tue, 2014-09-16 at 17:51 +0200, Petr Vobornik wrote:
> > > > > > > On 16.9.2014 17:26, Nathaniel McCallum wrote:
> > > > > > > > On Tue, 2014-08-19 at 17:10 -0400, Nathaniel McCallum
> > > > > > > > wrote:
> > > > > > > >> Admins need the ability to specify the token ID in the
> > > > > > > >> case of imports. However, generally, this ability is not
> > > > > > > >> needed.
> > > > > > > >>
> > > > > > > >> Is it possible to offload the ID generation to the
> > > > > > > >> ipa-uuid plugin? I'm not quite sure how to enable this
> > > > > > > >> (I think it involves passing a magic value?). But I'm
> > > > > > > >> not quite sure how this fits in with the IPA framework
> > > > > > > >> as the generated value is the DN.
> > > > > > > >>
> > > > > > > >> However, assuming this can be used, I propose the
> > > > > > > >> following. The token ID is removed from the UI for
> > > > > > > >> regular users (but retained for admins). We change the
> > > > > > > >> ACIs for token addition/modification to prevent regular
> > > > > > > >> users from specifying the ID in an add or mod operation.
> > > > > > > >> The CLI would retain the option to set it, but this
> > > > > > > >> option would only be usable by admins.
> > > > > > > >>
> > > > > > > >> Make sense?
> > > > > > > >
> > > > > > > > Nobody has responded to this. :)
> > > > > > > >
> > > > > > > > However, since investigating it a bit more, this approach
> > > > > > > > won't really work without further effort. Here is the
> > > > > > > > problem.
> > > > > > > >
> > > > > > > > First, the UUID plugin doesn't currently support this
> > > > > > > > kind of operation. Either it needs to be modified or a
> > > > > > > > new plugin needs to be created.
> > > > > > > >
> > > > > > > > Second, the client needs to know the ID in order to
> > > > > > > > generate the token URI. If we generate the UUID inside
> > > > > > > > the DS, the UUID is unknown to the client and the URI
> > > > > > > > can't be generated. This would mean a new control.
> > > > > > > >
> > > > > > > > As I see it we have three options:
> > > > > > > > 1. Remove the option to specify the ipatokenUniqueID from
> > > > > > > > the GUI. Don't make any change in the CLI.
> > > > > > > >
> > > > > > > > ENFORCEMENT: none
> > > > > > > > EFFORT: low
> > > > > > > >
> > > > > > > > 2. Perform a server-side check for admin membership.
> > > > > > > > Raise an exception if the ipatokenUniqueID is specified
> > > > > > > > and the user is not an admin.
> > > > > > > >
> > > > > > > > ENFORCEMENT: API-level
> > > > > > > > EFFORT: medium
> > > > > > > >
> > > > > > > > 3. Modify otptoken-add to create tokens with a magical
> > > > > > > > ipatokenUniqueID value by default. An ACI would prevent
> > > > > > > > normal users from adding tokens without this magic value.
> > > > > > > > Create/modify a plugin to generate UUIDs when the magic
> > > > > > > > value is found. Send a control back to the client
> > > > > > > > indicating the real ipatokenUniqueID value. Modify
> > > > > > > > otptoken-add to read this control.
> > > > > > > >
> > > > > > > > ENFORCEMENT: DS-level
> > > > > > > > EFFORT: high
> > > > > > > >
> > > > > > > > I think my preference for now is #1. Thoughts?
> > > > > > > >
> > > > > > > > Nathaniel
> > > > > > > >
> > > > > > > 
> > > > > > > I want to raise a question if we really want to disable this
> > > > > > > feature for normal users. Wouldn't it be better to improve
> > > > > > > error message or check if the name is taken?
> > > > > > > 
> > > > > > > UUID is the value which is displayed in the Soft. Token
> > > > > > > application as name. Sure, one can rename it there to match
> > > > > > > the description in IPA but that seems quite unpleasant to
> > > > > > > me. Also user has to know that he can rename the token in
> > > > > > > FreeOTP,....
> > > > > > > 
> > > > > > > Atm the existence check might be little problematic -
> > > > > > > disclose of information which is not readable to user by
> > > > > > > default. But current state is basically it, just unfriendly.
> > > > > > 
> > > > > > The existence check is impossible: users can't see tokens they
> > > > > > don't own.
> > > > > 
> > > > > Why don't we simply allow user to use an arbitrary name of their
> > > > > choosing (or a random string if not specified) and then append
> > > > > their uid string to it in the UI ?
> > > > > 
> > > > > I wonder if we have ACIs where the value must match a pattern
> > > > > defined by the bound user ...
> > > > 
> > > > This would present a strange scenario where a token is created and
> > > > then reassigned to another user.
> > > 
> > > 
> > > The idea is to strip away the tail when visualizing to the user, so
> > > it wouldn't really bee seen ?
> > > Or perhaps the token can simply be renamed when assigned.
> > 
> > It just feels like a kludge to me. Also, automated stripping may run
> > afoul of names assigned by the imports.
> > 
> > What is wrong with the three options I provided above? We can do #1
> > now and #2 or #3 later.
> 
> Nothing wrong, really, #1 is fine by me.

I settled on #2 since it turned out to be somewhat easy. I think we also
need an update to the UI (aka #1).

This should be an easy review:
https://www.redhat.com/archives/freeipa-devel/2014-September/msg00351.html

Nathaniel