If you have full control of access to Accumulo, then it's just as easy to
remove that visibility from the incoming set.


On Wed, Mar 19, 2014 at 11:22 AM, Jeff Kunkle <[email protected]> wrote:

> My particular use case meets both of those conditions. I'd like to use a
> not operator to soft delete things for specific groups of users, which are
> assigned a given authorization. For example, assume I have two groups of
> users: group1 and group2. If I want to temporarily hide something from
> group1 I would add "& !group1" to the visibility. In my case I'm not really
> using the NOT operator for access control. The users in the group have
> access to the data; they've just chosen to hide it from their view.
>
> On Mar 19, 2014, at 10:47 AM, Sean Busbey <[email protected]>
> wrote:
>
>
> On Wed, Mar 19, 2014 at 9:36 AM, kunklejr <[email protected]> wrote:
>
>> So is there any consensus on whether this should be included? I would use
>> it
>> right away on a current project if it were. I understand the security
>> risks
>> that have been discussed with having a NOT operator, but I see its use as
>> a
>> decision to be made by the development team. If the project deems use of a
>> the NOT operator as too risky, then they should implement a design that
>> doesn't use it. I don't think you can prevent people from making poor
>> decisions/implementations simply by limiting the functionality. It could
>> be
>> misused as is today.
>>
>
>
> Could you describe the use case you have in mind? In order for NOT to be
> usable today, you'd need one of two things:
>
> 1) Use of visibility labels for something other than enabling access
> control (because you'd have to expressly design for users being trusted to
> only misrepresent themselves appropriately)
>
> 2) Client requests would need to pass through an broker application that
> controlled the set of user authorizations and knew not to leave off any of
> the ones used in NOT expressions. This would require a hard network
> boundary between all untrusted clients and Accumulo
>
>
> -Sean
>
>
>

Reply via email to