rave-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Carlucci, Tony" <acarlu...@mitre.org>
Subject RE: [Proposal] Spring Permissions Change
Date Thu, 04 Oct 2012 15:53:23 GMT
>-----Original Message-----
>From: Chris Geer [mailto:chris@cxtsoftware.com]
>Sent: Thursday, October 04, 2012 11:32 AM
>To: dev@rave.apache.org
>Subject: Re: [Proposal] Spring Permissions Change
>
>On Thu, Oct 4, 2012 at 8:22 AM, Carlucci, Tony <acarlucci@mitre.org> wrote:
>
>> >-----Original Message-----
>> >From: Chris Geer [mailto:chris@cxtsoftware.com]
>> >Sent: Thursday, October 04, 2012 11:08 AM
>> >To: dev@rave.apache.org
>> >Subject: Re: [Proposal] Spring Permissions Change
>> >
>> >Tony,
>> >
>> >On Thu, Oct 4, 2012 at 6:58 AM, Carlucci, Tony <acarlucci@mitre.org>
>> wrote:
>> >
>> >> Hi Chris, could you go a little more into your use case (I think what
>> >> you've hinted at it with your Widget->add_comment block)?  I believe
the
>> >> spirit of that Permission enum was to define the context of the security
>> >> check to keep in line with CRUD actions.  The detailed business logic of
>> >> the Model/Permission Context combination can then be customized as
>> needed
>> >> in the various Default<Model>PagePermissionEvaluator.hasPermission
>> >> functions.  So if there is some specific security logic related to
>> adding a
>> >> comment to a widget, I believe you can put it in the appropriate
>> >> PermissionEvaluator class.
>> >>
>> >
>> >I understand the current model and I think it works great for top level
>> >objects but it doesn't work all so well for subordinate objects, or for
>> >business logic checks that are beyond CRUD. Right now everything is a top
>> >level object (everything has it's own repository for example) but as part
>> >of the object model restructure we have proposed to change that slightly.
>> >If you view WidgetComment as a subordinate object to a Widget, the
>> security
>> >checks are different. Instead of checking WidgetComment => "Create" as a
>> >standalone check, you really want to check Widget => "can_add_comment"
>> >which is at the Widget level since the Comment doesn't exist yet. This
>> >check would check to make sure the Widget is published, that the user has
>> >access to the Widget, etc. Once the WidgetComment exists, the current
>> >checks in place make sense (mostly).
>> >
>> >I know currently we could just check the widget in the
>> >WidgetCommentPermissionEvaluator because the WidgetComment has an
>> >attribute
>> >of "widget_id" but that is another thing we are proposing to change in the
>> >object model restructure. As we try and restructure things so that we can
>> >support backends other than JPA we need to tweak the object model at the
>> >interface level. For example, WidgetComment would no longer have an
>> >attribute of widget_id, it is just associated with whatever widget it's
>> >part of. This cleans up a few things like being able to create a
>> >WidgetComment with a widget_id of 3 but adding it to the WidgetComment
>> >collection of Widget id 2.
>> >
>> >Does that make sense?
>> >
>> >Beyond the WidgetComment example I still think there is a need for more
>> >fine grained permission checks. For example:
>> > - can_publish_widget
>> > - can_reset_other_users_password (low level admin who can't do some other
>> >functions)
>> > - can_delete_other_users_comment (like a moderator)
>> > - ...
>> >
>> >I know those functions can be covered by "admin" but I know our product
>> >needs a finer level of control than just "admin". This will become much
>> >more important as we start talking multi-tenancy which I'll bring up again
>> >soon where you need multiple levels of admins.
>> >
>> >Chris
>>
>> Ok yes, this definitely makes sense to me given the context of the model
>> refactoring changes.  I think changing the enum to strings should be fine,
>> so long as we don't see hard-coded "can_add_comment" strings in lots of
>> spots, which could make maintenance a little difficult.
>>
>
>The only place the string would show up would be in the hasPermission
>annotation (which is already a string) and in the PermissionEvaluator. In
>the permission  evaluators we could still define/use enums to define the
>valid strings, the interface just wouldn't be bound to a single enum. Sound
>reasonable?
>

Yup, +1

Tony

>>
>> Also we should keep in mind the current ability to override the default
>> security behaviors[1] as part of this change, and make sure only the most
>> common of changes should go into the Rave code base.
>>
>> [1] http://rave.apache.org/documentation/model-permission-override.html
>>
>> Tony
>>
>> >
>> >>
>> >> Am I understanding your use case or completely off the mark? :)
>> >>
>> >> Tony
>> >>
>> >> >-----Original Message-----
>> >> >From: Chris Geer [mailto:chris@cxtsoftware.com]
>> >> >Sent: Tuesday, October 02, 2012 7:50 PM
>> >> >To: dev
>> >> >Subject: [Proposal] Spring Permissions Change
>> >> >
>> >> >I would like to propose we change how the spring permission checks work
>> >> >slightly. Right now the "Permission" value (i.e. Create, Update...)
is
>> >> >defined as part of a enum named Permission defined in
>> >> >the org.apache.rave.portal.security.ModelPermissionEvaluator interface.
>> >> The
>> >> >various hasPermissions methods take an instance of that Permission enum
>> >> >(created from a string on the check permission annotation). Having the
>> >> >permissions defined in an enumeration limits what we are able to check
>> >> >permissions for in my opinion. Right now we have two choices, 1) limit
>> our
>> >> >permission checks to the list there is now, 2) add new permissions to
>> the
>> >> >generic Permissions enum which could lead to a bunch of permissions
>> stored
>> >> >on a generic enum that aren't really reusable (i.e. Widget ->
>> add_comment
>> >> >permission). I would like to propose we change the way we define
>> >> >permissions to remove the enum and just pass along the string defined
>> in
>> >> >the annotations. The only real downside of that is that we can't use
a
>> >> >switch/case statement during permissions checks unless we use Java7.
>> >> >
>> >> >Thoughts/concerns?
>> >> >
>> >> >Thanks,
>> >> >Chris
>> >>
>>

Mime
View raw message