cubicweb #3154558 [security] rdefs using default read permissions: just do nothing [open]

Comments

• 2013/09/19 08:23, written by acampeas

  First problem seen while implementing this: the test unittest_security.test_update_rql_permission breaks.

  It works on the following (reduced) schema, with read perms shown:

  class Affaire:
      __permissions__ = ('managers', 'X owned_by U', 'X concerne S?, S owned_by U')
      sujet = String()
      concerne = Relation(Societe)
  
  class Societe:
      __permissions__ = ('managers', 'users', 'guests')
  

  We have a scenario where a "user" (belonging only to the users group):

  1. creates an Affaire
  2. creates a Societe and links to the Affaire through "concerne"
  3. wants to set the "sujet" attribute of the Societe

  As of today it all works. Let's detail point 3 security check:

  • the querier performs a read group check on the "sujet" attribute * since the default sec on attributes is ('managers', 'users', 'guests') it passes (!) * later, the rqlrewriter interpolates the rql expressions, hence since the user owns both ends of the "concerne" relation it works ok

  It think this way of life is a pb., because:

  • when writing a schema like this, the typical end user NEVER thinks she just gave ('managers', 'users', 'guests') group permissions to the Societe attributes,
  • instead she believes the permission rule at entity level controls the whole entity (that is, attributes-security), unless a specific attribute permission rule is written

  I have the following scenario to implement: a "readonly" profile. Unfortunately, it cannot be made to work (with a "readonly" group at least) unless every attribute of every entity type is explicitly given a non-default, e.g. ('managers', 'users', 'guests', 'readonly') rule. Which is a bit unfriendly...

  Opinions ?
  • 2013/09/19 09:48, written by acampeas

    Well, doing nothind wrt rdefs read permissions == default read permissions "just works".
    
• 2013/09/28 10:24, written by acampeas

  this actually must be extended to write security checks
  
• 2014/08/01 13:23, written by acampeas

  As of 01-08-2014, the need rises again. Let's do it !