[teiid-issues] [JBoss JIRA] (TEIID-2311) Add simple row based security to data roles

Steven Hawkins (JIRA) jira-events at lists.jboss.org
Fri Dec 14 11:40:17 EST 2012


    [ https://issues.jboss.org/browse/TEIID-2311?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12741535#comment-12741535 ] 

Steven Hawkins commented on TEIID-2311:
---------------------------------------

Added the initial check-in of row-based security.  The approach is basically as outlined above.  A permission element can now have a condition child:

<permission>
  <resource-name>SCHEMA.TABLE</resource-name>
  <condition>COLUMNA=user()</condition>
  <allow-read>true</allow-read>
  ...
</permission> 

The affected resource can be a table/view or procedure.  Multiple conditions against the same resource are accumulated disjuctively.  Unlike data roles, the conditions are always applied (not just at the user query).  So it may be typical that the permissions are written for an any authenticated role and generalized, such as shown above with the use of the user function.

The condition can be any valid sql referencing the projected columns, but the current implementation will not correctly handle subqueries for bulk inserts (that could be addressed later if needed).

The condition is applied conjunctively to update/delete/select where clauses.  That means that those queries will only ever be effective against the subset of rows that pass the condition, i.e. SELECT * FROM TBL WHERE blah AND condition

inserts and updates against tables are further validated so that the insert/change values must pass the condition (evaluate to true) for the insert/update to succeed - this is effectively the same a sql constraint.  This will happen for all styles of insert/update - insert with query expression, bulk insert/update, etc.  However this can inhibit pushdown in some circumstances as we have to check the values for each row.  inserts/updates against views are not checked with regards to the constraint.

Further changes may be necessary to ensure that performance is not unduly harmed when the condition cannot be pushed down - most optimizations are already flexible about this case, but we could more aggressively look at raising the criteria to produce a better plan.
                
> Add simple row based security to data roles
> -------------------------------------------
>
>                 Key: TEIID-2311
>                 URL: https://issues.jboss.org/browse/TEIID-2311
>             Project: Teiid
>          Issue Type: Feature Request
>          Components: Query Engine
>    Affects Versions: 8.2
>            Reporter: Steven Hawkins
>            Assignee: Steven Hawkins
>             Fix For: 8.3
>
>
> A common request is to implement row based security.  The common workaround of modifying transformations is generally not a good solution.
> We should look at adding support for simple table filters and column masks.
> To be effective, filtering permissions however would have to act differently than normal data roles.  They would need to be applied all the time - and not just against the end user queries.
> For example, for tables:
>  <permission>
>    <resource-name>SCHEMA.TABLE</resource-name>
>    <filter>COLUMNA=2</filter>
>  </permission> 
> Meaning allow the CRUD of the given row only if COLUMNA has the value of 2.  Any valid predicate against just the referenced table would be allowed as a filter.  Each such permission would be applied as an additional predicate any time the table is referenced (in views, inserts, updates, deletes, etc.).  
> Allows would not be specified here as we want the filter to always specify inclusion.  Any applicable permissions in additional roles would be applied disjunctively - filter OR filter.  
> We could possibly support column masks via case expressions, such as:
>  <permission>
>    <resource-name>SCHEMA.TABLE.COLUMN</resource-name>
>    <mask>CASE WHEN ...</mask>
>  </permission> 
> However this is slightly more complicated.  Presumably the mask would only apply to projection and makes more sense to be applied at the final output/user query (more like a data role).  
> If we work the issue to specify the object type of a permission, then the name could alternatively refer to datatype or even an extension property to make the masking a little easier.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira


More information about the teiid-issues mailing list