Details
-
New Feature
-
Status: Open
-
Major
-
Resolution: Unresolved
-
0.9.0
-
None
-
None
-
None
Description
In this issue, we would like to discuss the possibility of supplementing the Hive authorization model with authorization at the storage level. As discussed in HIVE-1943, Hive should also check for operation permissions in hdfs and hbase, since otherwise, data and metadata can be in an inconsistent state or be orphaned. Going a step further, some of the setups might not need the full featured auth model by Hive, but want to rely on managing the permissions at the data layer. In this model, the metadata operations are checked first from hdfs/hbase and it is allowed only if they are allowed at the data layer. The semantics are documented at https://cwiki.apache.org/confluence/display/HCATALOG/Hcat+Security+Design.
So, the goals of this issue are:
- Port storage handler specific authorization providers, and the StorageDelegationAuthorizationProvider from
HCATALOG-245and HCATALOG-260 to Hive. - Keep current Hive's default authorization provider, and enable user to use this and/or the storage one. auth providers are already configurable.
- Move the manual checks that had to be performed about authorization in Hcat to Hive, specifically:
- CREATE DATABASE/TABLE, ADD PARTITION statements does not call
HiveAuthorizationProvider.authorize() with the candidate objects, which means that
we cannot do checks against defined LOCATION. - HiveOperation does not define sufficient Privileges for most of the operations,
especially database operations. - For some of the operations, Hive SemanticAnalyzer does not add the changed
object as a WriteEntity or ReadEntity.
- CREATE DATABASE/TABLE, ADD PARTITION statements does not call