HDFS Background

• When a file or directory is created, its owner is the user identity of the client process, and its group is inherited from parent (the BSD rule). Permissions are taken from default umask. Extended Acl's are taken from parent unless they are set explicitly.

Goals

To reduce need to set fine-grain file security props after every operation, users may want the following Hive warehouse file/dir to auto-inherit security properties from their directory parents:

• Directories created by new database/table/partition/bucket
• Files added to tables via load/insert
• Table directories exported/imported (open question of whether exported table inheriting perm from new parent needs another flag)

What may be inherited:

• Basic file permission
• Groups (already done by HDFS for new directories)
• Extended ACL's (already done by HDFS for new directories)

Behavior

• When "hive.warehouse.subdir.inherit.perms" flag is enabled in Hive, Hive will try to do all above inheritances. In the future, we can add more flags for more finer-grained control.
• Failure by Hive to inherit will not cause operation to fail. Rule of thumb of when security-prop inheritance will happen is the following:
  • To run chmod, a user must be the owner of the file, or else a super-user.
  • To run chgrp, a user must be the owner of files, or else a super-user.
  • Hence, user that hive runs as (either 'hive' or the logged-in user in case of impersonation), must be super-user or owner of the file whose security properties are going to be changed.

JIRA