[
https://issues.apache.org/jira/browse/HBASE-3025?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12916817#action_12916817
]
Andrew Purtell commented on HBASE-3025:
---------------------------------------
Attached is a first cut at a coprocessor based access controller. It requires
the patches pending for HBASE-2001 and HBASE-2002/HBASE-2321, and has
additional dependency on secure Hadoop classes. Therefore we expect this to be
developed in a feature branch until HBase compilation and operation on secure
vs. nonsecure Hadoop flavors can be made seamless.
Package documentation will be forthcoming.
This access controller coprocessor should be associated with all tables as a
system extension. (These are coprocessors listed in hbase-site.xml and loaded
into the regionserver at an early time. See package docs in the patch on
HBASE-2001 for more information pertaining to coprocessors specifically.)
Permissions for user tables are stored in a new {{acl:}} family in META. The
access controller is active on META and user tables. When active on META, the
access controller mirrors the contents of this family into a znode tree in
zookeeper and updates the mirrored permissions information when values in
{{acl:}} are added, changed, or deleted. When active on user tables, the access
controller reads the permissions for its table from the appropriate znode,
caches them, and sets a watch on the znode, updating the local cache whenever
the znode data changes. The {{acl:}} family in META serves as persistent
storage for access policy and as the canonical interface for defining access
permissions. ZooKeeper serves to immediately and atomically propagate policy
changes into the local permissions caches of all nodes in the cluster. For the
typical user operation neither META nor ZooKeeper need be consulted when
determining if a user has sufficient access privilege; up to date information
will be found in the local cache.
A new shell command, {{grant}}, is added to support granting or revoking
specific rights to tables. This is accomplished by puts or deletes into the new
{{acl:}} family in META.
When tables are created the current (creating) user is given ownership of it. A
table owner has full access to the table and can grant additional access.
Ownership is tracked using an attribute of HTableDescriptor. Therefore
currently a table must be disabled and modified to change ownership.
This patch also contains a small modification to HMaster that introduces the
concept of _superuser_, a specially privileged principal defined via
configuration variable, or by default the principal under which the Master or
RegionServer processes are running. (Currently, for proper functioning of the
cluster, the superuser must be the principal under which the processes are
running.) Only the superuser can modify a table descriptor. This prevents a
user from arbitrarily reassigning ownership and therefore bypassing access
control. We may keep the superuser concept or replace it with an explicit ALTER
permission. Perhaps even only the superuser should be allowed to enable and
disable tables, though this patch does not at present prevent any user from
taking those actions.
> Coprocessor based simple access control
> ---------------------------------------
>
> Key: HBASE-3025
> URL: https://issues.apache.org/jira/browse/HBASE-3025
> Project: HBase
> Issue Type: Sub-task
> Reporter: Andrew Purtell
> Assignee: Eugene Koontz
> Attachments: HBASE-3025.1.patch
>
>
> Thanks for the clarification Jeff which reminds me to edit this issue.
> Goals of this issue
> # Client access to HBase is authenticated
> # User data is private unless access has been granted
> # Access to data can be granted at a table or per column family basis.
> Non-Goals of this issue
> The following items will be left out of the initial implementation for
> simplicity:
> # Row-level or per value (cell) This would require broader changes for
> storing the ACLs inline with rows. It's still a future goal, but would slow
> down the initial implementation considerably.
> # Push down of file ownership to HDFS While table ownership seems like a
> useful construct to start with (at least to lay the groundwork for future
> changes), making HBase act as table owners when interacting with HDFS would
> require more changes. In additional, while HDFS file ownership would make
> applying quotas easy, and possibly make bulk imports more straightforward,
> it's not clean it would offer a more secure setup. We'll leave this to
> evaluate in a later phase.
> # HBase managed "roles" as collections of permissions We will not model
> "roles" internally in HBase to begin with. We will instead allow group names
> to be granted permissions, which will allow some external modeling of roles
> via group memberships. Groups will be created and manipulated externally to
> HBase.
> While the assignment of permissions to roles and roles to users (or other
> roles) allows a great deal of flexibility in security policy, it would add
> complexity to the initial implementation.
> After the initial implementation, which will appear on this issue, we will
> evaluate the addition of role definitions internal to HBase in a new JIRA. In
> this scheme, administrators could assign permissions specifying HDFS groups,
> and additionally HBase roles. HBase roles would be created and manipulated
> internally to HBase, and would appear distinct from HDFS groups via some
> syntactic sugar. HBase role definitions will be allowed to reference other
> HBase role definitions.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
