Mailing-List: contact commits-help@cassandra.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@cassandra.apache.org
Date: Mon, 19 Nov 2012 15:08:58 +0000 (UTC)
From: "Aleksey Yeschenko (JIRA)" <jira@apache.org>
To: commits@cassandra.apache.org
Message-ID: <1920628953.3024.1353337738778.JavaMail.jiratomcat@arcas>
In-Reply-To: <1027293777.42669.1351568892551.JavaMail.jiratomcat@arcas>
Subject: [jira] [Commented] (CASSANDRA-4874) Possible authorizaton handling
 impovements
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit


    [ https://issues.apache.org/jira/browse/CASSANDRA-4874?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13500289#comment-13500289 ] 

Aleksey Yeschenko commented on CASSANDRA-4874:
----------------------------------------------

Actually, that wouldn't work (a single GRANT permission).
I looked at two options, but both don't work:
1. Allow GRANT owners GRANT and REVOKE any permission on the resource (even those they don't have) - now we get the old FULL_ACCESS and all the associated problems (you have to check for GRANT and not just the requested permission and, secondly, REVOKE <perm> won't work as expected if a user still has GRANT permission on the resource).
2. Allow GRANT owners GRANT and REVOKE only the permissions they already have on the resource. This one is trickier:
2.1 Make P.GRANT non-recursive (disallow granting P.GRANT). Now you need to involve a superuser every time you want to allow someone grant permissions. This is bad since it involves superusers unnecessarily.
2.2 Make P.GRANT recursive (allow granting P.GRANT). Now we've got ourselves easy permission escalation. User A has every permission but P.GRANT. User B has P.SELECT and P.GRANT (or just P.GRANT). User B grants P.GRANT to user A. A grants B every permission he has. Now the two of them together have more permissions than the sum of their old ones. 
                
> Possible authorizaton handling impovements
> ------------------------------------------
>
>                 Key: CASSANDRA-4874
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-4874
>             Project: Cassandra
>          Issue Type: Improvement
>    Affects Versions: 1.1.6, 1.2.0 beta 1
>            Reporter: Aleksey Yeschenko
>            Assignee: Aleksey Yeschenko
>              Labels: security
>             Fix For: 1.2.0 rc1
>
>         Attachments: 4874-4875.txt, 4874-v2.txt, 4874-v3.txt, v1-v2.diff
>
>
> I'll create another issue with my suggestions about fixing/improving IAuthority interfaces. This one lists possible improvements that aren't related to grant/revoke methods.
> Inconsistencies:
> - CREATE COLUMNFAMILY: P.CREATE on the KS in CQL2 vs. P.CREATE on the CF in CQL3 and Thrift
> - BATCH: P.UPDATE or P.DELETE on CF in CQL2 vs. P.UPDATE in CQL3 and Thrift (despite remove* in Thrift asking for P.DELETE)
> - DELETE: P.DELETE in CQL2 and Thrift vs. P.UPDATE in CQL3
> - DROP INDEX: no checks in CQL2 vs. P.ALTER on the CF in CQL3
> Other issues/suggestions
> - CQL2 DROP INDEX should require authorization
> - current permission checks are inconsistent since they are performed separately by CQL2 query processor, Thrift CassandraServer and CQL3 statement classes.
> We should move it to one place. SomeClassWithABetterName.authorize(Operation, KS, CF, User), where operation would be a enum
> (ALTER_KEYSPACE, ALTER_TABLE, CREATE_TABLE, CREATE, USE, UPDATE etc.), CF should be nullable.
> - we don't respect the hierarchy when checking for permissions, or, to be more specific, we are doing it wrong. take  CQL3 INSERT as an example:
> we require P.UPDATE on the CF or FULL_ACCESS on either KS or CF. However, having P.UPDATE on the KS won't allow you to perform the statement, only FULL_ACCESS will do.
> I doubt this was intentional, and if it was, I say it's wrong. P.UPDATE on the KS should allow you to do updates on KS's cfs.
> Examples in http://www.datastax.com/dev/blog/dynamic-permission-allocation-in-cassandra-1-1 point to it being a bug, since REVOKE UPDATE ON ks FROM omega is there.
> - currently we lack a way to set permission on cassandra/keyspaces resource. I think we should be able to do it. See the following point on why.
> - currently to create a keyspace you must have a P.CREATE permission on that keyspace THAT DOESN'T EVEN EXIST YET. So only a superuser can create a keyspace,
> or a superuser must first grant you a permission to create it. Which doesn't look right to me. P.CREATE on cassandra/keyspaces should allow you to create new
> keyspaces without an explicit permission for each of them.
> - same goes for CREATE TABLE. you need P.CREATE on that not-yet-existing CF of FULL_ACCESS on the whole KS. P.CREATE on the KS won't do. this is wrong.
> - since permissions don't map directly to statements, we should describe clearly in the documentation what permissions are required by what cql statement/thrift method.
> Full list of current permission requirements: https://gist.github.com/3978182

--
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