Oracle read-only user can insert/update/delete data via specially crafted views
Dear bugtraq-Reader
Last Thursday 6th April 2006, Oracle released a note on the Oracle
knowledgebase Metalink with details about an unfixed security vulnerability
(=0day) and a working test case (=exploit code) which effects all versions of
Oracle from 9.2.0.0 to 10.2.0.3. This note "363848.1 ? A User with SELECT
Object Privilege on Base Tables Can Delete Rows from a View" was available last
week to Metalink customers. The note was also displayed in the daily headlines
section of the Metalink.
That?s why this information can be assumed as public knowledge and
DBAs/Developers which missed the note on Metalink should know this
vulnerability in order to avoid/mitigate the risk (if possible) whilst waiting
for a patch from Oracle.
After noticing the note, I informed Oracle secalert that releasing such
information on Metalink is not a wise idea. Oracle normally criticises
individuals and/or companies for releasing information about Oracle
vulnerabilities (like David Litchfield from NGSSoftware for releasing
information an ever not fixed bug in mod_plsql gateway). In this case, not only
Oracle released detailed information on the vulnerability; they also included
the working exploit code on the Metalink.
In an interview, the Oracle CSO stated: ?I?ve known customers to terminate
contracts ? for releasing exploit code? you might get applause from hackers?
but business will not pay you to slit their throats. With knowledge comes
responsibility.?
After my email, Oracle removed the note from Metalink.
Problem:
In Oracle versions (9.2.0.0-10.2.0.3) exists an unpatched vulnerability which
allows users with ?SELECT? only privileges on a base table to insert/update/
delete data via a specially crafted view.
The impact of this vulnerability on the Oracle data dictionary is low because
most data dictionary tables don?t have a primary key which is a requirement for
this vulnerability.
The impact on custom applications can be huge and eliminate the entire role
concept because in well designed applications there is normally a read-only
role for low-privilege users (e.g. reporting or external auditors). If these
low-privileged users are able to create a view, which is standard in Oracle
9.2.x to 10 g R1, they could also insert, update and delete data via a
specially crafted view. Depending on the architecture, it is possible to modify
data, escalate privileges, ?
Test cases:
Oracle provided a complete test case in note 363848.1. I decided not to publish
such code on the internet as long as patches are not available. If you need
additional information you could contact me via email. A test case (without the
specially crafted view) is available on my website:
http://www.red-database-security.com/advisory/oracle_modify_data_via_views.html
Patches:
Currently there are no patches available. According to Oracle secalert Oracle
will provide patches in a future critical patch update.
Red-Database-Security is not convinced that the April 2006 CPU will contain
patches against this vulnerability.
Workarounds / Risk Mitigation:
Sanitize the connect role (9i - 10g R1) and remove the CREATE VIEW (and CREATE
DATABASE LINK, ?) privilege from the connect role.
Removing the primary key from the base table solves the problem too. Be aware
that this could cause performance and integrity issues on the application.
Oracle recommends creating views the option ?WITH CHECK OPTION?. This
recommendation helps against accidental modification but not against hackers.
Credits:
Special thanks to Jens Flasche who made Red-Database-Security aware of the
Metalink note and for the first analysis + additional test cases.
URLs:
Interview: Oracle CSO - Mary Ann Davidson
http://news.com.com/When+security+researchers+become+the+problem/2010-1071_3-5807074.html
Metalink Hacking
http://www.red-database-security.com/wp/oracle_metalink_hacking_us.pdf
----------------------------------------------------------------------
Are you interested in additional information about Oracle security?
Our next Oracle Anti-Hacker-Training:
23-may ? 26-may (4 days (english) ? Milano / Italy)
29-may ? 2-june (5 days (english) ? Cupertino [CA] / U.S.A)
19-june ? 23-june (5 days (german) ? Oberursel/Frankfurt / Germany)
----------------------------------------------------------------------