=====================================================================

                              CERT-Renater

                   Note d'Information No. 2023/VULN057

_____________________________________________________________________

DATE                : 03/02/2023

HARDWARE PLATFORM(S): /

OPERATING SYSTEM(S): Systems running Apache AGE PostgreSQL 11,
                     AGE for PostgreSQL 12 versions prior to 1.1.1.

=====================================================================
https://lists.apache.org/thread/455f0ntg5jdmomv8oc552pmtqjjt574y
_____________________________________________________________________


CVE-2022-45786: Apache AGE: Python and Golang drivers allow data
manipulation and exposure due to SQL injection


Severity: important

Description:

There are issues with the AGE drivers for Golang and Python that
enable SQL injections to occur. This impacts AGE for PostgreSQL
11 & AGE for PostgreSQL 12, all versions up-to-and-including 1.1.0,
when using those drivers.

The fix is to update to the latest Golang and Python drivers in
addition to the latest version of AGE that is used for
PostgreSQL 11 or  PostgreSQL 12.

The update of AGE will add a new function to enable
parameterization of the cypher() function, which, in conjunction
with the driver updates, will resolve this issue.

Background (for those who want more information):

After thoroughly researching this issue, we found that due to the
nature of the cypher() function, it was not easy to parameterize
the values passed into it. This enabled SQL injections, if the
developer of the driver wasn't careful. The developer of the
Golang and Pyton drivers didn't fully utilize parameterization,
likely because of this, thus enabling SQL injections.

The obvious fix to this issue is to use parameterization in the
drivers for all PG SQL queries. However, parameterizing all PG
queries is complicated by the fact that the cypher() function
call itself cannot be parameterized directly, as it isn't a real
function. At least, not the parameters that would take the graph
name and cypher query.

The reason the cypher() function cannot have those values
parameterized is because the function is a placeholder and never
actually runs. The cypher() function node, created by PG in the
query tree, is transformed and replaced with a query tree for the
actual cypher query during the analyze phase. The problem is that
parameters - that would be passed in and that the cypher() function
transform needs to be resolved - are only resolved in the execution
phase, which is much later. Since the transform of the cypher()
function needs to know the graph name and cypher query prior to
execution, they can't be passed as parameters.

The fix that we are testing right now, and are proposing to use,
is to create a function that will be called prior to the execution
of the cypher() function transform. This new function will allow
values to be passed as parameters for the graph name and cypher
query. As this command will be executed prior to the cypher()
function transform, its values will be resolved. These values can
then be cached for the immediately following cypher() function
transform to use. As added features, the cached values will store
the calling session's pid, for validation. And, the cypher()
function transform will clear this cached information after
function invocation, regardless of whether it was used.

This method will allow the parameterizing of the cypher() function
indirectly and provide a way to lock out SQL injection attacks.

References:

https://age.apache.org
https://www.cve.org/CVERecord?id=CVE-2022-45786



=========================================================
+ CERT-RENATER        |    tel : 01-53-94-20-44         +
+ 23/25 Rue Daviel    |    fax : 01-53-94-20-41         +
+ 75013 Paris         |   email:cert@support.renater.fr +
=========================================================