Penn State’s QFilter puts security in XML database queries


Author: Jay Lyman

New software developed at Penn State University promises to protect
XML database queries and filter out unauthorized requests, thereby
boosting query performance as many as 100 times, researchers said.

QFilter, created by PSU School of Information assistant professor Dongwon
Lee and three others at the university, can bypass the typical access
control modules built into individual databases for security. The software
can be deployed with off-the-shelf databases without requiring substantial
changes to them, according to Lee.

“In XML access control, many proposed solutions are available recently,”
he said. “But they are not practical in our opinion since they usually
assume that underlying XML database has some kind of built-in security
features. This is usually not true — there is no known XML database product
with security features, not to mention that there are not many XML database
products themselves.”

To make matters worse, Lee explained, when object relational databases
such as DB2, Oracle, or SQL Server are used as the underlying database
engine, the proposed solutions become useless because they use an XML model, as opposed to the relational model of the underlying database.

QFilter, Lee said, can sit between the users and the database and filter out
unauthorized requests for data before a database responds to a query. The
shift from data filtering to query filtering provides a practical solution
to access control issues and also boosts query-response time by rejecting
unauthorized requests earlier, Lee said.

Although QFilter is not yet in its final version, Lee said he is working
with school officials to release the software in open source form next year.

“I’m currently discussing this with the Penn State IP office while my
student is polishing the software,” Lee said. “It may take awhile —
sometime in 2005, I hope. If ready, then we probably will go by either
Berkeley [license] style or GNU style, but I’m not sure at this point.”

A second lens on stream processing

Developed on Windows XP and written with Java and Galax — an open
source XML kernel developed at Bell Labs — QFilter may take stream
processing to a new level, Lee said. The process whereby a stream of data is
viewed through the lens of query is reinforced by a second lens through
another query with QFilter, Lee explained.

“The current implementation supports most of Xpath query language,” Lee said. “In the future, we want to add more features to the list.”

The Penn State assistant professor and the rest of the QFilter team —
PSU doctoral student Bo Luo, assistant professor Peng Liu, and associate
professor Wang-Chien Lee — presented a paper on the software in early
November at the ACM Conference on Information and Knowledge Management in
Washington D.C. Lee said the paper, “QFilter: Fine-Grained Run-Time XML
Access Control with NFA-based Query Rewriting,” was well received. The
software, however, is still in its alpha stage, despite the authors having begun work in late 2002.

More efficient query response

There are other technologies to restrict access to databases, such as the
view-based approach that creates different data views for each user. Since
user credentials do not have to be checked once views are created, this can
increase query response speed. However, as the number of users requesting
access grows or if views need to be frequently updated, this method can
cause maintenance and storage headaches, QFilter researchers contend.

By rejecting unauthorized queries early on, QFilter can improve query
processing time dramatically, depending on different queries and data, Lee

“When users ask invalid queries, QFilter detects and rejects them
outright,” he said. “Thus, it saves a lot of unnecessary database query

When users ask valid queries, QFilter detects that the query is
authorized using non-deterministic finite automata (NFA) — which store
access control policies and monitor query flows — and informs the
database. In turn, the underlying database does not require a security
check, leaving it free to focus on regular query processing, Lee said.

When users send some queries that request both unauthorized and
authorized data, QFilter “prunes out” the unauthorized request part upfront,
Lee said.

“As a side effect, usually this pruning results in more optimized query
processing,” he said.

According to the Penn State team’s experimentation, end-to-end processing
time from when query was issued to when answers were returned improved
between 10 and 100 times with QFilter, depending on the XML query and data
types, Lee said.