SQL injection exploitation internals

SQL injection
exploitation internals
How do I exploit this web
application injection point?
Intercon III, London
January 9, 2009
Bernardo Damele A. G.
bernardo.damele@gmail.com

Intercon III, London – January 9, 2009 2
About me
Bernardo Damele A. G.
● Proud father
● Penetration Tester and Security Researcher
● Currently working at Portcullis Computer Security Ltd
● sqlmap lead developer

SQL... what? (1/2)
● From the OWASP Testing Guide:
“SQL injection attacks are a type of injection attack, in
which SQL commands are injected into data-plane input in
order to affect the execution of predefined SQL commands”
● There are plenty of resources on the Net about SQL injection
concept: it is a high-risk web application security flaw
● A long list of resources can be found on my delicious profile,
http://delicious.com/inquis/sqlinjection
● I keep it updated with stuff I consider valuable
● A wise man once told me:
“An image is worth thousands words”

SQL... what? (2/2)
Source http://xkcd.com/327/

All right, tons of resources and I am still presenting about SQL
injection, why?
Because:
● New techniques have been released in the last year
● Some aspects have been over-looked in the past
● It is fun!
State of art

Basically the steps to go through are:
● Detection of a possible SQL injection flaw
● SQL query syntax detection
● Back-end database management system fingerprint
● Depending on the session user privileges, back-end DBMS
and some possible security settings in place server-side, a
SQL injection issue leads on the DBMS server to:
● DBMS data unauthorized access
● File system read and write access
● Operating system command execution
How does it work?

sqlmap is an automatic SQL injection tool:
● Developed in Python. Started on July 2006 initially by
Daniele Bellucci, then I took over in December 2006
● Licensed under the terms of GPLv2
● Detects and take advantage of SQL injection vulnerabilities
in web applications. The user can choose to:
● Perform an extensive back-end DBMS fingerprint
● Enumerate users, password hashes, privileges,
databases, tables, columns and their datatypes
● Dump entire or user's specified database tables' entries
● Run custom SQL statements and more
sqlmap

sqlmap key features:
● Full support for MySQL, Oracle, PostgreSQL and
Microsoft SQL Server back-end DBMS software
● Full support for three SQL injection techniques:
● Inferential blind SQL injection
● UNION query SQL injection
● Stacked queries (multiple statements) support
● Target aquisition: from user, by parsing WebScarab/Burp
proxies requests log files, by Google dorking
● Tests for injection flaws on GET and POST parameters,
HTTP User-Agent header and Cookie values
sqlmap features (1/2)

More features:
● Silent to verbose output messages
● Granularity in the user's options
● Support for concurrent HTTP requests (multi-threading)
● Estimated time of arrival
● Session save and resume
● Options from command line and/or configuration file
● Integration with Metasploit and w3af
● File system read and write access and operating system
command execution by providing own queries, depending
on the session user privileges and back-end DBMS
sqlmap features (2/2)

Real world
Have you ever had a dream, Neo, that you were so
sure was real? What if you were unable to wake from
that dream? How would you know the difference
between the dream world and the real world?
Morpheus, The Matrix

In the real world web applications are often complex
Usually the page content changes at each refresh
● They have inline counters, advertisement banner, clocks,
etc.
Inferential blind SQL injection algorithm is based on the
concept that the HTTP responses differ depending on the SQL
query
Dealing with advertisements (1/3)

Obstacle
If the page content does not depend only on the
SQL statement and changes at each refresh then
the algorithm may not work

Python library helped to solve this problem: for each HTTP
response sqlmap calls a function that compares the returned
page content with the untouched original page content:
● Return a measure of the page contents' similarity as a float
in the range [0, 1] with a radio of 3.
● It works also when the original page is stable, but the
injected query with a valid condition (True) differs
If the automatic comparison fails, the user can provide a
string or a regular expression to match on both original and
True page contents and to not match on False page contents

● In standard SQL language NULL is allowed as a value for a
table column field
● In the inferential blind SQL injection technique usually a
bisection algorithm is used to identify if the ordinal value of
the Nth query output character is higher of a certain ASCII
table number: this causes the page content to be True or
False
● The SQL statement used by sqlmap, depending on the back-
end DBMS, is similar to:
ASCII(SUBSTR((SQL query), Nth SQL query
output char, 1)) > Bisection algorithm
number
To NULL or not to NULL (1/4)

Obstacle
On some DBMS the substring function can not be
used on NULL

● A possible solution for this problem consists in modifying all
SQL query's columns explicitly:
● Casting its output to be a string
● Returning value ' ' (space) if the casted value is still
NULL
● Example on MySQL 5.0. The SQL query to enumerate the
column name first entry is:
SELECT name FROM test.users LIMIT 0, 1
● Casted SQL query:
SELECT IFNULL(CAST(name AS CHAR(10000)),
CHAR(32)) FROM test.users LIMIT 0, 1

The inferential blind SQL injected statement will be then:
ORD(MID((SELECT IFNULL(CAST(name AS
CHAR(10000)), CHAR(32)) FROM test.users
LIMIT 0, 1), Nth SQL query output character,
1)) > Bisection algorithm number
URL encoded:
ORD%28MID%28%28SELECT%20IFNULL%28CAST%28name
%20AS%20CHAR%2810000%29%29%2C%20CHAR
%2832%29%29%20FROM%20test.users%20LIMIT
%200%2C%201%29%2C%201%2C%201%29%29%20%3E
%2063

You have got an injection point
The injection point is in a SQL statement as follows:
SELECT * FROM users WHERE id LIKE ((('%" .
$_GET['id'] . "%'))) LIMIT 0, 1
SQL payload (1/3)

Obstacle
The injection is after a LIKE clause, within three
parenthesis, the statement terminates with a
LIMIT clause
SQL payload (2/3)

In this example the SQL payload that sqlmap identifies is:
id=1'))) AND ((('RaNd' LIKE 'RaNd
In the inferential blind SQL injection algorithm will be:
id=1'))) AND ORD(MID((SQL query), Nth SQL
query output character, 1)) > Bisection
algorithm number AND ((('RaNd' LIKE 'RaNd
In the UNION query SQL injection technique will be:
id=1'))) UNION ALL SELECT NULL, Concatenated
SQL query, NULL# AND ((('RaNd' LIKE 'RaNd
SQL payload (3/3)

First demo
I did every demo possible to see if the things would
do what they were promising they would do
Doug Hall

You have got an injection point
It is vulnerable to UNION query SQL injection:
● sqlmap detected it for you by NULL bruteforcing or by
ORDER BY clause bruteforcing, depending on your options
Bypass columns limitation (1/4)

Obstacle
The number of columns in the web application
SELECT statement is lower than the number of
columns of your UNION ALL SELECT statement

● A possible solution consists in concatenating your SELECT
statement columns in a single output by using the
specific DBMS string concatenation operator or function
● Example on PostgreSQL 8.3 to retrieve users privileges
● The SQL query to inject is:
SELECT usename, usecreatedb, usesuper,
usecatupd FROM pg_user

The injection will be:
UNION ALL SELECT NULL, CHR(83)||CHR(114)||
CHR(108)||CHR(71)||CHR(86)||CHR(116)||
COALESCE(CAST(usename AS CHARACTER(10000)),
CHR(32))||CHR(104)||CHR(100)||CHR(122)||
CHR(81)||CHR(121)||CHR(90)||
COALESCE(CAST(usecreatedb AS
CHARACTER(10000)), CHR(32))||CHR(104)||
CHR(100)||CHR(122)||CHR(81)||CHR(121)||
CHR(90)||COALESCE(CAST(usesuper AS
CHR(100)||CHR(122)||CHR(81)||CHR(121)||
CHR(90)||COALESCE(CAST(usecatupd AS
CHR(121)||CHR(80)||CHR(65)||CHR(68)||
CHR(102), NULL FROM pg_user

Obstacle
You have got an injection point vulnerable to
UNION query SQL injection. Only the query
output's first entry or a range of entries is
displayed in the page content
Going partial.. UNION (1/3)

sqlmap automatizes a known technique:
● Changes the parameter value to its negative value causing
the original query to produce no output
● Inspects and unpack the provided SQL statement:
● Calculates its output number of entries
● Limits it after the UNION ALL SELECT to return one
entry at a time
● Repeat the previous action N times where N is equal to
the number of entries

● Example on MySQL 4.1 to enumerate list of databases
● The SQL query to inject is:
SELECT db FROM mysql.db
● sqlmap identified the injection point as being an non-quoted
parameter (integer) in the WHERE clause with the equal
operator (simple scenario)
● The injection will be:
id=1 UNION ALL SELECT NULL,
CONCAT(CHAR(100,84,71,69,87,98),IFNULL(CAST
(db AS CHAR(10000)), CHAR(32)),
CHAR(65,83,118,81,87,116)), NULL FROM
mysql.db LIMIT 0, 1# AND 6972=6972

● Back-end DBMS fingerprinting is a mandatory step to go
through to take full advantage of a SQL injection flaw
● There are a few well known techniques and a few over-looked
techniques
● sqlmap implements up to four techniques, three of these are
in use by other tools:
● The user can force the back-end DBMS software value: no
HTTP requests are sent to identify the software
● By default a basic DBMS fingerprint based on one or two
techniques is done: only two HTTP requests are sent
● The user can choose to perform an extensive DBMS
fingerprint based on four techniques: numerous
(30+) HTTP requests are sent
DBMS fingerprint (1/4)

● The techniques implemented to perform an extensive back-
end DBMS fingerprint are:
● Inband error messages
● Banner parsing
● SQL dialect
● Specific functions static output comparison
● On a default installation all of them are reliable
● On a hardened installation the last two are reliable

Example of basic back-end DBMS fingerprint on PostgreSQL
8.3
● The techniques in use are two
● The two SQL queries injected to identify it are:
AND integer::int=integer
● SQL dialect
AND COALESCE(integer, NULL)=integer
● Specific function static output comparison

Example of extensive back-end DBMS fingerprint on Microsoft
SQL Server 2005
● The techniques in use are three
● The result is:
active fingerprint: Microsoft SQL Server 2005
banner parsing fingerprint: Microsoft SQL Server
2005 Service Pack 0 version 9.00.1399
html error message fingerprint: Microsoft SQL Server
● Active fingerprint refers to SQL dialect and
specific functions static output comparison

● Fingerprinting is a key step in penetration testing
● It is not only about back-end DBMS software
● There are techniques and tools to fingerprint the web server,
the web application technology and their underlying system
● What about the back-end DBMS underlying system?
● sqlmap can fingerprint them without making extra requests:
● Web/application server and web application technology: by
parsing the HTTP response headers (Server, X-AspNet-
Version, X-Powered-By, etc.) – known technique
● Back-end DBMS operating system: by parsing the
DBMS banner – over-looked technique
More on fingerprint

Second demo
A demo, as in "demolish", or "demonstration"?
Cyclops, X-Men: Evolution

It might comes in handy sometimes to be able to run your own
SQL queries, mainly for file system read and write access
and operating system command execution
The tool inspects the provided statement:
● If it is a SELECT statement sqlmap uses, depending on
user's options, the inferential blind or the UNION query
technique to retrieve its output
● If it is a data manipulation statement, a transaction
statement or any other valid SQL statement, it uses
stacked queries to run it if the web application supports
them
Give me a SQL shell!

Automation vs granularity (1/2)
sqlmap has been developed to make it simple for a busy
penetration tester to detect and exploit SQL injection
vulnerabilities in web applications
● Providing it with a source of targets, it can automatically:
● Detect all possible SQL injections and confirm them
● Identify the SQL query syntax
● Fingerprint the back-end DBMS
● The user does not have to look on the Net for DBMS
specific queries then manually inject them to enumerate
users password hashes, check if the session user is a
DBA, enumerate table columns' datatype, etc.
● There is an option to dump the whole back-end DBMS

Automation vs granularity (2/2)
● The tester is a professional, he knows what he does and why
● There are options to specify:
● How to compare True and False HTTP responses
● A single or more testable parameters
● The SQL payload prefix and postfix
● A single or a range of entries to dump from a table
● A single or multiple columns to dump from a table
● Custom SQL statements to run
● Options can be specified from both command line and/or
configuration file
● Options are documented in the user's manual with examples

Third demo
I get tons of uninteresting mail, and system
announcements about babies born, etc. At least a
demo MIGHT have been interesting
Richard Stallman

Limitations
Can sqlmap fail to detect or to exploit a SQL injection
vulnerability?
● Yes, in some cases mainly because it does not support:
● SQL injection on SQL clauses other than WHERE
● Time based blind SQL injection technique
...but I am working on these and others!

Want to contribute?
I am always looking forward to code contributions
Try it, find bugs, send feature requests, review the code and
the documentation, contribute on the mailing lists!
I can provide you with:
● Details on code internals
● Write access to the Subversion repository
● Access to the development platform
● A beer if you are in London area

Links and contacts
Homepage: http://sqlmap.sourceforge.net
Documentation:
● http://sqlmap.sourceforge.net/dev/index.html
● http://sqlmap.sourceforge.net/doc/README.pdf
Mailing lists:
● https://lists.sourceforge.net/lists/listinfo/sqlmap-users
● https://lists.sourceforge.net/lists/listinfo/sqlmap-devel
Personal contacts:
● E-mail / Jabber: bernardo.damele@gmail.com
● Blog: http://bernardodamele.blogspot.com

References
● OWASP Testing Guide, Open Web Application Security Project
● Exploit of a Mom, xkcd
● Deep Blind SQL Injection, Ferruh Mavituna (Portcullis Computer Security Ltd)
● Microsoft SQL Server sp_replwritetovarbin limited memory overwrite vulnerability,
Bernhard Mueller (SEC Consult Vulnerability Lab)
● Metasploit Framework, H D Moore and the Metasploit development team
● w3af, Andres Riancho and the w3af development team
● Data-mining with SQL Injection and Inference, David Litchfield (NGS Software)
● Advanced SQL Injection, Victor Chapela (Sm4rt Security Services)
● Python difflib, Python Software Foundation
● NULL (SQL), Wikipedia
● Agent oriented SQL abuse, Fernando Russ and Diego Tiscornia (CORE Security)
● Insight on UNION query SQL injection, Bernardo Damele A. G.
● DBMS Fingerprint, Daniele Bellucci (OWASP Backend Security Project)

Questions?
Thanks for your attention

SQL injection exploitation internals

More Related Content

What's hot

Viewers also liked

Similar to SQL injection exploitation internals

Recently uploaded

SQL injection exploitation internals