Message-ID: <913200DB-38FF-11D9-BE43-000A95E0092E@juniper.net>
Date: Wed, 17 Nov 2004 17:17:08 -0800
From: Bryan Burns <bburns@...iper.net>
To: bugtraq@...urityfocus.com
Subject: A Brief Analysis of Bofra/MyDoom.AG/AH


Overview of Bofra
==============

Bofra (AKA MyDoom.AG/AH) is a worm that was first discovered on November
8th circulating in the wild.  Bofra spreads by sending e-mail to victims
with a URL pointing back to a special webserver running on the infected
machine.  Bofra runs this small webserver on port 1639[1] (0x666 + 1)
and provides two pieces of content:

1. A request for /reactor will return a copy of the worm
2. A request for anything else will return an HTML document that
attempts to exploit the recent and unpatched IE iframe/frame/embed name
overflow vulnerability[2].

The payload of the IE exploit causes the /reactor URL to be downloaded
and executed, thereby infecting the host.

Upon infection, an infected host will do the following:

1. Add keys in the registry to start Bofra at startup
2. Start the infection web server on port 1639[1]
3. Gather e-mail addresses from the host's hard-drive and send e-mails
to them to spread the infection.
4. Connect to a random IRC server[3] and join a specific channel, based
on the date.

The IRC connection acts as a simple backdoor into the system, allowing
the author of the worm to further control the infected hosts.

IRC Backdoor
============

Machines infected with Bofra will select a random IRC server from a
provided list[3] and join a channel determined by passing the current
date into a channel name generation algorithm (detailed below).  Once
joined to the channel, the backdoor will accept a handful of commands
provided via PRIVMSG (either to the channel or to the client directly.)

Each command sent to the IRC backdoor needs to be prefixed with a
command word and is obfuscated with a key, both of which are changed
daily along with the channel.  Someone who knows the command word, key,
and obfuscation algorithm can send commands to the backdoor to gather
information about the infected host or instruct the backdoor to download
and execute an arbitrary executable from the internet.  The executable
download capability can be used to re-infect a host with the latest
Bofra variant, or install any additional malware the author desires.

By idling in one of the Bofra channels, I was able to catch someone
"upgrading" the bots using the execute command.  The URL provided (after
de-obfuscation) pointed to an executable with a jpg extension hosted on
a russian webserver (http://kjh0.narod.ru/pic9.jpg).  ClamAV identified
the executable as being MyDoom.AD.

Channel/Key Generation Algorithm
================================

In order to avoid easy containment, Bofra picks a new IRC channel,
command word and key every day.  Bofra uses the GetSystemTime function,
which uses UTC time, to determine the date, so instead of switching at
midnight, infected machines gradually migrate to the new channel over
the course of the day based on the local timezone.

The algorithm uses the year, month and day passed through some
trigonometric functions as a seed to srand, then uses the output of
rand() in series to generate:

1. channel name length (at most 8 chars)
2. channel name
3. command word (always 8 chars)
4. key (int between 0 and 25)

The input to srand is (sin(month) * cos(day)) + sin(1/year) interpreted
as an int.  Due to the particulars of the compiler used to generate the
Bofra executable, this portion of the algorithm is expressed in assembly
language in order for VC++ to generate equivalent code.

Here is functionally equivalent C code for generating the channel name,
command word and key:

-----------------------------------------------------------------------
void random_string(char *buf, int len){
         int i;

         for(i = 0; i < len; i++){
                 buf[i] = (rand() % 26) + 'a';
         }

         buf[len] = 0;
}

void get_chan_word_and_key(){
         SYSTEMTIME t;
         GetSystemTime(&t);

         double x = 1.0;
         double d;
         unsigned int n;

         // (sin(month) * cos(day)) + sin(1/year)
         _asm {
                 FILD t.wMonth
                 FSIN
                 FILD t.wDay
                 FCOS
                 FMULP ST(1), ST
                 FILD t.wYear
                 FDIVR x
                 FSIN
                 FADDP ST(1), ST
                 FSTP QWORD PTR d
                 MOV ECX, DWORD PTR d
                 MOV n, ECX
         };

         srand(n);

         int len = rand() & 0x80000003;
         if(len < 0){
                 len--;
                 len |= 0xFFFFFFFC;
                 len++;
         }

         len += 5;

         char chan[9];
         random_string(chan, len);

         char command[9];
         random_string(command, 8);

         int key = rand() % 26;

         printf("%hd/%hd/%hd: #%s %s %d\n", t.wYear, t.wMonth, t.wDay,
	chan, command, key);
}
-----------------------------------------------------------------------

Some example values:
11/17/2004: channel: #ygufwz command word: ahxvvnyx key: 14
11/18/2004: channel: #gjfuzg command word: htcyswlz key: 12
11/19/2004: channel: #eysyrtc command word: hdhcxqus key: 14

A list of channel names has already been provided to dalnet and undernet
OPERs and future Bofra channels have been disabled on both networks.

Obfuscation Algorithm
=====================

All commands and their arguments (if any) are obfuscated using a simple
algorithm.  The daily "key" determines the first character's
transformation.  Each subsequent character is transformed by the
previous character.  There is a small flaw in this algorithm: while you
do need to know the key to obfuscate a string, you do not need to know
they key to return it to the original text, as only the first character
requires knowledge of the key to transform.  Of course, with a total of
26 different keys, this is certainly not a robust algorithm anyway.

Here is functionally equivalent C code for obfuscating/de-obfuscating a
string:

-----------------------------------------------------------------------
char transform(char c, int key){
   char *uppers = "ABCDEFGHIJKLMNOPQRSTUVWXYZ";
   char *lowers = "abcdefghijklmnopqrstuvwxyz";
   char *idx;

   if(key < 0)
     key += 26;

   if(idx = strchr(uppers, c)){
     return uppers[(idx - uppers + key) % 26];
   } else if(idx = strchr(lowers, c)){
     return lowers[(idx - lowers + key) % 26];
   } else {
     return c;
   }
}

void decrypt(int key, char *s, char *d){

   while(*s){
     char c = *s;

     *d = transform(c, -key);

     if(c >= 'a' && c  <= 'z')
       key = ((int)c) - 'a';
     else if(c >= 'A' && c <= 'Z')
       key = ((int)c) - 'A';

     s++;
     d++;
   }
}

void encrypt(int key, char *s, char *d){

   while(*s){
     char c = *s;

     *d = transform(c, key);

     c = *d;

     if(c >= 'a' && c  <= 'z')
       key = ((int)c) - 'a';
     else if(c >= 'A' && c <= 'Z')
       key = ((int)c) - 'A';

     s++;
     d++;
   }
}
-----------------------------------------------------------------------

A sample obfuscated URL: vohw://sok.lzseewptk.mam/yyjffwa.ebf

Even though this URL was encoded with the key 14, the de-obfuscation
with the wrong key (in this case 1) is still obvious:

URL above de-obfuscated w/ key 1: uttp://www.botmaster.com/malware.exe

----------
[1] Bofra attempts to bind to port 1639 first, but if that fails it will
try 1640, 1641, etc.
[2] http://secunia.com/advisories/12959/
[3] Bofra picks a random IRC server from the following list to connect
to:
    broadway.ny.us.dal.net
    brussels.be.eu.undernet.org
    caen.fr.eu.undernet.org
    ced.dal.net
    coins.dal.net
    diemen.nl.eu.undernet.org
    flanders.be.eu.undernet.org
    graz.at.eu.undernet.org
    london.uk.eu.undernet.org
    los-angeles.ca.us.undernet.org
    lulea.se.eu.undernet.org
    ozbytes.dal.net
    qis.md.us.dal.net
    vancouver.dal.net
    viking.dal.net
    washington.dc.us.undernet.org


-Bryan

Powered by blists - more mailing lists