Windows Executable Infection
by
Qark and Quantum [VLAD]
This document attempts to explain technically NewExe infection for the
virus writer. This isn't for the novice coder and you'll need to
understand DOS EXE infection before being able to use any of it.
The infection described in detail here is the same as used by the
accompanying WinSurfer virus, there are other methods that can be done
but they aren't our concern (it's up to you to develop new code!)
You will want a copy of the New Exe header information which is available
in Ralf Browns interrupt listings under Int21h AH=4Bh, a copy of which is
included at the end of this article.
This is a map of what we are trying to do:
Before infection After Infection
0h--------------------- 0h---------------------
| | | |
| DOS EXE header | | DOS EXE header |
| | | |
--------------------- ---------------------
| | | |
| DOS Code | | DOS Code |
| | | |
| | 3F8h---------------------
400h--------------------- | |
| | <- Move | New EXE Header |
| New EXE Header | this | and some tables |
| and some tables | up | |
| | | ----------------- |
| ----------------- | | Segment |
| Segment | | Table |
| Table | | |
| | | |
| | Insert -> | ----------------- | Virus segment
| ----------------- | this | ----------------- | entry
| Misc Other | | Misc Other |
| Tables | | Tables |
| | | |
x x x x
x x x x
x x x x
CS:IP --------------------- ---------------------
| | | |
| Windows Code | | Windows Code |
| and Misc | | and Misc |
| Segments | | Segments |
| | | |
End --------------------- CS:IP ---------------------
| Virus segment |
Append all | with code etc |
this -> | |
--------------------- Virus
| | Relocation
End --------------------- Entry
Infection Theory:
-----------------
Test the EXE to make sure its windows and that the NE is at offset 1024.
Reduce the NE pointer at 3ch by eight.
Reduce DOS SP by eight.
Any NE offsets which are larger than the segment table offset must be
increased by eight.
Increment the segment counter.
Save the CS:IP.
Point the CS:IP to the new segment entry
Calculate the position of the end of the segment table, move all data up to
and including the segment table up by eight bytes.
Add another segment entry
Write the virus to the end of the file
Write relocation entry to the end of the file.
---
That is a very basic overview of what is wanted.
The main idea of it all is moving the NE header forward to add a new
segment table entry, and changing the CS:IP to point to it.
We'll go through all the stages step by step.
Testing for a Windows executable.
---------------------------------
A NewEXE file always begins with a DOS header and some DOS executable
code. With a windows executable:
The word at offset 0 is 'MZ'.
The word at offset 18h is 40h or greater.
assuming that these conditions are met then we also want the pointer to the
NE header (3ch) to be equal to 400h. The reason for this is that room is
needed to move the NE header forward by 8.
Rewriting the DOS header.
-------------------------
Before rewriting the DOS header, reduce the DOS SP field (10h) by 8, because
if it was pointing to the end of the DOS code section it would point to
the start of the NE header. (No real reason for doing it really)
The NE header pointer at 3ch must be reduced by 8, because we intend on
moving most of the NE header forward to that position.
Modifications to the NE header.
-------------------------------
From this point onwards we are referring to the offsets in relation to the
NE header.
The pointer to the segment table is a word at 22h. If any of the pointers
at 4,24h,26h,28h,2ah are larger than [22h], then increase that
pointer by eight. The reason for this is that since we are inserting a
new segment table entry, all tables behind the segment table will be
eight bytes further from the start of the header.
The segment counter is a word at offset 1ch. Increment it by one because
since we are adding another segment, we need to indicate this in the
header.
At offset 14h is the dword pointer to the program entry point, CS:IP
where the CS is actually the segment table entry number. Save this pointer
for use in your relocation entry (that part will be explained in detail
later on). Now point the CS:IP instead to the virus segment. Do this
by writing the necessary starting IP value to 14h (offset of the entry into
your segment) and the segment counter value into 16h (since the virus
segment is the last in the segment table it will be equal to the segment
counter value which we have incremented).
You can work this out for yourself, but what you need to do now is move
the entire NE header, upto and including the segment table, but not
the data behind it, forward by eight bytes. Heres an equation on
calculating how much to move:
((segmentcounter-1)*8)+word ptr [22h]
Directly after this position is where to insert the new virus segment entry.
The Virus Segment Entry.
------------------------
Format of new executable segment table record:
00h WORD offset in file (shift left by alignment shift for byte offs)
02h WORD length of image in file (0000h = 64K)
04h WORD segment attributes (see below)
06h WORD number of bytes to allocate for segment (0000h = 64K)
Offsets 2 and 6 in the segment record are just the virus size. Offset
4 is the segment attribute, we use 180h, which makes the segment executable
with relocations.
Calculating offset 0 is more difficult. It is necessary to get the file
length and shift it right by the alignment shift value which was saved
earlier. I wouldn't know how to do a dword shift so I convert it to a
division.
File length into DX:AX :
mov ax,4202h
xor cx,cx
cwd
int 21h
Shift right DX:AX by alignment shift:
mov cl,byte ptr alignment_shift
push bx
mov bx,1
shl bx,cl
mov cx,bx
pop bx
div cx
AX=required offset 0 value
Write this eight byte table behind all the moved header.
Writing the Virus.
------------------
Lseek to the end of the executable and write the virus.
Writing the Relocation Entry.
-----------------------------
You need a relocation entry so that you can return control to the host
after the virus has done its dirty work. The relocation entries follow
the segment itself.
Format of relocation entry:
Offset Size Description
00 WORD Number of relocation entries
Offset Size Description
00 BYTE relocation type
01 BYTE relocation flags
02 WORD offset within segment
04 WORD target address segment
06 WORD target address offset
The first word will be 1 because only 1 relocation entry.
First byte is 3 to signal a 32bit pointer (a far jump).
Second byte is 4 to signal additive relocation. (doesn't work without this)
Word at offset 2 is the offset of the dword after the 'far jump' opcode.
Word at offset 4 is the CS of the original host CS:IP
Word at offset 6 is the IP of the original host CS:IP
To put all this into code:
db 0eah ;Opcode of JMP FAR PTR
ret_ip dw 0
dw 0ffffh
relocation dw 1 ;One relocation entry
db 3 ;32bit pointer relocation
db 4 ;Additive relocation
dw offset ret_ip ;Offset of the relocation in segment
hostcs dw 0 ;CS:IP of place we want our relocation to
hostip dw 0 ; point.
Note: the relocation address data (ret_ip) _must_ be FFFF:0000 as above
because windows treats it as a linked list. If you don't understand
don't worry, just do it! It won't work otherwise. It is important
that before writing the virus body to the executable that you set
this address to FFFF:0000 so that executable will be setup correctly.
Write the relocation entry to the end of the file.
Windows allows a standard Int 21h call just fine so all file manipulation
is as simple as ever, with a few minor changes.
Writting TSR/ISR's under Windows.
---------------------------------
Windows is a protected mode environment and thus to set a vector under it we
must manipulate the various tables that protected mode requires to be
updated. Of these tables one is of interest to us. The "Interrupt
Descriptor Table" (IDT). The IDT holds all the Protected Mode interrupt
vectors that are in the "Interrupt Vector Table" (IVT) - if they are
supported - and some more. This is all well and good - all we have to do is
set a vector in the IDT to point to our code like we do with dos and the
IVT but - there are complications.
V86 mode.
---------
When Windows is running in 386 enhanced mode the processor is locked into
v86 mode. In v86 mode each application has its own 1mb memory block. All
EMS/XMS is locked away from the application and (usually) each process has
its own IDT/IVT. The idea being to allow the application to think it is
alone on the system. Any attempt to access any other tasks that may be
running is a "violation of system integrity" and will cause an exception.
This is no good for a virus and by now you're thinking we're screwed but
windows solves the problem for us. Windows has but one IDT and of course
only one IVT to shadow it and thus if we are to a hook a vector in the IDT
it will be reflected in every task. (Ever seen that OS/2 warp advert where
they say "... and true multi-tasking" - well this simply means that OS/2
warp holds a seperate IDT for every task under v86 mode.)
Setting the Vector.
-------------------
So let's get this straight.. the IVT is the one at 0:0 that we all know and
love and the IDT is a protected mode/v86 shadow of it. So you're most
probably asking "why can't we just hook the IVT and let windows reflect it
into the IDT?" - well there's 2 reasons: firstly windows doesn't "reflect"
the IVT as we would like to think, actually it copies the IVT into the IDT
on startup then replaces certain routines with "protected mode call
structures" which call the real mode routines and some routines it doesn't
even do this, secondly trying to access the IVT directly is a big nono and
will cause an exception error.
Can we use DOS ? well yes, you can.. most dos functions have been reflected
up into windows although windows is supposed to be trying to get rid of
them. So you can call int 21,25/35 to set the vector but remember that this
will be setting a vector in the GVT and only at the discretion of windows.
A better way: use DPMI. Windows is NOT your host under protected mode or
even v86 mode as microsoft would have you believe. Windows has an
overlooker that provides windows time-slicing and exception abilities. In
fact all windows does is act as a mediator to DPMI and basically slow things
down. So how do we use DPMI to set the vector ? well.. by using functions
0204h/0205h - with the vector in BL - of the DPMI provider int 31. This
way windows has no say over what goes in its IDT and thus we have a LOT of
control.
Writing the Interrupt.
----------------------
This is perhaps one of the easier things to do - once you understand the
principles. In protected mode/v86 mode your code segment is supposed to be
read only and thus you're supposed to have a code, stack and data segment.
Everyone knows this is bad so what we need is a way to write to the code
segment. You won't find one - writing to the code segment just isn't
possible. BUT - if we were to have a data segment that pointed to the same
code segment then we could do it. (actually we don't have any segments in
protect mode.. we have a thing called a "selector" which gives access
parameters to areas of memory.)
To get an "alias selector" to the code segment we can use another DPMI
function and bypass windows altogether: function ax=000ah with the code
selector in bx of int 31h, it returns a read/writeable alias selector to
the code segment in ax.
So now Protected mode isn't protected and we're free to write to the code
segment. One problem arises from this idea. Qark wanted to put this code
to generate the alias in the loader part of the virus and store it in the
code segment for the interrupt routine. This seemed fine until I found out
that windows likes to move segments around. One minute your code might be
at one address the next it may be at another. Thus you should always
regenerate the alias on every execution of the ISR or lock down the segment
- we chose against this last approach as it can cause exception errors from
windows (DPMI has no problems with this strategy.)
For a detailed look at DPMI consult Ralf Brown's interrupt listing under
int 31h. These functions haven't been included in this text.
Staying Resident.
-----------------
Finally: don't go resident off something that can be closed. If you go
resident off something like mine sweeper or something and the user closes
the application, your code segment will be deleted and removed from the IDT
and will point to an empty segment. This will cause windows to hang.
Solution: look for a process that will never be closed, direct action
infect this as soon as you find it and only ever go resident off it.
In the 'system.ini' file in your windows directory, is a variable 'shell='
that points to a file that is never closed. Mostly it will be
'progman.exe' but sometimes other programs are used, so read it in and
infect it.
To find the path of the windows directory, search the environment for a
variable called 'windir=' (yes, lowercase!). On entry to any windows
executable ES will point to the PSP (or something functionally similar).
The word at 2ch is the segment (selector) of the environment segment.
Scan through this as you would within a DOS application. We didn't discover
this through any documentation, but when Quantum accidentally typed 'set'
from within a DOS window.
Facts and Possible Techniques.
------------------------------
At offset 0eh in the New Executable header is the 'auto data segment index'.
This is another segment table index, which will be automatically in DS
on execution entry. Although this is pure guesswork, if you set that
'auto data segment' to the same segment as CS, (ie in every way the same
except the segment attributes) then you could write to your CS without
using DPMI and would also open the gateway to polymorphic windows viruses.
(Polymorphism would be pretty useless if you had to put a DPMI call before
you could write to your code segment.)
Mark Ludwig's 'Windows Virus #2' uses a different form of infection in that
it extends the length of the code segment as indicated in the segment
table, and moves the entire host program starting from the end of the file
back by virus length until the end of the code segment is reached, at which
point the virus is written. The IP is then repointed to the virus.
There are some advantages to this, such as no need to add relocation entries
because the jmp is intra-segment but this is greatly outweighed by the
fact that moving an entire file is very slow and a much larger number of
pointers in the header need to be modified to account for the change.
There are DPMI calls for allocating memory (Int 31h ax=501h) but it is
unknown at this point whether the memory remains allocated after the program
is closed. Mark Ludwig makes extensive use of these calls for temporary
buffers in his direct action 'Windows Virus #2'.
Mark Ludwig also demonstrates use of the WinAPI calls, which involve adding
a new relocation entry for each call and pushing some stuff on the stack.
This may become important when win95 comes out but for the moment it is
more convenient to simply use int 21h.
Some food for thought.
----------------------
Microsoft has a dream [here we go], the dream is to control the world. They
intend to do this with applications like Windows [scary isn't it ?], and by
the looks of it they might succeed. A lot of people think computers should
be easy, that there should be no skill involved in it and that everyone
should HAVE to use one. Microsoft are using this idea to flood the market
with shit like Windows, in the hope that it will be accepted as a "standard".
Windows is already this but Microsoft aren't happy with one success, they
want to be on top of the hill.
This is where we come in. If Microsoft is going to succeed in taking over
the world then we should be there making their lives difficult. The
WinSurfer virus created by us is so stable that an average windows
user wouldn't even notice it. [Then again the average windows user
wouldn't notice if his hair caught on fire.] And thus will stay with us
for quite some time. At least until AVers recognise the windows market
and start releasing scanners.
The New Executable format. (From Ralf Browns Interrupt listing 21 AH=4B)
--------------------------
new executables begin running with the following register values
AX = environment segment (Wrong! AX=0)
BX = offset of command tail in environment segment
CX = size of automatic data segment (0000h = 64K)
ES,BP = 0000h (Wrong! ES=PSP)
DS = automatic data segment
SS:SP = initial stack
the command tail corresponds to an old executable's PSP:0081h and
following, except that the 0Dh is turned into a NUL (00h); new
format executables have no PSP (All wrong)
Format of .EXE file header:
Offset Size Description
00h 2 BYTEs .EXE signature, either "MZ" or "ZM" (5A4Dh or 4D5Ah)
02h WORD number of bytes in last 512-byte page of executable
04h WORD total number of 512-byte pages in executable (includes any
partial last page)
06h WORD number of relocation entries
08h WORD header size in paragraphs
0Ah WORD minimum paragraphs of memory to allocation in addition to
executable's size
0Ch WORD maxi paragraphs to allocate in addition to executable's size
0Eh WORD initial SS relative to start of executable
10h WORD initial SP
12h WORD checksum (one's complement of sum of all words in executable)
14h DWORD initial CS:IP relative to start of executable
18h WORD offset within header of relocation table
40h or greater for new-format (NE,LE,LX,PE,etc.) executable
1Ah WORD overlay number (normally 0000h = main program)
---new executable---
1Ch 4 BYTEs ???
20h WORD behavior bits
22h 26 BYTEs reserved for additional behavior info
3Ch DWORD offset of new executable (NE,LE,etc) header within disk file,
or 00000000h if plain MZ executable
Format of new executable header:
Offset Size Description
00h 2 BYTEs "NE" (4Eh 45h) signature
02h 2 BYTEs linker version (major, then minor)
04h WORD offset from start of this header to entry table (see below)
06h WORD length of entry table in bytes
08h DWORD file load CRC (0 in Borland's TPW)
0Ch BYTE program flags (see below)
0Dh BYTE application flags (see below)
0Eh WORD auto data segment index
10h WORD initial local heap size
12h WORD initial stack size (added to data seg, 0000h if SS <> DS)
14h DWORD program entry point (CS:IP), "CS" is index into segment table
18h DWORD initial stack pointer (SS:SP), "SS" is segment index
if SS=automatic data segment and SP=0000h, the stack pointer
is set to the top of the automatic data segment, just below
the local heap
1Ch WORD segment count
1Eh WORD module reference count
20h WORD length of nonresident names table in bytes
22h WORD offset from start of this header to segment table (see below)
24h WORD offset from start of this header to resource table
26h WORD offset from start of this header to resident names table
28h WORD offset from start of this header to module reference table
2Ah WORD offset from start of this header to imported names table
(array of counted strings, terminated with a string of length
00h)
2Ch DWORD offset from start of file to nonresident names table
30h WORD count of moveable entry point listed in entry table
32h WORD file alignment size shift count
0 is equivalent to 9 (default 512-byte pages)
34h WORD number of resource table entries
36h BYTE target operating system
00h unknown
01h OS/2
02h Windows
03h European MS-DOS 4.x
04h Windows 386
05h BOSS (Borland Operating System Services)
37h BYTE other EXE flags
bit 0: supports long filenames
bit 1: 2.X protected mode
bit 2: 2.X proportional font
bit 3: gangload area
38h WORD offset to return thunks or start of gangload area
3Ah WORD offset to segment reference thunks or length of gangload area
3Ch WORD minimum code swap area size
3Eh 2 BYTEs expected Windows version (minor version first)
Note: this header is documented in detail in the Windows 3.1 SDK
Programmer's Reference, Vol 4.
Bitfields for new executable program flags:
Bit(s) Description
0-1 DGROUP type
0 = none
1 = single shared
2 = multiple (unshared)
3 = (null)
2 global initialization
3 protected mode only
4 8086 instructions
5 80286 instructions
6 80386 instructions
7 80x87 instructions
Bitfields for new executable application flags:
Bit(s) Description
0-2 application type
001 full screen (not aware of Windows/P.M. API)
010 compatible with Windows/P.M. API
011 uses Windows/P.M. API
3 is a Family Application (OS/2)
5 0=executable, 1=errors in image
6 non-conforming program (valid stack is not maintained)
7 DLL or driver rather than application
(SS:SP info invalid, CS:IP points at FAR init routine called with
AX=module handle which returns AX=0000h on failure, AX nonzero on
successful initialization)
Format of new executable segment table record:
00h WORD offset in file (shift left by align shift to get byte offs)
02h WORD length of image in file (0000h = 64K)
04h WORD segment attributes (see below)
06h WORD number of bytes to allocate for segment (0000h = 64K)
Note: the first segment table entry is entry number 1
Bitfields for segment attributes:
Bit(s) Description
0 data segment rather than code segment
1 unused???
2 real mode
3 iterated
4 movable
5 sharable
6 preloaded rather than demand-loaded
7 execute-only (code) or read-only (data)
8 relocations (directly following code for this segment)
9 debug info present
10,11 80286 DPL bits
12 discardable
13-15 discard priority
Format of new executable relocation data (immediately follows segment image):
Offset Size Description
00h WORD number of relocation items
02h 8N BYTEs relocation items
Offset Size Description
00h BYTE relocation type
00h LOBYTE
02h BASE
03h PTR
05h OFFS
0Bh PTR48
0Dh OFFS32
01h BYTE flags
bit 2: additive
02h WORD offset within segment
04h WORD target address segment
06h WORD target address offset