Windows Named Pipes Part 4: Taking a Trip Down Static Analysis Lane

Windows Named Pipes Part 4: Taking a Trip Down Static Analysis Lane

Reversing & Exploiting Custom Windows Named Pipe Servers Continued

For the fourth and final blog post of our Windows Named Pipes series, VS-Lab Researchers demonstrate the following information:

  • Static Analysis

    • Reversing custom functionality exposed by the named pipe server
    • Reversing the named pipe server custom protocol
  • Exploitation

    • Walking through a single abuse case targeting given functionality within the named pipe server

These articles are not meant to be an exhaustive reverse engineering tutorial. To keep the material engaging and informative, some code will be covered at a high level to prevent this article from becoming extremely lengthy. If you have not read parts 1-3, skip to those articles using the links below.

Windows Named Pipe Servers Series:

Part 1: The Fundamentals of Windows Named Pipes
Part 2: Analysis of a Vulnerable Microsoft Windows Named Pipe Application
Part 3: Reversing & Exploiting Custom Windows Named Pipe Servers

Static Analysis – Reversing Custom Functionality

We will review the code after the call to CreateNamedPipeW() starting where we left off within part three . Quickly reviewing the graph view of the AAA_namedPipe_startup() function, the red arrow points to the code block where CreateNamedPipeW() is located. This graph can be seen in Figure 1 below.

Static Analysis
Figure 1 – CreateNamedPipeW() code block

After the valid creation of a named pipe server, the next stage is to identify where the “HANDLE” object associated with the specific \\.\pipe\NinjaReally named pipe is utilized. According to documentation, the next function we should be looking for is ConnectNamedPipe(). This function accepts two arguments, the first is the HANDLE that was returned from CreateNamedPipeW() earlier, and the second argument is potentially either a LPOVERLAPPED, which is a valid pointer to an OVERLAPPED structure, or a null pointer. The presence of ConnectNamedPipe() is seen in Figure 2 below.

ConnectNamedPipe Code Path
Figure 2 – ConnectNamedPipe Code Path

Reviewing the code within Figure 2, we notice that two different ConnectNamedPipe() function calls exist. The code block on the left is the valid code path that we are after. The code block on the right leads down a path related to debugging and testing code to make sure that the server code is properly implemented.

The ConnectNamedPipe() function will continuously run until either, an error is encountered or a client successfully connects to the named pipe server. Immediately after this call, a check is made to determine if a client successfully connected, and this is implemented by checking the return value from ConnectNamedPipe().

If everything is executed successfully this check will fail because the return value (EAX) does not equal zero, and the code diverts to the next code block at address offset 0x3D90. This code block is responsible for handling the initialization of a local buffer, and it also reads data sent from the client to the named pipe server. The local buffer is initialized with 1023 elements of a 0x00 (null byte) value via a call to memset() at address offset 0x3DDF. Immediately following at address offset 0x3E0E a call to ReadFile() is executed, and the handle to the \\.\pipe\NinjaReally named pipe is passed, and also the buffer initialized via memset() at address offset 0x3DDF, as arguments. An interesting observation to make during the call to ReadFile() is that the nNumberOfBytesToRead argument is one byte less than the total size of the local buffer that holds 1023 bytes. The behavior outlined in this paragraph can be seen in the image below within Figure 3.

ReadFile() Code Path
Figure 3 – ReadFile() Code Path

Immediately after the call to ReadFile(), another check is made to determine if the ReadFile() function is executed successfully. If the return value is nonzero, then code execution is diverted successfully to the next code block at address offset 0x3E09, and another check is detected. This check is responsible for ensuring that the value stored within the lpNumberOfBytesRead argument is not null. This is because sometimes, even if the function ReadFile() executes successfully, a chance still exists where ReadFile() reads zero bytes within the provided buffer. This behavior can be seen in Figure 4 below.

lpNumberOfBytesRead Check
Figure 4 – lpNumberOfBytesRead Check

After that small check, code execution is diverted to the next code block at address offset 0x3E1B. We can quickly identify two print statements at this location and the next interesting instructions begin at address offset 0x3E62. Starting at this location, it is observable that a conditional move if not zero (cmovnz) instruction is executed. The cmovnz instruction relies on the result of the ZF flag value after the previous instruction, cmp (compare), at address offset 0x3E71. The cmp instruction, executes a targeted subtraction of the second operand against the first operand, while altering some of the EFLAGS and, in this specific scenario, the ZF flag. The ZF flag commonly referred to as the Zero Flag is modified during the cmp instruction and is set to 1 if the values being compared are equal, or it is set to 0 if the values are not equal.

In the event of the cmovnz at address offset 0x3E62, a local vector object (var_4B0) has its MyLast and MyFirst fields compared. If they do not match, then a conditional move is executed where the MyLast field is set to the same as the MyFirst field, and code execution will continue to the next interesting code section starting at address offset 0x3E6F. This behavior can be seen in Figure 5 below.

Local Vector Number of Elements Check
Figure 5 – Local Vector Number of Elements Check

Before the next function call, the few instructions include setting up arguments to be passed to a std::vector insert() member function. The local vector object (var_4B0) will be initialized with the client data recorded within the local variable Buffer from the ReadFile() function call. This behavior can be seen in Figure 6 below.

Static Analysis
Figure 6 – Local Vector Initialization

After the insert member function call, instructions relating to the std::vector shrink_to_fit() member function are executed at address offset 0x3EE3 and shortly after, the first function of substance is encountered at address offset 0x3EEC. This function call is in relation to the beginning of client data parsing. This behavior can be seen in Figure 7 below.

Entering into first function relating to parsing client data
Figure 7 – Entering into first function relating to parsing client data

Upon entering the function sub_2D90, one of the first things we should do is give a quick look at the Graph overview window. This window can provide quick insight into a function such as code density, code flow, error paths, and a general idea of how large or small the function is. This window’s output can be seen in Figure 8 below.

Graphview Overview of Parsing Function
Figure 8 – Graphview Overview of Parsing Function

Reviewing the image above, some noticeable additions were manually added. These are two circles (left and right side) and three numbers. Since this is the start of the client packet parsing, it can be safe to assume that potentially the early error branches that lead to the early return of the function are because of a malformed client packet. These error branches are captured within the circle on the left, and the checks are also captured within the numbers one, two, and three. These code blocks are important because it appears that the majority of all the denser code (within the right circle) is avoided if these checks fail.

Starting at the function prologue some local stack variables and buffers are being initialized to null via call’s to memset() and variables being assigned null via the r14 register. The r14 register is set to null via being previously xor’ed against itself at address offset 0x2DDC. After initializing these local variables and buffers, the first check is made at the address offset 0x2E2F.
The first check is comparing the first BYTE that RAX points to, against, a static value of 0x99. The RAX register appears to come from a dereference of RDI at address offset 0x2E0C. RDI is first initialized by a simple local copy made at address offset 0x2D85, which correlates to RCX which is the vector of client data passed into this function as an argument.

Windows Named Pipes Static Analysis – Introduction to C++ Vector Objects

Understanding and covering the entirety of how all kinds of different C++ containers are operated on within assembly is out of scope for this blog post; however, a quick cover of the memory layout for C++ vectors is approachable. Starting, the vector object will be represented as three pointers within memory.

First Member:

  • Pointer to the first element in the Vector

Second Member:

  • Pointer to one element past the end of the total initialized elements in the Vector

Third Member:

  • Pointer to the end of the total capacity for the Vector

By dereferencing the first pointer within the Vector object, we are essentially operating on the First Member which is a pointer to the first element in the client data packet. This pointer is assigned to RAX at address offset 0x2E0C and shortly after, the first byte in this Vector is compared against 0x99 (within blog post number three in this series, we covered how we determined this was a vector of uint8_t elements). If the check passes then code execution is diverted to the next code block at address offset 0x2E7A; however, if the check fails then the jump is not taken, and the function exits prematurely. This behavior is seen in Figure 9 below.

Checking first element within Vector
Figure 9 – Checking first element within Vector

Static Analysis – Identifying Usage of Second Member Field within Vector Objects

The next two code blocks appear to be another check; however, reviewing a string referenced by aServerChecking_0; it appears that these two code blocks are in reference to checking the overall length of the client packet. This is seen in Figure 10 below.

String Information regarding Client Packet Length Check
Figure 10 – String Information regarding Client Packet Length Check

At address offset 0x2E9E, another operation involving the local vector is executed. However, instead of focusing on the first member of the vector object, the second member is operated on. As mentioned earlier, the second member is a pointer to one element past the end of total initialized elements within the local vector object. After the second member pointer is moved into RAX a subsequent SUB (subtraction) instruction is executed where the second member (the address of one element past the end of initialized elements within the vector object) is subtracted from the first member (the address of the first element within the vector object). An important note is that this behavior is commonly related to when the member function size() is executed on a vector object. This member function executes the following operation at a high level:

  • Second member – First member =

    • total size of elements currently initialized within the vector object

This resulting operation is then stored within RAX, and a CMP instruction is executed utilizing this value at address offset 0x2EAC. If the result of the size() function returns a value either equal to 3 or above, then the check passes, and code execution diverts to address offset 0x2F2E; however, if the check fails, then the function exits prematurely. This can be seen in Figure 11 below.

Checking Size of Client Packet
Figure 11 – Checking Size of Client Packet

Static Analysis – First Optimization Encountered

After passing the previous check, the next two code blocks executed are at address offsets 0x2F2E and 0x2F35. With some of the optimizations we have already seen thus far, we are about to encounter another compiler optimization. This optimization will reduce functions down to simple instructions that are inline, and no external function calls are made. The first optimization removes a function call in relation to parsing the first byte within the client data’s local vector. This is observed by reviewing the instructions starting at address offset 0x3013. The first member of our local vector (RDI) is dereferenced and stored within RAX. Then at address offset 0x3024, the first element within our local vector is dereferenced, and the byte is stored within the ESI register via the movzx instruction. Finally, at address offset 0x3047, the register SIL is utilized, which coincides with the usage of a byte because SIL (which is the 8 least significant bits of RSI) is used to represent the byte extracted from the first element of our local vector (at this time is held within a pointer stored in RAX). Utilizing the SIL register, a comparison is made against the static value 1, and if these two values are equal, then the check will pass, and we will continue to the next code block at address offset 0x312F. This behavior can be seen within figure 12 below.

Function Optimization Detected
Figure 12 – Function Optimization Detected

The associated source code of the function that this inline assembly represents can be seen in the table below:

parseClientPkt() – Source Code
1    // Extract values from client packet
2    pcpClientType = clientPktVect[0];
3    pcpClientChallengeNumber = clientPktVect[1];
4    pcpClientLength = clientPktVect[2];
5
6    std::cout << "nt[+] Server: Detecting which "Type" of challenge client sent." << std::endl;
7
8  // call Type parsing function.
9     parseTypeField(pcpClientType,   pcpClientParsingStruct);
10    switch (pcpClientParsingStruct.typeFieldParsingResponse) {‘
11        [SNIP]
12    }

Reviewing the code snippet above, the direct source code relation within the annotated assembly snippet below is immediately noticeable.

parseClientPkt() - Assembly
1 .text:0000000000003013                 mov     rax, [rdi]      ; RDI = Our Local Vector
[SNIP]
2 .text:0000000000003024                 movzx   esi, byte ptr [rax] ; Move the first element of our Vector into ESI
3 .text:0000000000003027                 movzx   ebx, byte ptr [rax+1] ; Move the second element of our Vector into ebx for later use
[SNIP]
4 .text:0000000000003047                 cmp     sil, 1          ; Compare first element against 1
5 .text:000000000000304B                 jz      loc_312F

Reviewing the annotated assembly, it is apparent that within the parseClientPkt() – Assembly snippet, lines 1, 2, and 3 directly relate to the lines 2 and 3 within the parseClientPkt() – Source Code snippet. Next, we will look into the function parseTypeField() and we will notice that this entire function call was optimized away by the compiler, and a simple cmp sil, 1 instruction replaced it. The parseTypeFile() function source code can be reviewed in the code snippet below.

parseTypeField() – Source Code
1  const uint8_t ptfLogicChallenges = 0x01;
2  const uint8_t ptfMemoryCorruptionChallenges = 0x02;
3
4  switch (ptfTypeValue) {
5     case ptfLogicChallenges:
6         std::cout << "tt[!] Server: "TYPE" field value recorded is valid." << std::endl;
7         std::cout << "tt[!] Server: Value is related to "LOGIC" challenges." << std::endl;
8         ptfClientParsingStruct.typeFieldParsingResponse = 0x01;
9         return;
10    default:
11        std::cout << "tt[!] Server: "TYPE" field value recorded is NOT valid." << std::endl;
12        std::cout << "tt[!] Server: Value is not related to any known type." << std::endl;
13        ptfClientParsingStruct.typeFieldParsingResponse = 0x03;
14        return;
15  }

So, reviewing the code snippet above, it appears that this function is quite simple as all it does is take in the ptfTypeValue argument and perform a switch case statement where the only actual value being checked is if the ptfTypeValue equals ptfLogicChallenges where ptfLogicChallenges equals 0x01. The other default case results in the function simply returning to parseClientPkt(). So, the compiler decided to optimize this entire function call away and inline the function behavior instead.

Static Analysis – Second Optimization Encountered

Now, back to the static analysis, if the first byte of the client packet (stored within our local vector) is equal to the numerical value of 0x01 then, code execution will divert to the next code block at address offset 0x312F.
Upon entering this code block, it is apparent that another compiler optimization has been made. This time, the optimization has removed the function call to parseChallengeFieldLogic().This functions source code can be found within the code snippet below:

parseChallengeFieldLogic() – Source Code
1 void parseChallengeFieldLogic(uint8_t pcflChallengeValue, clientParsingStruct& pcflClientParsingStruct) {
2 [SNIP]
3 // Variables
4 	const uint8_t pcflWriteFileChallenge = 0x01;
5	const uint8_t pcflDeleteFileChallenge = 0x02;
6	const uint8_t pcflCreateRegistryKeyChallenge = 0x03;
7	const uint8_t pcflCreateRegistryKeyEntryChallenge = 0x04;
8
9	switch (pcflChallengeValue) {
10	case pcflWriteFileChallenge:
11		std::cout << "tt[!] Server: "Challenge" field value recorded is valid." << std::endl;
12		std::cout << "tt[!] Server: Value is related to "Write File" challenge." << std::endl;
13		pcflClientParsingStruct.challengeFieldParsingResponseLogic = 0x01;
14		return;
15	case pcflDeleteFileChallenge:
16		std::cout << "tt[!] Server: "Challenge" field value recorded is valid." << std::endl;
17		std::cout << "tt[!] Server: Value is related to "Delete File" challenge." << std::endl;
18		pcflClientParsingStruct.challengeFieldParsingResponseLogic = 0x02;
19		return;
20	case pcflCreateRegistryKeyChallenge:
21		std::cout << "tt[!] Server: "Challenge" field value recorded is valid." << std::endl;
22		std::cout << "tt[!] Server: Value is related to "Create Registry Key" challenge." << std::endl;
23		pcflClientParsingStruct.challengeFieldParsingResponseLogic = 0x03;
24		return;
25	case pcflCreateRegistryKeyEntryChallenge:
26		std::cout << "tt[!] Server: "Challenge" field value recorded is valid." << std::endl;
27		std::cout << "tt[!] Server: Value is related to "Create Registry Key Entry" challenge." << std::endl;
28		pcflClientParsingStruct.challengeFieldParsingResponseLogic = 0x04;
29		return;
30	default:
31		std::cout << "tt[!] Server: "Challenge" field value recorded is NOT valid." << std::endl;
32		std::cout << "tt[!] Server: Value is related to no known "LOGIC" challenge." << std::endl;
33		pcflClientParsingStruct.challengeFieldParsingResponseLogic = 0x05;
34		return;
35	}
36 }

A quick review of this function provides the insight that it is quite similar to the previous function, parseTypeField(). So, instead of calling into this function to perform a switch statement on the argument pcflChallengeValue, the compiler chooses to optimize this and utilize the two assembly instructions sub and jz to perform this switch statement. The reason for this is because if we look at the value of the local variables:

    • pcflwriteFileChallenge

      • Value -> 0x01
    • pcflDeleteFileChallenge

      • Value -> 0x02
    • pcflCreateRegistryKeyChallenge

      • Value -> 0x03
    • pcflCreateRegistryKeyEntryChallenge

      • Value -> 0x04>

By simply performing a subtraction of the user-supplied byte (pcflChallengeValue) against these four different local variables used as cases for the switch statement, the sub instruction will only set the ZF flag to 1 if the result of the subtraction is zero. So, by utilizing the conditional jump if zero (jz) instruction with the subtraction (sub) instruction, the compiler can create a simulated switch statement within the assembly.

So, reviewing the check within the assembly at address offset 0x3196, we can trace ECX to the previous instruction at address offset 0x3196, where a mov instruction is executed, and ECX is populated with the value that the EBX register currently holds. By tracing where EBX comes from, we will land at the address offset 0x3027. In this previously mentioned address, at this location, a movzx instruction is executed with the EBX register as the destination, and byte ptr [rax+1] as the source. This is important because the RAX register is the first member of our local vector object, which points to the first element within our vector, and by incrementing the pointer by the numerical value of 0x01, we will be reading the second byte in the client packet (local vector).

So, at this moment, we understand that the client packet is stored within a local vector of uint8_t elements and the first two elements within the vector are:

Client Packet -> [0x00: -> Type of Challenge (Logic or Memory Corruption)]
Client Packet+1 -> [0x00: -> Type of Logic Challenge]

Now, returning to the beginning of the optimized switch statement at address offset 0x3193, we subtract the second element within our client packet (which is a single byte) against the value of 1 and storing the results in ECX. The switch statement optimization can be seen in the table below, where four condition jumps are encountered in sequence.

Optimized Switch Statement - Assembly
.text:0000000000003191                 mov    ecx, ebx
.text:0000000000003193                 sub      ecx, 1
.text:0000000000003196                 jz         loc_37FA
.text:000000000000319C                 sub      ecx, 1
.text:000000000000319F                 jz         loc_3509
.text:00000000000031A5                 sub     ecx, 1
.text:00000000000031A8                 jz         loc_33CC
.text:00000000000031AE                 cmp    ecx, 1
.text:00000000000031B1                 mov     rcx, cs:?cout@std@@3V?$basic_ostream@DU?$char_traits@D@std@@@1@A ; a1
.text:00000000000031B8                 jz      loc_329B

We noted that four values, which were 0x01, 0x02, 0x03, and 0x04, were all utilized as cases within the switch statement and the switch statement executed in order from 0x01 to 0x04.

This is important because we notice in the assembly in the snippet above that if the client provided value was 0x04, the first conditional jump would fail, and ECX would equal 0x03. Then the second conditional jump would also fail, causing ECX to equal 0x02, then the third conditional jump would also fail and ECX would finally equal 0x01. The final conditional jump is executed using a cmp (comparison) based conditional jump if equal (jz).

With a hard-coded value of 0x01 on the final comparison, the check will pass if the provided value is 0x04. Otherwise, if it is above 0x04 code execution will divert to an error path, and the application will terminate. If the value is below 0x04 then the check would execute one of the previous conditional jumps instead.

If we wanted to verify that only four possible paths exist, we can look at the code after the final check and see what would happen if the conditional jump failed and the value of ECX was not 0x01. You can see this by reviewing the code block at address offset 0x31BE. This function performs what appears to be some error logging, and finally, it moves the value of 0x03 into AL and jumps to the next code block at address offset 0x397A.

Instead of reviewing this code path further, reviewing the graph outline below in Figure 13 will provide adequate information about how this code path leads directly to this function returning.

Error Code Path
Figure 13 – Error Code Path

Reviewing the image above, we see an error path highlighted in green. This is the direct path that is executed as the default path within the switch case statement that correlates to the parseChallengeFieldType() function within the source code. However, in this case, this is the assembly. This is important because if no source code were available this would provide valuable information about the second byte accepted values within the client packet. With this new information about the accepted values, one would create a mental image of the potential values such as in the image below.

Visualization of Vector Object
Figure 14 – Visualization of Vector Object

Reviewing the image above, it is noticeable that the first three squares labeled 1, 2, and 3 represent our local vector’s member fields in memory. These member fields are all pointers and were discussed earlier in this blog post. By reviewing the previous information about the optimization and switch case layout for the parseChallengeFiledType() function, we can deduce that the four valid values for the second byte within the client packet are:

  • 0x01
  • 0x02
  • 0x03
  • 0x04

Any other value provided will result in the early termination of this function, and no further parsing of the client packet within the named pipe server will happen. This same technique we used to derive the four total possible valid values for the second byte within the protocol can also be applied to the first byte. Reviewing the image below will showcase this new information.

Updated Valid Protocol Values
Figure 15 – Updated Valid Protocol Values

Now, back to the conditional check at address offset 0x3193, we are going to assume that the client sent a value of 0x01 and this conditional check passed, and we enter the next code block at address offset 0x37FA.
This is an interesting code block because, unlike the others so far, this is the first time we will be entering a function that was not optimized away. This can be seen by the function call at address offset 0x384B. Looking at the arguments being passed, we notice a stack pointer, var_c80, is being passed via RCX, the pointer to our local vector object is being passed via RDX, and a null byte via the r8d register. This can be seen in Figure 16 below.

getName() function call
Figure 16 – getName() function call

Static Analysis – Reversing Structures

Reversing Structures – Introduction

Reviewing the instructions immediately after the function call to sub_12E0(), it is apparent that a large array of some data type is being populated via a call to the function memcpy(). The source argument is RAX, which in this case is the return value from “code>sub_12E0(), the destination argument is a pointer to some type of array located on the stack, and finally, the number of bytes being copied is ”0x624”.

Before continuing further, we need to review some documentation that Microsoft provides regarding how functions with a return type of a structure are treated since the compiled binary was created using MSVC.

MSDN – Documentation (x64 Calling Convention | Return Values)
Return values
A scalar return value that can fit into 64 bits, including the __m64 type, is returned through RAX. Non-scalar types including floats, doubles, and vector types such as __m128, __m128i, __m128d are returned in XMM0. The state of unused bits in the value returned in RAX or XMM0 is undefined.
User-defined types can be returned by value from global functions and static member functions. To return a user-defined type by value in RAX, it must have a length of 1, 2, 4, 8, 16, 32, or 64 bits. It must also have no user-defined constructor, destructor, or copy assignment operator. It can have no private or protected non-static data members, and no non-static data members of reference type. It can't have base classes or virtual functions. And, it can only have data members that also meet these requirements. (This definition is essentially the same as a C++03 POD type. Because the definition has changed in the C++11 standard, we don't recommend using std::is_pod for this test.) Otherwise, the caller must allocate memory for the return value and pass a pointer to it as the first argument. The remaining arguments are then shifted one argument to the right. The same pointer must be returned by the callee in RAX.
These examples show how parameters and return values are passed for functions with the specified declarations:
struct Struct1 {
   int j, k, l;    // Struct1 exceeds 64 bits.
};
Struct1 func3(int a, double b, int c, float d);
// Caller allocates memory for Struct1 returned and passes pointer in RCX,
// a in RDX, b in XMM2, c in R9, d pushed on the stack;
// callee returns pointer to Struct1 result in RAX.

So, using the knowledge that functions returning a structure that exceeds 64 bits in size, place the responsibility of creating the structure and allocating adequate memory on the Caller and the Caller passes the structure via RCX before the normal arguments passed to the function. We can observe this behavior happening before the call to the sub_12E0() function and can be reviewed by the image Figure 16.

Armed with this knowledge, we will embark on attempting to reverse an undocumented structure statically and start understanding what the potential members are and what their associated data types are.The technique used here will be purely static analysis; one could use dynamic analysis to perform research into reversing an undocumented structure; however, this entire post is focused on a static analysis approach.

It is also important to note that accurately reversing structures is dependent upon one’s knowledge of how the compiler compiles source code using different compiler flags and optimization levels; these techniques can vary between compilers and the optimization levels. While reversing undocumented structures; understanding how higher-level concepts translate to assembly is also important. I will do my best to cover as much as possible; however, this is not an entire break down and is more of an introduction to reversing structures in this blog post section.

Reversing Structures – What is VAR_C80?

If we look at the stack for the function sub_2D90() using the Stack window page within IDA Pro, it is observable that this function holds two large local variables. This can be seen in the table below.

sub_2D90 – Stack
-0000000000000CA0 ; D/A/*   : change type (data/ascii/array)
-0000000000000CA0 ; N       : rename
-0000000000000CA0 ; U       : undefine
-0000000000000CA0 ; Use data definition commands to create local variables and function arguments.
-0000000000000CA0 ; Two special fields " r" and " s" represent return address and saved registers.
-0000000000000CA0 ; Frame size: CA0; Saved regs: 8; Purge: 0
-0000000000000CA0 ;
-0000000000000CA0
-0000000000000CA0                 db ? ; undefined
-0000000000000C9F                 db ? ; undefined
[SNIP]
-0000000000000C80 var_C80         db ?
-0000000000000C7F                 db ? ; undefined
-0000000000000C7E                 db ? ; undefined
-0000000000000C7D                 db ? ; undefined
[SNIP]
-0000000000000651                 db ? ; undefined
-0000000000000650 pszPath         db 256 dup(?)           ; string(C)
-0000000000000550 var_550         dd ?
-000000000000054C var_54C         dd ?
-0000000000000548                 db ? ; undefined
-0000000000000547                 db ? ; undefined
-0000000000000546                 db ? ; undefined
[SNIP]
-0000000000000354 var_354         dq ?
-000000000000034C var_34C         dd ?
-0000000000000348 var_348         db ?
[SNIP]
-0000000000000048 var_48          dq ?
-0000000000000040 var_40          xmmword ?
-0000000000000030 var_30          dd ?
-000000000000002C                 db ? ; undefined
-000000000000002B                 db ? ; undefined
-000000000000002A                 db ? ; undefined
-0000000000000029                 db ? ; undefined
-0000000000000028                 db ? ; undefined
-0000000000000027                 db ? ; undefined
-0000000000000026                 db ? ; undefined
-0000000000000025                 db ? ; undefined
-0000000000000024                 db ? ; undefined
-0000000000000023                 db ? ; undefined
-0000000000000022                 db ? ; undefined
-0000000000000021                 db ? ; undefined
-0000000000000020 Stack_Cookiez   dq ?
-0000000000000018                 db ? ; undefined
-0000000000000017                 db ? ; undefined
-0000000000000016                 db ? ; undefined
-0000000000000015                 db ? ; undefined
-0000000000000014                 db ? ; undefined
-0000000000000013                 db ? ; undefined
-0000000000000012                 db ? ; undefined
-0000000000000011                 db ? ; undefined
-0000000000000010                 db ? ; undefined
-000000000000000F                 db ? ; undefined
-000000000000000E                 db ? ; undefined
-000000000000000D                 db ? ; undefined
-000000000000000C                 db ? ; undefined
-000000000000000B                 db ? ; undefined
-000000000000000A                 db ? ; undefined
-0000000000000009                 db ? ; undefined
-0000000000000008                 db ? ; undefined
-0000000000000007                 db ? ; undefined
-0000000000000006                 db ? ; undefined
-0000000000000005                 db ? ; undefined
-0000000000000004                 db ? ; undefined
-0000000000000003                 db ? ; undefined
-0000000000000002                 db ? ; undefined
-0000000000000001                 db ? ; undefined
+0000000000000000  s              db 8 dup(?)
+0000000000000008  r              db 8 dup(?)
+0000000000000010                 db ? ; undefined
+0000000000000011                 db ? ; undefined
+0000000000000012                 db ? ; undefined
+0000000000000013                 db ? ; undefined
+0000000000000014                 db ? ; undefined
+0000000000000015                 db ? ; undefined
+0000000000000016                 db ? ; undefined
+0000000000000017                 db ? ; undefined
+0000000000000018 arg_8           dq ?
+0000000000000020 arg_10          dq ?
+0000000000000028
+0000000000000028 ; end of stack variables

Reviewing the table above, the two large local variables are referred to as VAR_C80 and pszPath.
Now, at this exact moment, it is hard to tell exactly what data type these variables are or even if they are structures to begin with. At this time, all we can tell is that starting at offset 0xC80 is a local variable, and the next local variable on the stack is 0x630 bytes away, and that is pszPath at offset 0x650. For simplicity we can assume right now that VAR_C80 is some array that holds 0x630 bytes.

Reversing Structures – Navigating to the Structures Window

Knowing that the VAR_C80 local variable could potentially be a structure of 0x630 bytes in length, we will create a new structure within IDA Pro and assign this new structure to the local variable VAR_C80.
To create a new structure, first, open the Structures window within IDA Pro. This is accomplished either through navigating to the View -> “code>Open subviews menu option and manually selecting the Structures option, or executing the hotkey combination Shift + F9. This behavior can be seen in Figure 17 below.

Opening the Structures Window
Figure 17 – Opening the Structures Window

After opening the Structures window, we will be presented with the current structures that IDA Pro has registered. This can be seen in figure 18 below.

IDA Pro Structures Window
Figure 18 – IDA Pro Structures Window

Reviewing the image above, the important information is within the open and closed brackets ([]), and we will use the _FILTTIME structure as an example for reading.
Starting off, the first row/element within the brackets is an integer, which displays the structure’s total size. The next important row/element is after the COLLAPSED STRUCT/UNION row and this row/element shows the structure’s name. In this event, it easy to tell that the _FILETIME structure is only 0x08 bytes in length.

Reversing Structures – Creating a New Structure

After navigating to the Structures window, we can create a new structure by either navigating to the whitespace at the bottom of the structure list with our cursor and right clicking to bring up a sub menu where we can select the menu option Add struct type… or we can hit the Insert (shorthand named INS) hot key as demonstrated in the figure below.

Add New Structure Menu
Figure 19 – Add New Structure Menu

After selecting the Add struct type… menu option, another new window will appear called create structure/union. Within this window, we will select the text within the text box associated with Structure Name and call this new structure ourStruct. Then select the option ok, and a new structure will now be created within the Structures window. This behavior can be seen in the image below.

Newly Created Structure ourStruct
Figure 20 – Newly Created Structure ourStruct

Reversing Structures – Creating New Member Fields within ourStruct

After creating the initial structure, the next step is to modify our structure and add member fields. As mentioned earlier within the What is VAR_C80 sub section, the VAR_C80 local structure/variable is potentially around 0x630 bytes in length. Knowing the potential size, we will start by adding a blank array of 0x630 bytes within the ourStruct structure.

To accomplish this, select the ourStruct structure (make sure that it is highlighted yellow) and right-click with your mouse. This will bring up a new window full of menu options. These options can be seen in the image below.

Creating New Array Member”
Figure 21 – Creating New Array Member”

Reviewing Figure 21 above, the options such as adding new member are:

  • Data – D
  • String – A
  • Array – Numpad+* (Shift+8)

By selecting the Array… menu option, a new window called convert to array will appear. We will want to select the Array Size field and modify it to be “code>0x630 (1,584) elements. Then select Ok at the bottom. This can be seen in the figure below.

Modifying Number of Array Elements”
Figure 22 – Modifying Number of Array Elements”

After selecting Ok, the ourStruct structure within the Structures window, will be modified to reflect the new addition of the first member field within the structure as seen in the image below.

Newly Created Member Field”
Figure 23 – Newly Created Member Field”

Reversing Structures – Modifying VAR_C80 to be of ourStruct data type

After the creation of the ourStruct structure within the Structures window within IDA Pro. The next step is to modify the local variable var_C80 within the stack of the sub_2D90 function.

To perform this operation, navigate to the start of the sub_2D90 function within either text or graph view (or find a reference to var_C80 within the assembly of the function) and locate var_C80 within the stack information, as seen in the image below.

Local Reference of var_C80 on the stack
Figure 24 – Local Reference of var_C80 on the stack

After finding var_C80, double click with the left mouse button, and IDA Pro will present the associated stack within a new window called stack of , as seen in the image below.

Reviewing sub_2D90 stack
Figure 25 – Reviewing sub_2D90 stack

Next, highlight var_C80 with the mouse pointer and right click. After that, a new window pop-up will appear and select the menu option Struct var… (or execute the hot-key combination, Alt+Q), as seen in the image below.

Modifying var_C80 into a Structure”
Figure 26 – Modifying var_C80 into a Structure”

Upon selecting the Struct var… menu option, a new window will appear, titled Choose a structure (not the current!). Within this window, we will see a list of available structures within our IDA Pro session. At the bottom, we notice the ourStruct structure we created earlier. Please select this option, then click the Ok icon, as seen in the image below.

Selecting ourStruct
Figure 27 – Selecting ourStruct

After selecting OK, the Choose a structure window will disappear, and the var_C80 variable within the sub_2D90 stack window will be modified so that instead of a continuous array of single bytes, it will instead now be represented as an ourStruct structure. This behavior can be seen in the image below.

Modified var_C80 variable”
Figure 28 – Modified var_C80 variable”

Now that we have successfully modified var_C80 to be represented as an ourStruct structure, any modifications we make to the ourStruct structure will be represented in the disassembly where var_C80 member fields may be utilized.

Reversing Structures – Where is VAR_C80 Utilized?

After identifying VAR_C80 on the stack and getting a rough idea of potentially how large of an array it is, the next step is finding where it is first referenced or utilized. The reason for this is because at this moment, all we see is a large single byte array of 0x630 (0xC80 – 0x650) bytes in length. Without the context of how this local structure/variable is being utilized, we will never know the structure’s true layout and its members fields, and their associated data types.

To start, we will cross reference the VAR_C80 variable and see all locations where it’s either being written to or maybe used as an argument within the sub_2D90() function. Through navigating to the start of the function within IDA Pro’s IDA View window, select the variable VAR_C80, and while selected, press the hotkey x on your keyboard. This will open the xrefs to window. This behavior can be seen in the image below.

Reviewing var_C80 Xrefs
Figure 29 – Reviewing var_C80 Xrefs

Reviewing the image above, we identify seven locations within the sub_2D90 function where var_C80 is passed presumably as an argument to a function via a LEA instruction where the destination operand is RCX. This correlates to the information we covered earlier within the Reversing Structures – Introduction section.

It is important to note that although each of these cross-references depicts var_C80 as being utilized as an argument being passed to a function, it is possible that var_C80 could be passed to another local variable (operated on by being stored within another register) and operations performed on that variable instead. This information is important because, we are trying to understand HOW the structures internal member fields are being utilized. Without source code or symbols, structures are commonly treated as just arrays of bytes when reviewing the structures within disassemblers. Depending on how each disassembler performs their post-analysis, they might be able to provide insight into internal member fields.

Looking at an example of usage at address sub_2D90+AB3, we will notice that all the other cross references share an exact same template (template in this manner is being used generically and refers to the overall layout of the instructions surrounding this address).

Generic Usage of var_C80
Figure 30 – Generic Usage of var_C80

Reviewing the image above, we notice a similar pattern between this single use case and the other six use cases. The pattern is that we have a function call where var_C80 is passed via RCX, then immediately after the function call, a memcpy() call is made where the return value is copied into the local variable pszPath, of a total size of 0x624 bytes.

Reversing Structures – Why such little usage of VAR_C80 within sub_2D90

An invested reader might have noticed that we recorded little usage of the var_C80 variable within the sub_2D90 function. This can be for many reasons; however, in this example the reason relates directly to the information covered within the Reversing Structures – Introduction section; regarding how MSVC handles functions that return a structure value larger than 64bits. We should look within the functions being called where var_C80 is being passed as an argument with this information.

Reversing Structures – Digging into sub_12E0

Although we are diving into functionality relating to reversing structures within these sections of the blog post, we will also be reversing sub_12E0 in its entirety as an example.
Now, by performing the cross-reference check within the Where is VAR_C80 Utilized? section of this blog post, it is apparent that two functions exist where var_C80 is being passed as an argument. These two functions are, sub_12E0 and sub_1760. Starting at the address offset 0x384B, a function call to sub_12E0 is executed. Upon entering this function, we will be tracing usage of RCX, which correlates to var_C80, RDX, which correlates to our local vector object (client packet), and r8 within the caller function (sub_2D90).

Digging into sub_12E0 – Initial Impressions

Upon entering the sub_12E0 function, the first operation we will perform is to glance at the overall layout of the function by reviewing the graph seen in the image below.

sub_12E0 Graph
Figure 31 – sub_12E0 Graph

Initial observations of the function, provide insight such as:

  • Two potential code paths that are similar in behavior
  • Total of five checks are implemented per path
  • The overall function is not all that complicated upon first glance

With these observations made the first step is to begin analysis with the first code block of the sub_12E0 function.

Digging into sub_12E0 – First Code Block Overview

Reviewing the first code block at address offset 0x12E0, the first few instructions are related to the function prologue. Next we notice some calls to the memset() function, then finally a check is made against the static value of 0x01. This can all be seen in the image below.

sub_12E0’s initial code block
Figure 32 – sub_12E0’s initial code block

Digging into sub_12E0 – Identifying the First Check

Reviewing the image above, at address offset 0x12FC, a mov instruction is executed, and the value that r8d holds is placed into the EBX register. Then at the end of this code block at address offset 0x139D, EBX is compared against the static value 0x01 and if the two values do not equal each other, the check passes; however, if they do equal each other, the check fails.

The r8d register coincides with the third argument passed to the sub_12E0 function within the caller function at address offset 0x3840, where it is nulled out, as seen in the table below.

Sub_29D0 – r8d argument
.text:0000000000003840                 xor     r8d, r8d        ; static boolean value of False argument #2
.text:0000000000003843                 lea     rcx, [rsp+20h]  ; void *
.text:0000000000003848                 mov     rdx, rdi        ; clientPktVector argument #1
.text:000000000000384B                 call    sub_12E0        ; our call to getName()

Digging into sub_12E0 – First initialization to our structure

After the assignment of r8d to EBX, the next instruction that follows immediately is an assignment for alignment purposes (we will verify the behavior later in this section), at address offset 0x12FF as seen in the table below.

RCX (var_C80)+0xFF Alignment
.text:00000000000012FF                 mov     [rcx+0FFh], al

It is important to note that structures in the assembly are treated as an array of bytes, so with RCX equaling the very first byte in our array, if we wish to access a member at x offset, we must add that value to the base address of the structure. An example could be:

  • RCX Base Address of Structure = 0x00004100
  • We wish to access a char pointer located at offset 0x22 into the structure
  • We would operate on RCX+0x22
  • Char Pointer Address = 0x00004122

Updating ‘ourStruct’ Structure

After seeing that at offset 0xFF into our structure, a null byte is initialized, we can go ahead and update ‘ourStruct’ structure with this new information. To update ourStruct, please select the menu option View -> Open subviews -> Structures, this will open up the Structures window within IDA Pro.

Upon navigating to the Structures window, we will be looking at modifying the member fields within our ourStruct structure. To update the members field, please highlight the first field field_0 and right click, then select Array…. This should open a new window called Convert to array as seen in the image below.

Static Analysis
Figure 33 – Modifying ourStruct Structure

Reviewing the information above, we are modifying the first array to be of size 0xFF (255) because we know that at offset 0xFF into the structure, we initialize a value to Null at address offset 0x12FF within the sub_12E0 function. The new ourStruct structure will represent a single array of size 0xFF as seen in the image below.

Static Analysis
Figure 34 – ourStruct Structure after update of First Field

Reviewing the image above, we notice that the ourStruct structure is no longer 0x630 bytes in length; however, it is now 0x100 in total length.

The total length has been modified because we operated on that single member field, and at the beginning starting out, we are not aware of what each member field really is. So, we created a single member field that was of 0x630 bytes in length. However, now that we are gaining some further context of the type of data within the structure, we start adding and manipulating the member fields within as we gain more understanding of what each member field may represent.
Fixing member fields of ourStruct Structure
Now, after modifying this structure’s single member field, we must add the remaining 0x531 bytes starting at offset 0x100. To create a member field at that offset, please highlight offset 0x100, then right click and select the option Array… as seen in the image below.

Static Analysis
Figure 35 – Creating Second Member Field within ourStruct

This will open the Convert to array window, and we will create a second member field with a total size of 0x531 (1,329) as seen in the image below.

Static Analysis

Figure 36 – Second Member Field Size within ourStruct

After selecting OK, we will notice that the newly create byte array starting at offset 0x100 is of size 0x531; however, a new issue has appeared. The total size of ourStruct structure is now larger than the original 0x630 bytes. So, we need to account for the 0x04 byte alignment we manually added earlier to ourStruct and modify the newly added member field field_100 to be of size 0x52D (1,325), and this will fix the alignment issue. We are using an alignment of 0x04 because Visual Studio 2019 uses a default structure member alignment option within the project’s properties. If you wish to modify the alignment used manually, you can directly modify this option, which correlates to the flag Zp.

Now, the newly created ourStruct structure should be like the image below.

Static Analysis
Figure 37 – Second Member Field Added within ourStruct

Recap of ourStruct structure insights

At this moment, the ourStruct structure is starting to show some interesting values. We have two large single byte arrays where one is of size 0xFF and another is 0x52D. We also know that at offset 0xFF into ourStruct, we have a single byte of value null, and this appears to be related to potential alignment behavior executed by the compiler via the default value of the Zp compiler flag. Right now, we can take few guesses as to what field_0 may be; however, to verify this assumption, we will have to investigate how this member field is operated on within the binary itself. This concept applies to both fields within our structure.

Digging into sub_12E0 – Local Vector & First memset()

So, at this point, we know that at offset 0xFF into our structure (var_C80) within RCX, a value is initialized with a null value. After this, at address offset 0x1305, our local vector object (client packet) is placed into the RSI register. The local vector object is only used once within the first code block, which is near the end, starting at address offset 0x1393.

The reason we know this is our vector object is because RDX is assigned via RDI at address 0x3848 within the caller function, sub_2D90. The RDI register at that moment holds our vector object (client packet).

Next, we encounter the first memset() function, where 0x1F4 bytes with the value of null are written, starting at offset 0x108 into our structure (var_C80) within RCX. This behavior can be seen in the table below.

First Memset() – var_C80
.text:000000000000130B                 xor     edx, edx        ; Val
.text:000000000000130D                 add     rcx, 108h       ; void *
.text:0000000000001314                 mov     r8d, 1F4h       ; Size
.text:000000000000131A                 call    memset

Local Vector & First memset()

Looking at this first memset() function call, it is apparent that at offset 0x100, we have some type of value(s) that take up 8 bytes within ourStruct structure. At offset 0x108, we initialize a larger member field of size 0x1F4 (500) bytes. You can see a general layout of the new information in the table below, where the associated pseudo source is also available as a reference.

Static Analysis
Example 1 – Updated ourStruct + Pseudo Source

At this moment, we are not exactly sure what any of these member fields are or if these are even the right data types. A common theme so far is that we are modifying the structure as we go, and as we uncover my information about HOW the member fields are utilized, we will begin to understand what each member field is and its purpose.

Digging into sub_12E0 – Second memset() & Further initializations

The second memset() function call is at address offset 0x133F; however, unlike the previous section, we will also address two more single byte initialization within the ourStruct structure. The first two single byte initialization’s are at address offset’s 0x2FC and 0x304. The memset() function call is for 0x201 bytes with a value of null, starting at offset 0x407 into ourStruct structure. The assembly responsible for this behavior can be seen in the table below.

Second Memset & Initializations
.text:000000000000131F                 xor     r14d, r14d
.text:0000000000001322                 lea     rcx, [rdi+407h] ; void *
.text:0000000000001329                 xor     edx, edx        ; Val
.text:000000000000132B                 mov     [rdi+2FCh], r14 ;
.text:0000000000001332                 mov     r8d, 201h       ; Size
.text:0000000000001338                 mov     [rdi+304h], r14d
.text:000000000000133F                 call    memset          ;

These assembly instructions modify our ourStruct structure in a specific manner and their results can be seen in the table below with the updated structure.

Static Analysis
Example 2 – Updated ourStruct + Pseudo Source

Now, reviewing and maintaining a mental image of our structure’s layout would be difficult; so, I suggest keeping a notepad available to keep track of the initializations so far. It is easy to become confused during initial reversing when first starting; after a while, one will become more efficient though.

Digging into sub_12E0 – Third memset & further initializations

Entering the third memset() function call at address offset 0x1364, three single byte assignments are executed using the r14 register, which has previously been xor’ed to null at address offset 0x131F. Along with these three single null byte initializations, the memset() function call creates a 0xFF (255) byte array starting at offset 0x00 of ourStruct structure. The assembly responsible for this behavior can be seen in the table below.

Third Memset & Initializations
.text:0000000000001344                 xor     edx, edx        ; Val
.text:0000000000001346                 mov     [rdi+608h], r14
.text:000000000000134D                 mov     r8d, 0FFh       ; Size
.text:0000000000001353                 mov     [rdi+614h], r14
.text:000000000000135A                 mov     rcx, rdi        ; void *
.text:000000000000135D                 mov     [rdi+61Ch], r14
.text:0000000000001364                 call    memset

These assembly instructions modify our ourStruct structure in a specific manner, and you can view their results in the table below with the updated structure.

Updated ourStruct + Pseudo Source
Example 3 – Updated ourStruct + Pseudo Source

Now, at this moment, the pseudo code is becoming even more convoluted; however, I promise, starting out, it is important to do this and document each change manually within the structure. An important note here is that these single byte initializations could be exactly that, a single byte; however, they could also be an integer or a pointer. At this exact moment, we have no context of how these fields are utilized, so we have no real understanding if they are a byte, integer, or pointer.

Digging into sub_12E0 – Final memset & Initializations

Entering the last memset() function call within the first code block, executed at address offset 0x1380, it is observed that an array of size 0xFF with a value of null is initialized at offset 0x308 into the ourStruct structure. Along with the memset, another single byte initialization occurs where at offset 0x100 into ourStruct structure, a null value is written via the EAX register. The assembly responsible for this behavior can be seen in the table below.

Final Memset & Initializations
.text:0000000000001369                 xor     eax, eax
.text:000000000000136B                 lea     rcx, [rdi+308h] ; void *
.text:0000000000001372                 xor     edx, edx        ; Val
.text:0000000000001374                 mov     [rdi+100h], eax
.text:000000000000137A                 mov     r8d, 0FFh       ; Size
.text:0000000000001380                 call    memset

These assembly instructions modify our ourStruct structure in a specific manner and their results can be seen in the table below with the updated structure.

Static Analysis
Example 4 – Updated ourStruct + Pseudo Source

Digging into sub_12E0 – Final Initialization within ourStruct in First Code Block

After the last memset() function call, two more members are initialized within ourStruct structure. These fields are at offsets 0x610 and 0x104. The initialization happens via a xor instruction executed at address offset 0x1385 where the EAX registered is xor’ed against itself, resulting in a null value. The assembly responsible for this behavior can be seen in the table below.

Final Initializations
.text:0000000000001385                 xor     eax, eax
.text:0000000000001387                 mov     [rdi+610h], eax
.text:000000000000138D                 mov     [rdi+104h], eax

These assembly instructions modify our ourStruct structure in a specific manner and their results can be seen in the table below with the updated structure.

Static Analysis
Example 5 – Updated ourStruct + Pseudo Source

Reviewing the newly created structure within the table above, it is observable that it is still hard to read and decipher. Now, we have all the single byte initializations set to a single byte and every other potential value assigned to a byte array. For simplicity of reading, we will convert every single byte member field into a four-byte integer data type. This newly created structure can be seen in the table below.

Static Analysis
Example 6 – Updated ourStruct + Pseudo Source

After some modifications and renaming of member fields within the ourStruct structure, we are finally starting to reach a point where readability of our structure is increasing. We are still a long way from a complete 1:1 source and this approach may be tedious, but it is effective.

Digging into sub_12E0 – A quick look at compiler optimizations and associated behavior

Before we dive into why we initially set all those single byte initializations to a data type of an INT, lets talk about compiler optimizations and its behavior. This section is short and is not an entire in-depth review of compiler internals; however, it should provide readers with adequate initial guidance in the right direction. As mentioned earlier in this blog post, the compiler we are basing our analysis on is MSVC.

Starting out, we must understand that compilers can compile source code into machine code using multiple different levels of optimizations. Each different compiler will behave slightly different when it comes to compiling source into machine code. Along with different compilers having different behavior, different compiler versions may also exhibit slightly different behavior.

Understanding that everything is compiler version dependent (to some degree), we will utilize the online tool Compiler Explorer created by Mat Godbolt. We will review the disassembly of the associated compiled source code in the table below.

Generic Example Source Code
// Type your code here, or load an example.

#include <windows.h>;
#include <stdio.h>

struct dataType {
    BYTE wow = 0;
    short wow2 = 0;
    int wow3 = 0;
};

int main() {
    struct dataType dog = {};
    dog.wow = 1;
    dog.wow2 = 2;
    dog.wow3 = 4;
    printf("Dog.wow = %d", dog.wow2);
    return 1;
}

We will utilize the MSVC compiler and specify a manual compiler flag of Od, to disable all optimization levels. The reason for this is because, if we leave it at default (O2), all the important aspects for learning are optimized away and all we will see is a printf() function call where the second argument is a hardcoded value of 0x02.

Now, reviewing the source code above, we have the main() function defined where a local struct called dog, based on the dataType structure, is created. The dataType structure has three different member fields, each with a data type of BYTE, short, and int. Within main(), the member fields of the dog structure are initialized with 1, 2, and 4.

According to the MSDN documentation, each data type has a specified size (width), meaning that a BYTE data type is represented as a single BYTE commonly represented as either _int8 or unsigned _int8, is only 1 byte in size (width). Each potential data type has its own size, and the three we utilize are:

  • BYTE ([signed/unsigned] _int8) – 1 byte in size (width)
  • SHORT [signed/unsigned] _int16 – 2 byte in size (width)
  • INT ([signed/unsigned] int – 4 bytes in size (width)

Understanding the data type representations is important because the compiler will operate on each of these potential data types using specific instructions.

So, if we take the source code we presented earlier and compile it using the MSVC compiler and the optimization flag set to be disabled (Od), we should be able to identify some assembly relating to specific data type usage. The assembly can be seen in the table below.

Reviewing assembly of main() function
0: dog$ = 32
1: main    PROC
2: $LN3:
3:         sub     rsp, 56                             ; 00000038H
4:         mov     BYTE PTR dog$[rsp], 0
5:         xor     eax, eax
6:         mov     WORD PTR dog$[rsp+2], ax
7:         mov     DWORD PTR dog$[rsp+4], 0
8:         mov     BYTE PTR dog$[rsp], 1
9:         mov     eax, 2
10:        mov     WORD PTR dog$[rsp+2], ax
11:        mov     DWORD PTR dog$[rsp+4], 4
12:        movsx   eax, WORD PTR dog$[rsp+2]
13:        mov     edx, eax
14:        lea     rcx, OFFSET FLAT:$SG92624
15:        call    printf
16:        mov     eax, 1
17:        add     rsp, 56                             ; 00000038H
18:        ret     0
19: main    ENDP

Reviewing the table above, we notice that starting at line 4, we have our first assignment to the dog structure through the usage of the Pointer Directive of BYTE. This signals that the first member within the dog structure, is of data type BYTE.

The next interesting instructions are lines 5 and 6. The first instruction at line 5 is a xor where we null the value of eax. The second instruction at line 6, then moves the lower 16bits of eax, which is represented as ax, into the second member field of dog using the WORD PTR Pointer Directive.

The final interesting instruction executed is at line 7. This instruction uses the DWORD Pointer Directive to initialize the third member field within the dog structure, using four bytes. This corresponds to the INT data type width within the MSVC documentation.

Now, this is important because, when it comes to reversing undocumented structures, all our analysis is based on understanding how the compiler behaves. If all we were presented with were the assembly in the table above, we would be able to recreate the general structure of the dog variable used within the main() function.

However, if we rerun this side example, using maximum optimizations we will notice a quite drastic change between the disassembler output as seen in the table below.

Full Optimizations Enabled – MSVC (O2)
0: main    PROC                                            ; COMDAT
1: $LN4:
2:         sub     rsp, 40                             ; 00000028H
3:         mov     edx, 2
4:         lea     rcx, OFFSET FLAT:`string'
5:         call    printf
6:         mov     eax, 1
7:         add     rsp, 40                             ; 00000028H
8:         ret     0
9: main    ENDP

Reviewing the newly optimized disassembly, created using the compiler flag of O2, it is apparent that a lot of code has been eliminated. This is important because, when reversing binaries, our methodology will change drastically depending on the optimization level utilized during compilation. At this level, we have no insight into direct usage of the dog structure, and instead all we are presented with is a call to printf() where the second argument is 0x02.

Understanding why the compiler eliminated the extra code relating to the initialization of the dog structure is left as an exercise for the reader.

Digging into sub_12E0 – Quick review of our new structure

Now, returning to the ourStruct structure within the named pipe binary application. Hopefully, the decisions to modify all the single byte member fields into integers is clear. Although we modified the structure quite drastically, some member fields are still not 100% accurate in representation.

Some fields are still a mystery because we have not identified associated usage with each member field. All we have covered so far has been the initialization of some arrays and integer-based member fields. To achieve an accurate representation of the structure’s member fields, we will have to identify associated usage with each member field that provides adequate context to understand exactly what each member field should represent regarding data type.

An example of why this analysis takes a lot of time is because if we look at array5 and array7 within the newly created structure. We could split these fields into smaller arrays, turn them into multiple arrays and multiple different data types; the possibilities are honestly endless.
Associated table with the ourStruct structure can be seen below.

Updated ourStruct + Pseudo Source
Example 7 – Updated ourStruct + Pseudo Source

Digging into sub_12E0 – Encountering First Check

Moving forward to the next instructions executed within the first code block of the sub_12E0() function, a familiar construct is presented at address offset 0x1393. This construct relates to the size() member function of vector objects, as documented earlier in this blog post. However, in this specific case, the RSI register is the local vector object passed in via RDX and passed as a reference to RSI at address offset 0x1305, as seen in the table below.

Local Vector Object – Size() member function
.text:0000000000001305                 mov     rsi, rdx        ;
.text:0000000000001308                 mov     rdi, rcx        ;
.text:000000000000130B                 xor     edx, edx        ; Val
.text:000000000000130D                 add     rcx, 108h       ; void *
.text:0000000000001314                 mov     r8d, 1F4h       ; Size
.text:000000000000131A                 call    memset
.text:000000000000131F                 xor     r14d, r14d
.text:0000000000001322                 lea     rcx, [rdi+407h] ; void *
.text:0000000000001329                 xor     edx, edx        ; Val
.text:000000000000132B                 mov     [rdi+2FCh], r14 ;
.text:0000000000001332                 mov     r8d, 201h       ; Size
.text:0000000000001338                 mov     [rdi+304h], r14d
.text:000000000000133F                 call    memset          ;
.text:0000000000001344                 xor     edx, edx        ; Val
.text:0000000000001346                 mov     [rdi+608h], r14
.text:000000000000134D                 mov     r8d, 0FFh       ; Size
.text:0000000000001353                 mov     [rdi+614h], r14
.text:000000000000135A                 mov     rcx, rdi        ; void *
.text:000000000000135D                 mov     [rdi+61Ch], r14
.text:0000000000001364                 call    memset
.text:0000000000001369                 xor     eax, eax
.text:000000000000136B                 lea     rcx, [rdi+308h] ; void *
.text:0000000000001372                 xor     edx, edx        ; Val
.text:0000000000001374                 mov     [rdi+100h], eax
.text:000000000000137A                 mov     r8d, 0FFh       ; Size
.text:0000000000001380                 call    memset
.text:0000000000001385                 xor     eax, eax
.text:0000000000001387                 mov     [rdi+610h], eax
.text:000000000000138D                 mov     [rdi+104h], eax
.text:0000000000001393                 mov     rcx, [rsi]      ; move first member into RCX
.text:0000000000001396                 mov     rax, [rsi+8]    ; move second member into RAX
.text:000000000000139A                 sub     rax, rcx
.text:000000000000139A                 sub     rax, rcx
.text:000000000000139D                 cmp     ebx, 1
.text:00000000000013A0                 jnz     loc_1574

After the size() member function is executed, a check is encountered at address offset 0x139D. The check is based on the third argument passed to sub_12E0 via r8d. The r8d register value is assigned to the EBX register at address offset 0x12FC within the sub_12E0 function. Before entering this function, the caller performs an xor instruction on the r8d register, causing the value to become null.

With a check being performed where code execution continues if the value EBX holds is not equal to 1, and EBX being hard coded to a null value, this check appears always to pass when reaching via this code path. Other code paths within the caller function could modify r8d to hold a value other than null.

Digging into sub_12E0 – The Second Check

After passing the first check at address offset 0x13A0, another check is presented at address offset 0x1574. This check is responsible for ensuring that the total number of elements within our local vector object is greater than 0x03. The exact instructions can be seen in the table below.

Local Vector Object – Size() > 0x03
.text:0000000000001574 loc_1574:                               ; CODE XREF: sub_12E0+C0↑j
.text:0000000000001574                 cmp     rax, 3
.text:0000000000001578                 jbe     loc_16B3

The rax register at this time is the result of the size() member function, executed at address offset 0x139A. If the total number of elements provided is less than or equal to three, code execution will divert into an error branch, and the application will terminate.

Digging into sub_12E0 – The Third Check

Upon entering the next check at address offset 0x157E, the first member field within our vector object is incremented by three bytes forward, and a check is made to make sure that the resulting value is not null. The string at offset 0x03 into our vector object is also used as an argument to a strnlen() function call, executed at address offset 0x158D.

Digging into sub_12E0 – The Strnlen()

The strnlen() function call executes by reading the number of bytes within our vector object until either a max size of 0xFE, or a null byte is reached. Which ever event occurs first, the resulting number of bytes read will be placed into rbx for later use, and a check is made to ensure that the return value is not null. If the strnlen() function executed successfully, code execution diverts to another check at address offset 0x159F.
Before continuing, we should probably update our understanding of the client packet’s current possible layout being sent to the named pipe server. Earlier in this blog post we understood some of the first few elements within the client packet; however, now we understand a potential string up to 0xFE bytes in length may be present starting at offset 0x04 into our client packet. A visual representation can be seen in the image below.

Static Analysis
Figure 38 – Start of string within client packet

Digging into sub_12E0 – Client Packet String > 5 bytes in Length

Starting at address offset 0x159F, another check is present where the total number of bytes read (the return value from strnlen()) must be greater than 0x05 bytes, otherwise code execution will divert into an error, and the application will terminate.
At this moment, we have identified another boundary on the type of packet that the named pipe server expects to receive from a client. A visual representation of this new boundary can be seen in the image below.

Static Analysis
Figure 39 – Min String Length Boundary

Digging into sub_12E0 – Copying String from client packet to ‘ourStruct’ structures first member field

Upon a string that is greater than 0x05 bytes being provided within our client packet, the next code block executed I at address offset 0x01633. This code block is responsible for some interesting behavior such as:

  • Copying client packet string into first member field of ourStruct
  • Updating two other fields within the ourStruct structure

This is important because, as mentioned earlier, context is extremely valuable when it comes to reversing undocumented structures and their associated member fields. Being able to see HOW the member fields are utilized and operated on provided powerful insight during our static analysis.

The associated assembly within this code block at address offset 0x1633, can be seen in the table below.

Memcpy() & Updating ‘outStruct’ member fields
.text:0000000000001633 loc_1633:                               ; CODE XREF: sub_12E0+2C3↑j
.text:0000000000001633                 mov     rdx, [rsi]
.text:0000000000001636                 mov     r8, rbx         ; Size
.text:0000000000001639                 add     rdx, 3          ; Src
.text:000000000000163D                 mov     rcx, rdi        ; void *
.text:0000000000001640                 call    memcpy
.text:0000000000001645                 mov     [rdi+100h], ebx
.text:000000000000164B                 mov     [rdi+104h], r14d
.text:0000000000001652                 jmp     loc_1739

Quickly identifying the memcpy() call, we can see that the destination argument, is the first member within the ‘ourStruct’ structure. The source argument, is the string starting at element four into the client packet (local vector object) and finally the size argument is the return value from the previous strnlen() function call, executed at address offset 0x158D.

Digging into sub_12E0 – Verifying first member field of ‘ourStruct’ is of a char array data type

Circling back to reversing structures, earlier in this blog post, we identified that the first member field within the ‘ourStruct’ structure was possibly an array of 255 bytes. This observation occurred when reviewing a memset() function call, executed at address offset 0x1364. By observing the assembly around locations where the first member field within the ‘ourStruct’ structure is operated on, we can understand the specific data type this member field represents.

Starting at address offset 0x158D, a strnlen() function call is made where a string within the client packet is read, where the max amount of bytes that will be read is set to 0xFE. The next clue that provides us with some more information is at address offset 0x1364. At this address, the first member field within our ‘ourStruct’ structure is used as a destination buffer of a memcpy() function call where the max amount of bytes that can by copied, is equal or less than the total size of the first member fields byte array.

With this information, we can conclude with an educated decision that the first member field is a byte array that holds the client’s string located at the fourth element within the client packet (local vector object). Using this information, we can rename the first member field to a more descriptive name, such as clientPackStringValue for example. An updated ‘ourStruct’ structure can be seen in the image below.

First Member Field Name Updated
Figure 40 – First Member Field Name Updated

Digging into sub_12E0 – Updating two Integer Member Fields within ‘ourStruct’

After updating the first member field within ‘ourStruct’, immediately after the ‘memcpy()’ function call, executed at address offset 0x1640, two other member fields within ‘ourStruct’ are also operated on. The first member field is at offset 0x100 and the second member field is at offset 0x104. These two fields appear to both be integers because r14 is null, because of the xor instruction executed at address offset 0x131F, and the EBX register is an integer of size_t type and is the return value from the strnlen() function call at address offset 0x158D. The assembly relating to this functionality can be seen in the table below.

Memcpy() & Updating ‘outStruct’ member fields
.text:0000000000001633 loc_1633:                               ; CODE XREF: sub_12E0+2C3↑j
.text:0000000000001633                 mov     rdx, [rsi]
.text:0000000000001636                 mov     r8, rbx         ; Size
.text:0000000000001639                 add     rdx, 3          ; Src
.text:000000000000163D                 mov     rcx, rdi        ; void *
.text:0000000000001640                 call    memcpy
.text:0000000000001645                 mov     [rdi+100h], ebx
.text:000000000000164B                 mov     [rdi+104h], r14d
.text:0000000000001652                 jmp     loc_1739

With the knowledge that EBX, holds the total number of bytes within the string element of our client packet (local vector object), we can rename the member field at this offset to clientPacketStringLength. The updated structure can be seen in the image below.

Second Member Field Name Updated
Figure 41 – Second Member Field Name Updated

Looking into the second member field at offset 0x104, since the value assigned is null and we have very little context, we could rename this member field to possibly, potentailReturnValue. However, if we take a second and look at other code paths within the sub_12E0 function where this member field is used, we will identify an interesting pattern. Reviewing the image below, each code block where the member field at offset 0x104 is referenced is highlighted in a box.

All References to ‘ourStruct’ member field at offset ‘0x104’d
Figure 42 – All References to ‘ourStruct’ member field at offset ‘0x104’d

Reviewing the image above, it is apparent that eight code blocks are present where the member field at offset 0x104 of ‘ourStruct’ is referenced. Each of these code blocks also immediately returns into the function epilogue before returning code execution to the caller function. With this information, the third member field, at offset 0x104, within ‘ourStruct’ could actually be a possible return value for the sub_12E0 function. At each code block, a different value ranging from 0x00 to 0x08 is assigned, except for two code blocks where the value is null (passed via r14d).

Understanding that the return value of this function could be stored within the member field for the ‘ourStruct’ structure, it is possible that further in the application life cycle, this value may be checked for a specific return value. For now, we are going to assign the name clientPacketStringRetValue, until we can understand further how this member field is utilized. An updated ‘ourStruct’ structure can be seen in the image below.

Third Member Field Name Updated
Figure 43 – Third Member Field Name Updated

After modification of these last two member fields, code execution diverts to the function epilogue, and the local structure referenced within RDI is passed to RAX (the function return value), and the function returns to the caller (sub_2D90).

Reversing Structures – Returning to sub_2D90

Upon returning to sub_2D90, the return value (stored within RAX) is immediately used as the source buffer during a memcpy() function call. The destination argument is another local variable called pszPath, and the total number of bytes copied is 0x624. This behavior is seen in the table below.

Copying ‘ourStruct’ returned from ‘sub_12E0’ into ‘pszPath’
.text:0000000000003843                 lea     rcx, [rsp+0CA0h+var_C80] ; void *
.text:0000000000003848                 mov     rdx, rdi        ; clientPktVector argument #1
.text:000000000000384B                 call    sub_12E0        ; our call to getName()
.text:0000000000003850                 lea     rcx, [rbp+0BA0h+pszPath] ; void *
.text:0000000000003857                 mov     rdx, rax        ; Src
.text:000000000000385A                 mov     r8d, 624h       ; Size
.text:0000000000003860                 call    memcpy
.text:0000000000003865                 mov     rcx, cs:?cout@std@@3V?$basic_ostream@DU?$char_traits@D@std@@@1@A ; a1
.text:000000000000386C                 cmp     [rbp+0BA0h+var_54C], r14d
.text:0000000000003873                 jnz     loc_3957

Reviewing the behavior of the ‘memcpy()’ function call, it is apparent that pszPath is actually a direct copy of the var_C80 local structure that we have been modifying throughout this post. By modifying the pszPath local variable and assigning it the value of ourStruct, all the local references where an offset from pszPath is updated accordingly using the member fields within the ‘ourStruct’ structure. An example of this behavior can be seen in the table below.

Updated <code>pszPath</code> local variable
.text:000000000000384B                 call    sub_12E0        ; our call to getName()
.text:0000000000003850                 lea     rcx, [rbp+0BA0h+pszPath] ; void *
.text:0000000000003857                 mov     rdx, rax        ; Src
.text:000000000000385A                 mov     r8d, 624h       ; Size
.text:0000000000003860                 call    memcpy
.text:0000000000003865                 mov     rcx, cs:std::ostream std::cout ; a1
.text:000000000000386C                 cmp     [rbp+0BA0h+pszPath.clientPacketStringRetValue], r14d
.text:0000000000003873                 jnz     loc_3957

Reviewing the newly updated pszPath structure within the table above, it is apparent that at address offset 0x386C, a comparison is made between the clientPacketStringRetValue and the value the r14d register currently holds. If these two do not equal each other, code execution diverts to an error, and the application terminates. However, if they do equal each other, code execution diverts to address offset 0x3879.

At this moment, r14d is equal to null, via a xor instruction executed at address offset 0x2DCC. The clientPacketStringRetValue structure member holds a value of 0x00 at this moment, causing this check to fail.

Reversing Structures – Entering next code block – 0x3879

This code block is extremely relatable to the code block at address offset 0x37FA. Nearly identical code exists within the two code blocks, as seen in the table below.

Static Analysis
Figure 43 – Third Member Field Name Updated

Comparing the two code blocks, we see direct usage of var_C80 being passed via RCX, and immediately after returning from sub_[1760/12E0] function, a memcpy() function call is executed where the structure returned via RAX is copied into pszPath structure.

Another similar note is that at the end of the code, a comparison is made where in 0x3879 code block, the array3 member field is utilized instead of clientPacketStringRetValue was within the 0x37FA code block (this is a hint as to which data type array3 represents).
Full reversing of the sub_1760 function is out of scope for this blog post; however, avid readers are encouraged to use the techniques covered through this blog post so far to try reversing the entire function. Modification to the ‘ourStruct’ structure will be encouraged as the behavior between sub_1760 and sub_12E0 are quite similar in nature!

At the end of the code block at address offset 0x3879, a comparison is made where a potential return value stored within array3 is checked against r14d and if these two values are not equal, then code execution diverts, and the application terminates via an error. However, if the two values are equal, then code execution diverts to the next code block, starting at address offset 0x38CE.

Reversing Structures – Entering the ‘WriteFile()’ function

Upon entering the code block at address offset 0x38CE, a function call is made where the ‘pszPath’ structure is passed via the RCX register. It is important to note that at this moment, ‘pszPath’ holds the same information as ‘var_C80’ which both local structures are treated as an ‘ourStruct’ structure. This structure was created manually through the ‘Static Analysis – Reversing Structures sections of this blog post.

An example of the assembly instructions within the code block at address offset 0x38CE can be found in the table below.

Entering into the ‘WriteFile()’ function
.text:00000000000038CE                 lea     rdx, aServerGetvalue ; "t[+] Server: getValue() function succe"...
.text:00000000000038D5                 call    print_
.text:00000000000038DA                 mov     rcx, rax
.text:00000000000038DD                 lea     rdx, std::endl<char,std::char_traits<char>>(std::ostream &)
.text:00000000000038E4                 call    cs:std::ostream::operator<<(std::ostream & (*)(std::ostream &))
.text:00000000000038EA                 lea     rcx, [rbp+0BA0h+pszPath]
.text:00000000000038F1                 call    sub_1BD0
.text:00000000000038F6                 lea     rdx, aServerWriteFil ; "nt[+] Server: write_file() function s"...
.text:00000000000038FD                 mov     rcx, cs:std::ostream std::cout ; a1
.text:0000000000003904                 cmp     dword ptr [rbp+0BA0h+pszPath.array8], r14d
.text:000000000000390B                 jz      short loc_3914

The function sub_1BD0() is the actual function responsible for privileged operations (in this event, writing files using data obtained from parsing the incoming client packet) and is the example code path that we will exploit in this blog post.

Exploitation – Introduction

In this blog post, we covered the following information:

  • How to approach auditing an application using static analysis
  • How to understand code flow within an application
  • Understanding compiler optimizations
  • Understanding how to identify attacker input and trace their usage within an application
  • Reviewed reversing usage of Vector objects.
  • Reviewed reversing undocumented structures

All these topics lead into the second half of this blog post surrounding exploitation.

Overview of the WriteFile() function (sub_1BD0)

Reviewing sub_1BD0, the first task to perform is reviewing the graph view image of the function, using the same techniques as discussed earlier within this blog post. The graph view image overview can be seen in the image below.

Static Analysis
Figure 44 – sub_1BD0 overview

Reviewing the image above, it is apparent that more code exists within the left side. The right side could either be potential error paths or paths that do not require an initial setup of arguments before calling into potential callee functions. In either case, by reviewing the output of the WinGraph32 – Xref from window, we will be presented with the following information:

  • How to approach auditing an application using static analysis
    • CreateFile
    • WriteFile
    • CloseHandle

Since the sub_1BD0 function directly calls into these functions, it appears that functionality relating file operations are present within this function. By reversing this function, we will see if the incoming client packet data extracted from the local vector object, stored within pszPath structure within sub_2D90, is utilized as arguments to these privileged file operation functions.

Reversing sub_1BD0 – First Code Block

Upon entering sub_1BD0 and skipping the function prologue, the first interesting instruction executed is at address offset 0x1BE6 as seen in the table below.

NOPE

At address offsets 0x1BE6, 0x1BF0, and 0x1BF8 the xor instruction is executed.

  • 0x1BE6 – Used to initialize arguments to two different functions
    • At address offset 0x01BF3, the hTemplateFile argument is initialized to null
        • Used as an argument to CreateFileA()
    • At address offset 0x1C13, the NumberOfBytesWritten argument is initialized to null
      • Used as an argument to WriteFile()
  • 0x1BFO – Used to initialize the lpSecurityAttributes argument to the CreateFileA() function
  • 0x1BF8 – Used to initialize the dwShareMode argument to the CreateFileA() function

Noticing that a CreateFileA() function is being executed, we are interested in understanding where the lpFileName argument is coming from. Starting at the reference made at address offset 0x1C10 we notice that the value the RCX register holds is copied into the RBX register. Knowing that the calling conventions for 64-bit binaries on Windows utilize the first argument passed via RCX, we can trace any assignments made to this register.
Tracing RCX assignments within this first code block upon entering sub_1BD0, leads to discovering no assignments being executed. This means that the first argument, lpFileName (passed via RCX), is an argument to the sub_1BD0 function. By cross-referencing locations where the sub_1BDO function is called, we will be able to identify what the value of RCX is at the time sub_1BDO is called. The associated cross reference can be seen in the image below.

VerSprite Windows Named Pipe Static Analysis: Sub12E0
Figure 45 – Cross Refencing usage of sub_1BD0

Reviewing the image above depicting the cross-references to sub_1BD0 function, it appears only a single reference is identified, and that is at offset 0xB61 into the sub_2D90 function. By navigating to this location, we discover that the first member of the pszPath structure is passed. This can be seen in the image below.

VerSprite Windows Named Pipe Static Analysis: Sub12E0
Figure 46 – Identifying the Argument passed to sub_1BD0

Noticing that the pszPath local structure is being passed as the only argument to sub_1BD0 this means that the first argument passed to the function, CreateFileA(), utilizes the string at the fourth element of our client packet (local vector object) which has a max size of 0xFE. Navigating back to the CreateFileA() function call within sub_1BD0 at address offset 0x1C17, we will continue our analysis.

The CreateFileA() function returns a handle, and a check is performed against the handle to make sure that CreateFileA() is executed successfully. If CreateFileA() failed to execute accurately, then the return value is -1 (0xFFFFFFFF), which corresponds to the constant INVALID_HANDLE_VALUE within the handleapi.h source file.

Upon obtaining a handle to the file object relating to the attacker provided string, code execution diverts to the next code block at address offset 0x1C31. Within this code block, a call to WriteFile() is executed where the handle object passed via the RDI register, which corresponds to the handle object returned from the previous CreateFileA() function.

Reviewing the arguments that are passed to WriteFile() and their associated values can seen in the list below:

  • hFile – Assigned via RDI which is the return value from CreateFileA()
  • lpBuffer – Is a byte array member field within the ‘ourStruct’ structure at offset 0x108
    • This structure should have been reversed within the sub_1760
  • Understanding how to identify attacker input and trace their usage within an application
  • nNumberOfBytesToWrite – Is an integer member field within the ‘ourStruct’ structure at offset 0x2FC
  • lpNumberOfBytesWritten – This is null and updated before returning from WriteFile()
  • lpOverlapped – This is null and is not utilized during this function call

Reviewing the information in the above list, it is apparent that two more values from our client packet (local vector object) are being utilized during the call to WriteFile(). The most important arguments in this care are lpBuffer and nNumberOfBytesToWrite (both member fields were left as an assignment for the reader to reverse earlier in this blog post). The lpBuffer argument is the exact data that shall be written to the file referenced by the handle object, and the nNumberOfBytesToWrite tells the function WriteFile(), how many bytes to write from the provided buffer within the lpBuffer argument.

Exploitation – Why is this vulnerable?

The call to WriteFIle() is vulnerable because no checks are made by the privileged system service (in this event the named pipe server) to prevent unauthorized processes from sending arbitrary data to cause the creation of any file anywhere on the disk where normal users may not have adequate permissions to write normally.

Since attackers with fewer privileges can control the complete filename (including the extension) and file data, exploitation should be straight forward since attackers can write potential raw binaries or maybe create a malicious configuration file that normally the attacker would not be able to create.

If the developers intend for the privileged system service to remain accessible by less privileged processes, it is important that they properly implement checks on the incoming data to prevent these types of logic-based vulnerabilities. Some examples of these checks can be seen in the list below.

  • Verifying allowed folders to create files within
  • Controlling the extension of the newly created file
  • Verifying the format of the incoming file data to match a specific format before writing to disk

These are just examples and is not an exhaustive list of every possible solution.

Exploitation – Bringing Everything Together

Utilizing all the knowledge we have gained so far from reversing this custom named pipe server. We can utilize the information covered within the previous three blog posts to understand how to create a named pipe client that can send a malicious packet to the named pipe server to create a file in a restricted folder such as C:WindowsSystem32. A list of the blog posts can be seen below:

Conclusion

This blog post covered quite a bit of material to provide readers with a step-by-step approach to perform static analysis while reversing custom named pipe servers. Many different topics were either covered in-depth or lightly touched on. Some of these topics include vector objects, reversing structures, understanding compiler optimizations, understanding how certain data types are operated on within assembly, how to reverse custom protocols, how to trace attacker input through a binary, and the list goes on.

The previous three blog posts were designed to provide an overall high-level view of how to approach named pipe servers specifically; however, we created this post with the idea that a reader can take the skills learned here and apply them to any binary they encounter in the future. It is important to understand that this post is not a complete, exhaustive reference of everything. Many topics are still left uncovered within this post; however, this post should act as a great entry into building a foundation on which your skills can be built even further.

VerSprite Security Research Team

Maintain awareness regarding unknown threats to your products, technologies, and enterprise networks. Organizations that are willing to take the next step in proactively securing their flagship product or environment can leverage our zero-day vulnerability research offering. Our subscription-based capability provides your organization with immediate access to zero-day vulnerabilities affecting products and software. Learn More →