1. CAP6135: Malware and Software Vulnerability Analysis Fuzzing Test Example Cliff Zou Spring 2014

2. Objective Explain basic fuzzing with concrete coding example
Explain how the vulnerable code in programming project 2 is derived
Introduce several useful techniques in doing the fuzzing test on project 2

3. Example Code $ fuzzTest-target 200 “what is this?” 2
Example code needs three inputs
Int, string, Int
int inputInteger; /* global variable */
void (*foo)(short);
if (argc != 4){
fprintf(stderr, "fuzzTest needs 3 input parameters: int string int!\n");
exit(0); }
sscanf(argv[1], "%d", &inputInteger);
my_func(inputInteger, argv[2], argv[3]);
Subfun my_func() introduces 3 man-made bugs

4. Bug # 1: Integer Overflow
int my_func(short argLen, char *str, char *divStr) {
int denominator;
float x;
char buf[bufLen];
if (argLen != inputInteger) {
fprintf(stderr, "Bug #1: integer overflow triggered\n");
foo = (void *)0xbfffffff;
foo(argLen); /* trigger illegal instruction fault */
exit(1);
Int variable inputInteger changes to short
Overflow happens when inputInteger>32767
foo() is a function pointer
Give it an arbitrary address will cause illegal memory reference for executing code

5. Bug # 2: buffer Overflow char buf[10]; if (strlen(str) > 10){
fprintf(stderr, "Bug #2: buffer overflow triggered. strlen=%d\n", strlen(str));
strcpy(buf, str); /* trigger segmentation fault or stack smashing */
return 2; /*if overwriting return address, it will cause segmentation fault */ }

6. Bug #3: divide by zero sscanf(divStr, "%d", &denominator);
int denominator; float x;
sscanf(divStr, "%d", &denominator);
if (denominator == 0){
x = argLen / denominator;
fprintf(stderr, "Bug #3: division by zero triggered\n");
foo = (void *)0xbffbffff;
foo(argLen); /* trigger illegal instruction fault */
}else
return 0;

7. Fuzzer Outline Generate inputs (random or follow rules)
firstInt = rand()%50000; secondInt = rand() % 2;
arraySize = rand() % 20;
charArray = (char *) malloc(arraySize);
for (j=0; j< arraySize; j++) charArray[j] = 'A';
charArray[arraySize-1] = NULL;
Generate execution command line
sprintf(buffer, "./fuzzTest-target %d \"%s\" %d\n", firstInt, charArray, secondInt);
free(charArray); /* must free memory for repeat testing! */

8. Fuzzer Outline Execute target code ret = system(buffer);
Obtain target execution exit code
wait(&status);
retCode = WEXITSTATUS(ret);
Check abnormal exit code and record inputs that cause the abnormal
if ( retCode == || retCode ==128+4) /* segmentation fault (11) or illegal (4) */ {
printf("retCode=%d ## Input: firstInt = %d, arraySize = %d, secondInt = %d\n", retCode, firstInt, arraySize, secondInt);
fflush(stdout); /*make sure output is print out immediately ! */ }
Repeat from start in generating inputs

9. List of Unix Signal Number
You can find it at:
The WEXITSTATUS() returns a value that is the signal number that caused the termination plus 128

10. How to Record Fuzzing Result?
When abnormal happens, record down inputs that cause the abnormal
Record the corresponding abnormal message printout by target code
Unix OS I/O definition:
stdin (0), stdout (1), stderr (2)
I/O redirection:
$ Command < data.txt: let stdin get from file (instead of keyboard)
$ Command > output.txt: let stdout redirect to file
$ Command 2> error.txt: let stderr redirect to file
$ Command &> output.txt: let stdout and stderr redirect to file
For our example:
$./fuzzTest100 &> output.txt

11. Program Project 2 Introduction

12. Manual Read Sample.jpg File
To understand the jpeg file format and the project’s ‘sample.format’ description, you need a HEX Editor:
In Unix: use “$hexdump sample.jpg > hex.txt”
Each two-byte value is shown as ‘daff’ where the first byte is ‘ff’ and second byte is ‘da’ !
A bit confusing on the byte order
HexEdit for Win:
This program shows each byte value, so no confusion on big-endian or little-endian stuff.
You can use windows accessories “calculator” to translate between decimal and hexadecimal values
Use ‘programmer’ option in ‘view’ menu

13. One-Round Fuzzing Outline
In our fuzzer, we need to first read sample.jpg into a char buffer array
Then, modify the buffer (randomly or follow some format rules)
Then, write the content of the buffer to test.jpg file.
Then, invoke jpegconv on test.jpg to do fuzz test

14. Read sample.jpg into Buffer
char imageBuf[10000]; /*enough to hold sample.jpg */ int fSize; FILE *fin, *fout; fin = fopen(“./sample.jpg”, "rb"); fout = fopen(“./test.jpg”, “wb"); fseek(fin, 0, SEEK_END); /* set file pointer to the file end */ fSize=ftell(fin); /*get input file size */ fseek(fin, 0, SEEK_SET); /* rewind the pointer to the start of file fin */ fread(imageBuf, 1, fSize, fin); /* read byte stream of the file */ fclose(fin); /* then, modify imageBuf randomly, */ /* or follow jpeg format on the header section*/

15. Jpeg Header Format
Now the ‘sample.jpg’ is in the char array imageBuf[]
Check the ‘sample.format’ for the Jpeg format
For example:
imageBuf[0] = 0xff; imageBuf[1] = 0xd8;  SOI header
imageBuf[158]=0xff; imageBuf[159]=0xc0;  SOF header
imageBuf[609]=0xff; imageBuf[610]=0xda;  SOS header
Simple fuzzing: Mutation-based fuzzing
Only work on Jpeg Header section since all bugs are in here
You may only be able to find a few bugs in this way
Of course, trying millions of inputs may find all bugs if you are lucky
Advanced fuzzing: Protocol-aware fuzzing
Follow the guide in project description, modify format sections step-by-step
Modify different section could trigger different bugs

16. Write fuzzed image to test.jpg
fwrite (imageBuf , 1, fSize, fout );
/* if you modified the image size, then use the new fSize */
fclose (fout);
/* then, invoke jpegconv on test.jpg for testing */
Note that the ‘test.jpg’ will only save the newest fuzzed file if you repeatedly do the above fuzzing!
On the other hand, we cannot save all tested jpg files since most of them do not trigger bugs

17. Save Fuzzed Input That Causes Bug
int status, ret, retCode; int crashNum = 0; char fileName[20]; /* saved fuzzed image file name */ char comBuf[200]; /* save the command line string */ sprintf(comBuf, “./jpegconv -ppm -outfile foo.ppm test.jpg"); ret = system(buffer); wait(&status); retCode = WEXITSTATUS(ret); if (retCode == 139){ /* Segmentation fault for a bug */ crashNum ++; sprintf(fileName, “./crashed-%d.jpg”, crashNum); fout = fopen(fileName, “wb"); fwrite (imageBuf, 1, fSize, fout ); fclose (fout); }

18. Notes Remember, do not save every fuzzed input into image files!
There is no enough disk space for that on Eustis!
You will still have multiple fuzzed images saved for the same bug.
You can find smart way to only save one copy for each bug.
When one or two bugs are repeatedly triggered
Try to modify image on other format sections
Mutate image file in different ways
Change to different values
random, negative, zero, upper-bound…
Change different number of bytes
Consecutive n bytes, randomly picked n bytes, change value of n, ….

19. Unsolved Task How to Match crashed-x.jpg to its bug ID?
Hint: Jpegconv uses stderr to print out “BUG X TRIGGERED”
I will leave this task to you

20. Notes Do not directly copy code in this slide to your code!
The quotation mark has been changed by Word!
How many runs should I do?
No. of fuzzed input files
No. of saved fuzzed image files
In order to not blow your disk space quota in Eustis
No. of different bugs found
Need your code to process stderr message
Your code needs to check if fopen() succeeds or not!

21. Working Environment
You can do this project on Eustis, or any Linux machine you set up
Make sure ‘jpegconv’ works on your computer (see project description)
You can use any programming langrage in Linux for the project
But your code must be able to run under Eustis for project submission in order for me to verify your code!
Eustis support: Perl, Java, C, Python, Sbcl
Your report must explain how I can run your code under Eustis!

22. Last Words
After this detailed explanation and coding, the project should be not too hard
My own mutation-based fuzzer only contains less than 60 lines in C
Find two bugs in 1300 inputs
Protocol-aware fuzzer will be longer