Rop Analysis II
Author: David Zimmer
Date: 02.23.13 - 10:01pm
Here is another quick post on ROP analysis.
I didnt really know the best place to begin, so I started where I could.
First I built a quick app that would scan a directory for dlls and load the sections and map the virtual addresses each covered. It also scanned the import table of each dll and built a list of IAT addresses and what they corresponded to. I also added some other stuff to the IAT marker list to include scans of some known gadget addresses and their comments from various public exploits to help identify ROP start offsets.
So with the base info built, I took a rop exploit (MD5: 805538FF200EC714A735EF3BC1FFF1F0) got the shellcode, and extracted the ROP portion into a seperate file for simplicity. One problem is not knowing what alignment the ROP part starts on, so the app will also allow you to specify the start offset. When you scan the file, it will dump it to 32bit numbers and if any of the known markers are found, or if the addresses fit within the sections loaded, then you know you have the right alignment. (1-3 tries required max).
If the address is found to be within an executable section of a loaded dll, then we disassemble the address and extract the opcodes and disasm up until a terminating instruction is encountered.
Once all of the data is collected, then you can have the interface generate a scdbg .scmd file which will poke the utilized opcodes into memory at the necessary addresses, and set the start offset to the entry you specify.
You can download sample files and output here At first wasnt sure what was going on, but a little googling revealed that this sample is using a NtAccessCheckAndAuditAlarm egg hunter
Test 2 successful. Next comes a full rop only shellcode test and possibly adding a handler for int 0x2e to scdbg. Note: if you want to record all of the register values at every instruction, edit the scmd file and set verbosity to two (eg. -vv) I have uploaded the test tool shown above so I dont lose it.