qian-feng / gencoding Goto Github PK

We find it difficult to reproduce this work since currently released code is not complete.
Methods such as getSequences() and graph_analysis() is not defined in currently released code.
Hope you can fill in the missing code ;)

In file graph_analysis_ida.py, function def calTransferIns(bl):

def calTransferIns(bl):
	x86_TI = {'jmp':1, 'jz':1, 'jnz':1, 'js':1, 'je':1, 'jne':1, 'jg':1, 'jle':1, 'jge':1, 'ja':1, 'jnc':1, 'call':1}
	mips_TI = {'beq':1, 'bne':1, 'bgtz':1, "bltz":1, "bgez":1, "blez":1, 'j':1, 'jal':1, 'jr':1, 'jalr':1}
	arm_TI = {'MVN':1, "MOV":1}
	calls = {}
	calls.update(x86_TI)
	calls.update(mips_TI)
	start = bl[0]
	end = bl[1]
	invoke_num = 0
	inst_addr = start
	while inst_addr < end:
		opcode = GetMnem(inst_addr)
		re = [v for v in calls if opcode in v]
		if len(re) > 0:
			invoke_num += 1
		inst_addr = NextHead(inst_addr)
	return invoke_num

Here:

re = [v for v in calls if opcode in v]

It‘s supposed to use opcode == v instead of opcode in v.
The current code implementation here will make all the substrings of the strings in the above instructions in the assembly instructions counted.

The content of the .ida file generated by ./raw-feature-extractor/preprocessing_ida.py is quite hard to understand, would you please provide more information about it?

This code implementation has a lot of dead code that is not used for experiments, but it's not important.
I have an important finding that there is a serious error in the code implementation of this article, which will lead to serious problems in the topology of the ACFG extracted by the ACFG feature extraction code.
This problem is caused by the wrong code implemented in the function def getCfg(func, externs_eas, ea_externs): of cfg_constructor.py .

               if ref in control_blocks:
				dst_node = control_blocks[ref]
				if dst_node not in visited:
					visited[dst_node] = len(cfg)
				dst_id = visited[dst_node]
				cfg.add_edge(dst_id, src_id)
				cfg.node[dst_id]['label'] = dst_node

When there are several dst_nodes that is not visited before more than once, this part of the code will label these dst_nodes to the same label. This will have a serious impact on the topology of ACFG.

As the title, I want to know the version of the IDA pro that you used to extract the ACFG. Thank you!

Hi,

Do you have any plans about releasing the codebook generation and feature encoding component?

Thank you.