This is the repository illustrate how we construct the dataset and label the dataset.
Folder "construction" shows some scripts to compile the binaries. "construction\Dockerfile_source2binary" is a Dockerfile for compiling coreutils v8.29 using clang-10 and O0-O3 options. Run "docker build -t image_owner/image_name -f Dockerfile_source2binary ." to build an image containing the source and binary of coreutils.
Dataset will be open-sourced after the acceptance of the paper.
Folder "ground_truth_building" contains the code to automatically label the above dataset. In detail, the code structure is listed as follows:
| dir | file | function |
|---|---|---|
| IDA_pro_scripts | extract_binary_range.py | scripts to extract binary function boundary for IDA 7.0 and lower |
| extract_binary_range_75.py | scripts to extract binary function boundary for IDA 7.5 | |
| extract_debug_information | extract_debug_dump.py | extract the line mapping from .debug_line section in binary using readelf |
| extract_source_information | use_understand_to_extract_entity.py | use understand to extract the source line-to-function mapping. |
| mapping | binary2source_mapping.py | extend the line-mapping with binary address-to-function mapping and source line-to-function mapping to function level mapping. |
| - | binary2source_mapping_using_understand.py | main function to conduct labeling for all binaries and source projects. |
| summary_for_inline_staticstics.py | summary the metrics for all binaries. |
When using the above scripts for dataset labeling, some paths need to be set. ``binary2source_mapping_using_understand.py'' contains several paths including the path of ida, the path of understand python, the path of understand tool, the path of dataset, and paths of scripts. And the running of the scripts requires the install of IDA Pro, understand, readelf and python3. The current version is implemented in Linux, but using it in windows is also feasible.