versync

An experimental project for transfering symbols across srtripped binary versions using machine learning and other binary similarity techniques.

Planned Single Function Workflow

# 1. Extract ACFG features# 2. Extract ACFG disassembly# 3. Preprocess ACFG data tuple for embeddingcfg_feature_data=extract_cfg_features("path/to/binary", "func_name")
# 4. Generate embeddingcfg_embedding=generate_embedding(cfg_feature_data)

Planned Full Binary Workflow

# symboled binarybin_1_embeddings= {
addr: extract_embedding(cfg_feature_data) foraddr, cfg_feature_dataincfg_func_features("path/to/binary1").items()
}
# stripped binarybin_2_embeddings= {
addr: extract_embedding(cfg_feature_data) foraddr, cfg_feature_dataincfg_func_features("path/to/binary2").items()
}
recover_symbols(bin_1_embeddings, bin_2_embeddings)
# output:# {addr: (symbol, confidence), addr2: (symbol2, confidence2), ...}

TODO

• Fix the ACFG extraction for IDA Pro
• Get the embeddings produced in the test_model function