hacnho/tfrecord-features-feature-materialization-dos-poc

TFRecords Features feature materialization DoS PoC

This repository contains a benign security research PoC for a .tfrecords artifact that drives large protobuf materialization during tf.train.Features.FromString(...).

Files:

• control_same_size.tfrecords
• malicious_features_feature_1230000.tfrecords
• reproduce.py

Observed behavior:

• control artifact:
  • parses successfully with one feature entry
• malicious artifact:
  • both files are about 18.6 MB
  • parses successfully with 1,230,000 feature entries
  • increases peak RSS by about 243848 kB
  • takes about 38.47x the control parse time

Tested runtime:

• tensorflow==2.21.0
• clean interpreter: /tmp/tf221clean/bin/python

Public files:

• https://huggingface.co/hacnho/tfrecord-features-feature-materialization-dos-poc/resolve/main/control_same_size.tfrecords
• https://huggingface.co/hacnho/tfrecord-features-feature-materialization-dos-poc/resolve/main/malicious_features_feature_1230000.tfrecords
• https://huggingface.co/hacnho/tfrecord-features-feature-materialization-dos-poc/resolve/main/reproduce.py

Reproduction:

TF_PYTHON_BIN=/tmp/tf221clean/bin/python python3 build_poc.py
TF_PYTHON_BIN=/tmp/tf221clean/bin/python python3 reproduce.py