Division by zero in TFLite's implementation of `GatherNd`

Summary

TensorFlow is an end-to-end open source platform for machine learning. The reference implementation of the `GatherNd` TFLite operator is vulnerable to a division by zero error(https://github.com/tensorflow/tensorflow/blob/0d45ea1ca641b21b73bcf9c00e0179cda284e7e7/tensorflow/lite/kernels/internal/reference/reference_ops.h#L966). An attacker can craft a model such that `params` input would be an empty tensor. In turn, `params_shape.Dims(.)` would be zero, in at least one dimension. The fix will be included in TensorFlow 2.5.0. We will also cherrypick this commit on TensorFlow 2.4.2, TensorFlow 2.3.3, TensorFlow 2.2.3 and TensorFlow 2.1.4, as these are also affected and still in supported range.

Timeline

1. May 14, 2021, 7:22 PM UTCCVE published
2. Aug 3, 2024, 10:11 PM UTCRecord last updated

Affected versions

• tensorflow tensorflow

  fixed in 2.1.4; From 2.2.0 · fixed in 2.2.3; From 2.3.0 · fixed in 2.3.3; From 2.4.0 · fixed in 2.4.2

• tensorflow-cpuPyPI

  fixed in 2.2.0rc0; From 2.2.0 · fixed in 2.3.0rc0; From 2.3.0 · fixed in 2.3.4; From 2.4.0 · fixed in 2.4.3

• tensorflowPyPI

  fixed in 2.1.4

• tensorflowPyPI

  From 2.2.0 · fixed in 2.2.3

• tensorflowPyPI

  From 2.3.0 · fixed in 2.3.3

• tensorflowPyPI

  From 2.4.0 · fixed in 2.4.2

• tensorflow-cpuPyPI

  fixed in 2.1.4

• tensorflow-cpuPyPI

  From 2.2.0 · fixed in 2.2.3

• tensorflow-cpuPyPI

  From 2.3.0 · fixed in 2.3.3

• tensorflow-cpuPyPI

  From 2.4.0 · fixed in 2.4.2

• tensorflow-gpuPyPI

  fixed in 2.1.4

• tensorflow-gpuPyPI

  From 2.2.0 · fixed in 2.2.3

• tensorflow-gpuPyPI

  From 2.4.0 · fixed in 2.4.2

• tensorflow-gpuPyPI

  From 2.3.0 · fixed in 2.3.3

• tensorflow-gpuPyPI

  fixed in 2.2.0rc0; From 2.2.0 · fixed in 2.3.0rc0; From 2.3.0 · fixed in 2.3.4; From 2.4.0 · fixed in 2.4.3

• tensorflowPyPI

  fixed in 2.2.0rc0; From 2.2.0 · fixed in 2.3.0rc0; From 2.3.0 · fixed in 2.3.4; From 2.4.0 · fixed in 2.4.3

References 2