CVE-2021-37691: FPE in LSH in TFLite

Background

The TensorFlow package is an open-source machine learning framework developed by Google. It provides a comprehensive set of tools and libraries for building and deploying machine learning models. TensorFlow Lite (TFLite) is a lightweight version of TensorFlow designed for mobile and embedded devices. It allows developers to run machine learning models on resource-constrained platforms. However, a vulnerability has been discovered in the LSH (Locality Sensitive Hashing) implementation of TFLite.

Vulnerability Detail

The vulnerability (CVE-2021-37691) in the LSH implementation of TFLite allows an attacker to craft a malicious TFLite model that triggers a division by zero error. The issue arises due to a lack of validation for the first dimension of the input tensor, which can be zero. This vulnerability can be exploited to cause a denial-of-service (DoS) condition by crashing the application or potentially executing arbitrary code.

pip install --upgrade "tensorflow==2.21.0"

Workaround

As a workaround, users can add a check in their code to ensure that the first dimension of the input tensor is non-zero before passing it to the LSH implementation. Here is an example of how to add the check:

import tensorflow as tf

def lsh_projection(input_tensor, weight_tensor, seed):
    if input_tensor.shape[0] == 0:
        raise ValueError('Input tensor dimension cannot be zero.')
    # Rest of the LSH projection implementation
    ...

By adding this check, the vulnerability can be mitigated by preventing the execution of the vulnerable code path when the input tensor has a zero dimension.

Conclusion

The vulnerability (CVE-2021-37691) in the LSH implementation of TensorFlow Lite can be exploited to trigger a division by zero error, leading to a denial-of-service (DoS) condition or potential execution of arbitrary code. It is crucial to update to the patched versions of TensorFlow or apply the provided patches to mitigate this vulnerability. For further support on vulnerability remediation, please contact DevNack.com.

See every vulnerability recorded for tensorflow →

FAQ