CVE-2021-29546: TensorFlow - Integer Division by Zero in QuantizedBiasAdd

Background

TensorFlow is an end-to-end open source platform for machine learning. It provides a comprehensive ecosystem for building and deploying machine learning models. TensorFlow is widely used in various industries and research domains due to its flexibility and scalability. However, a vulnerability has been discovered in TensorFlow that can lead to an integer division by zero undefined behavior in the tf.raw_ops.QuantizedBiasAdd function. This vulnerability affects multiple versions of TensorFlow, including 2.1.4, 2.2.3, 2.3.3, and 2.4.2.

Vulnerability Detail

The vulnerability arises from a division operation performed in the Eigen kernel implementation of QuantizedBiasAdd. The kernel divides the number of elements of the smaller input (based on shape) without checking if the divisor is zero. This can result in an integer division by zero, leading to undefined behavior. An attacker can exploit this vulnerability to cause a denial of service or potentially execute arbitrary code with the privileges of the TensorFlow process.

Workaround

As a temporary workaround, you can modify the affected code to include a check for zero divisor before performing the division operation. Here is an example of how you can add the check:

import tensorflow as tf

def safe_quantized_bias_add(x, y):
    if tf.reduce_prod(tf.shape(x)) == 0 or tf.reduce_prod(tf.shape(y)) == 0:
        raise ValueError('Input shape cannot be zero')
    return tf.raw_ops.QuantizedBiasAdd(x, y)

By adding this check, you ensure that the division operation is only performed when the divisor is non-zero.

Conclusion

The identified vulnerability in TensorFlow can lead to an integer division by zero undefined behavior, which can be exploited by attackers. It is crucial to upgrade to the fixed version of TensorFlow (2.5.0) as soon as it becomes available. In the meantime, you can apply the provided workaround to mitigate the risk. For further assistance and support in vulnerability remediation, please contact DevNack.com.

See every vulnerability recorded for tensorflow →

FAQ