CVE-2021-29566: TensorFlow-CPU - Out-of-bounds Write in Dilation2DBackpropInput

Background

TensorFlow is an end-to-end open source platform for machine learning. It provides a comprehensive ecosystem of tools, libraries, and resources for developing and deploying machine learning models. TensorFlow-CPU is a specific package that focuses on running TensorFlow on CPUs. It is widely used in various industries and research fields for machine learning tasks.

Vulnerability Detail

The vulnerability (CVE-2021-29566) exists in the tf.raw_ops.Dilation2DBackpropInput function of TensorFlow-CPU. By passing invalid arguments to this function, an attacker can trigger an out-of-bounds write operation outside the bounds of heap allocated arrays. The implementation of the function does not properly validate the input arguments before writing to the output array. While the loop indices for out_backprop are bounded by the size of the array, there are no similar guarantees for in_backprop. This vulnerability can be exploited to overwrite memory and potentially lead to arbitrary code execution. The issue has been identified and a fix will be included in TensorFlow 2.5.0. Additionally, the fix will be backported to TensorFlow 2.4.2, TensorFlow 2.3.3, TensorFlow 2.2.3, and TensorFlow 2.1.4, as these versions are also affected and within the supported range.

pip install --upgrade "tensorflow-cpu==2.21.0"

Workaround

As a temporary workaround, you can consider implementing input validation checks in your code before calling the tf.raw_ops.Dilation2DBackpropInput function. By validating the input arguments and ensuring they are within the expected range, you can mitigate the risk of triggering the out-of-bounds write vulnerability. Here's an example of how you can perform input validation in Python:

import tensorflow as tf

def safe_dilation_backprop_input(out_backprop, filter, strides, padding, dilations, input_shape):
    h_out, w_out = out_backprop.shape[1], out_backprop.shape[2]
    h_in_max = input_shape[1] + (filter.shape[0] - 1) * dilations[0] - 2 * padding[0]
    w_in_max = input_shape[2] + (filter.shape[1] - 1) * dilations[1] - 2 * padding[1]
    if h_out > h_in_max or w_out > w_in_max:
        raise ValueError('Invalid output shape')
    return tf.raw_ops.Dilation2DBackpropInput(out_backprop, filter, strides, padding, dilations, input_shape)

Conclusion

The vulnerability (CVE-2021-29566) in TensorFlow-CPU's tf.raw_ops.Dilation2DBackpropInput function allows an attacker to write outside the bounds of heap allocated arrays, potentially leading to arbitrary code execution. It is crucial to update TensorFlow-CPU to the patched versions or upgrade to TensorFlow-CPU 2.5.0 once it is released to mitigate this vulnerability. Implementing input validation checks in your code can also provide temporary protection. For further support and assistance in vulnerability remediation, you can contact DevNack.com.

See every vulnerability recorded for tensorflow-cpu →

FAQ