CVE-2022-21728: TensorFlow - Heap Out-of-Bounds Read in ReverseSequence Shape Inference

Background

TensorFlow is an open-source machine learning framework developed by Google. It provides a comprehensive ecosystem for building and deploying machine learning models. TensorFlow supports various operations and functions for manipulating tensors, which are multi-dimensional arrays. One of these operations is ReverseSequence, which allows reversing the elements of a tensor along a specified dimension. However, the implementation of shape inference for ReverseSequence in TensorFlow has a vulnerability that can lead to a heap out-of-bounds read. This vulnerability arises due to insufficient validation of the batch_dim parameter, allowing negative values that can result in accessing elements before the start of an array.

Vulnerability Detail

The vulnerability in TensorFlow's ReverseSequence shape inference can be exploited by providing a negative value for the batch_dim parameter. While negative dimensions are allowed in some cases to mimic Python's negative indexing, if the value is too negative, it can lead to accessing elements before the start of an array. This can result in a heap out-of-bounds read, potentially leading to information disclosure or a crash. The issue has been assigned CVE-2022-21728 and has a CVSSv3 severity rating of High.

Workaround

As a workaround, users can ensure that the batch_dim parameter passed to ReverseSequence is a non-negative value. By avoiding negative values, the vulnerability can be mitigated. Here's an example of how to check and enforce a non-negative batch_dim value in Python:

import tensorflow as tf

def reverse_sequence_with_non_negative_batch_dim(data, seq_lengths, batch_dim):
    if batch_dim < 0:
        raise ValueError('batch_dim must be non-negative')
    return tf.reverse_sequence(data, seq_lengths, batch_dim)

Conclusion

The vulnerability in TensorFlow's ReverseSequence shape inference has been addressed in the latest versions of TensorFlow. Users are advised to upgrade to the patched versions to ensure their systems are secure. If you need further assistance or support in vulnerability remediation, please contact DevNack.com.

See every vulnerability recorded for tensorflow →

FAQ