CVE-2021-29550: 
Python vulnerability analysis and mitigation

TensorFlow, an end-to-end open source platform for machine learning, was found to contain a vulnerability in the tf.raw_ops.FractionalAvgPool operation (CVE-2021-29550). The vulnerability was discovered and disclosed in May 2021, affecting TensorFlow versions prior to 2.5.0. The issue allows an attacker to trigger a runtime division by zero error, potentially leading to a denial of service condition (TF Advisory).

The vulnerability stems from the implementation of FractionalAvgPool where a divisor quantity is computed by dividing two user-controlled values. The values input_size[i] and pooling_ratio_[i] are controlled via the value.shape() and pooling_ratio arguments. If input_size[i] is smaller than pooling_ratio_[i], the floor operation results in output_size[i] being 0. While there is a DCHECK_GT check, it is a no-op outside of debug mode. The issue manifests when these computed values are passed to GeneratePoolingSequence, where a division operation is performed without proper validation (TF Advisory).

When successfully exploited, this vulnerability can cause a runtime division by zero error, resulting in a denial of service condition. This could affect systems running TensorFlow applications that use the FractionalAvgPool operation (TF Advisory).

The vulnerability can be triggered by crafting specific inputs to the tf.raw_ops.FractionalAvgPool operation. A proof-of-concept exploit involves setting up a tensor with specific shape parameters and pooling ratios that lead to a division by zero condition (TF Advisory).

The issue was patched in TensorFlow 2.5.0. The fix was also backported to TensorFlow versions 2.4.2, 2.3.3, 2.2.3, and 2.1.4. Users are advised to upgrade to these patched versions. The fix involves adding validation checks to prevent pooling ratios from being bigger than input tensor dimensions (TF Patch).