CVE-2021-29523: 
Python vulnerability analysis and mitigation

TensorFlow, an end-to-end open source platform for machine learning, was found to contain a vulnerability (CVE-2021-29523) that allows attackers to trigger a denial of service via a CHECK-fail in tf.raw_ops.AddManySparseToTensorsMap. The vulnerability was discovered and reported by Yakun Zhang and Ying Wang of Baidu X-Team, and was publicly disclosed on May 14, 2021. This issue affects TensorFlow versions prior to 2.5.0 (GitHub Security).

The vulnerability stems from the implementation taking values specified in sparse_shape as dimensions for the output shape. The TensorShape constructor uses a CHECK operation which triggers when InitDims returns a non-OK status. The issue occurs when adding a dimension from the argument results in overflow. This is a legacy implementation of the constructor, and operations should use BuildTensorShapeBase or AddDimWithStatus to prevent CHECK-failures in the presence of overflows. The vulnerability has been assigned a CVSS v3.1 score of 5.5 (Medium) and a CVSS v2.0 score of 2.1 (Low) (CISA Bulletin).

An attacker can exploit this vulnerability to trigger a denial of service attack through a CHECK-fail in the TensorFlow library. The issue occurs when processing specific input parameters that cause an overflow condition, leading to a program crash (GitHub Security).

The vulnerability can be exploited by providing specially crafted inputs to the tf.raw_ops.AddManySparseToTensorsMap operation. A proof of concept exists that demonstrates the vulnerability using specific tensor values and shapes that trigger the overflow condition (GitHub Security).

The issue has been patched in TensorFlow version 2.5.0. The fix was also backported to 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 was implemented in GitHub commit 69c68ecbb24dff3fa0e46da0d16c821a2dd22d7c (GitHub Security).