Missing validation causes `TensorSummaryV2` to crash

Impact

The implementation of tf.raw_ops.TensorSummaryV2 does not fully validate the input arguments. This results in a CHECK-failure which can be used to trigger a denial of service attack:

import numpy as np
import tensorflow as tf

tf.raw_ops.TensorSummaryV2(
  tag=np.array('test'),
  tensor=np.array(3),
  serialized_summary_metadata=tf.io.encode_base64(np.empty((0))))

The code assumes axis is a scalar but there is no validation for this.

    const Tensor& serialized_summary_metadata_tensor = c->input(2);
    // ...
    ParseFromTString(serialized_summary_metadata_tensor.scalar<tstring>()(),
                     v->mutable_metadata());

Patches

We have patched the issue in GitHub commit 290bb05c80c327ed74fae1d089f1001b1e2a4ef7.

The fix will be included in TensorFlow 2.9.0. We will also cherrypick this commit on TensorFlow 2.8.1, TensorFlow 2.7.2, and TensorFlow 2.6.4, as these are also affected and still in supported range.

For more information

Please consult our security guide for more information regarding the security model and how to contact us with issues and questions.

Attribution

This vulnerability has been reported by Neophytos Christou from Secure Systems Lab at Brown University and Hong Jin from Singapore Management University.