Missing validation causes denial of service via `UnsortedSegmentJoin`

Impact

The implementation of tf.raw_ops.UnsortedSegmentJoin 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 tensorflow as tf

tf.raw_ops.UnsortedSegmentJoin(
  inputs=tf.constant("this", shape=[12], dtype=tf.string),
  segment_ids=tf.constant(0, shape=[12], dtype=tf.int64),
  num_segments=tf.constant(0, shape=[12], dtype=tf.int64))

The code assumes num_segments is a scalar but there is no validation for this before accessing its value:

const Tensor& num_segments_tensor = context->input(2);
OP_REQUIRES(context, num_segments_tensor.NumElements() != 0,
            errors::InvalidArgument("Number of segments cannot be empty."));
auto num_segments = num_segments_tensor.scalar<NUM_SEGMENTS_TYPE>()();

Patches

We have patched the issue in GitHub commit 13d38a07ce9143e044aa737cfd7bb759d0e9b400.

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.