CVE-2022-21728

Tensorflow ReverseSequence Negative batch_dim Heap OOB Read

• CVSS 8.1
• CWE-125 Out-of-bounds Read
• Memory Corruption
• Remote

Tensorflow is an Open Source Machine Learning Framework. The implementation of shape inference for `ReverseSequence` does not fully validate the value of `batch_dim` and can result in a heap OOB read. There is a check to make sure the value of `batch_dim` does not go over the rank of the input, but there is no check for negative values. Negative dimensions are allowed in some cases to mimic Python's negative indexing (i.e., indexing from the end of the array), however if the value is too negative then the implementation of `Dim` would access elements before the start of an array. The fix will be included in TensorFlow 2.8.0. We will also cherrypick this commit on TensorFlow 2.7.1, TensorFlow 2.6.3, and TensorFlow 2.5.3, as these are also affected and still in supported range.

CVE

CVE-2022-21728

CWE

CWE-125 Out-of-bounds Read

CVSS base score

8.1

Published

2022-02-03

OWASP

A03 Injection

Orthogonal defect classification

Checking

Code defect classification

Missing Check

Category

Memory Corruption

Subcategory

Out-of-Bound Accesses

Accessibility scope

Remote

Impact

Arbitrary Code Execution

Affected component

TensorFlow

Fixed by upgrading

Yes

Solution

Ensure that negative values for batch_dim are properly checked and handled to prevent out-of-bounds access. Upgrade to TensorFlow 2.8.0 and is also being backported to TensorFlow 2.7.1, TensorFlow 2.6.3, and TensorFlow 2.5.3 to cover affected and supported versions

import tensorflow as tf

def reverse_sequence(input_tensor, seq_lengths, batch_dim):
    input_rank = tf.rank(input_tensor)

    if batch_dim < 0:
        if abs(batch_dim) >= input_rank:
            raise ValueError("Absolute value of batch_dim exceeds the rank of the input tensor")
    elif batch_dim >= input_rank:
        raise ValueError("batch_dim exceeds the rank of the input tensor")

    return tf.reverse_sequence(input_tensor, seq_lengths, seq_axis=batch_dim)

input_tensor = tf.constant([[1, 2, 3], [4, 5, 6]])
seq_lengths = tf.constant([2, 3])
batch_dim = -1

try:
    output = reverse_sequence(input_tensor, seq_lengths, batch_dim)
    print(output)
except ValueError as e:
    print(f"Vulnerability triggered: {e}")
except Exception as e:
    print(f"An unexpected error occurred: {e}")

import tensorflow as tf

def reverse_sequence(input_tensor, seq_lengths, seq_axis, batch_axis):
    input_rank = tf.rank(input_tensor).numpy()

    if seq_axis < 0:
        seq_axis += input_rank
    if batch_axis < 0:
        batch_axis += input_rank

    if not (0 <= seq_axis < input_rank):
        raise ValueError("Invalid seq_axis: out of bounds for input tensor rank")
    if not (0 <= batch_axis < input_rank):
        raise ValueError("Invalid batch_axis: out of bounds for input tensor rank")
    if seq_axis == batch_axis:
        raise ValueError("seq_axis and batch_axis must be different")

    return tf.reverse_sequence(input_tensor, seq_lengths, seq_axis=seq_axis, batch_axis=batch_axis)

input_tensor = tf.constant([[1, 2, 3], [4, 5, 6]])
seq_lengths = tf.constant([2, 3])
seq_axis = 1
batch_axis = 0

try:
    output = reverse_sequence(input_tensor, seq_lengths, seq_axis, batch_axis)
    print(output.numpy())
except ValueError as e:
    print(f"Validation error: {e}")
except Exception as e:
    print(f"Unexpected error: {e}")

@misc{vaitp:cve202221728,
  title        = {{
Tensorflow ReverseSequence Negative batch_dim Heap OOB Read}},
  author       = {Bogaerts, Fr\'ed\'eric and Ivaki, Naghmeh and Fonseca, Jos\'e},
  year         = {2022},
  note         = {VAITP Python Vulnerability Dataset, entry CVE-2022-21728},
  howpublished = {\url{https://netpack.pt/vaitp/vulnerability/CVE-2022-21728/}}
}