[CORE][REFERENCE] Add bounds checking for SpaceToBatch by EgorDuplensky · Pull Request #34667 · openvinotoolkit/openvino

EgorDuplensky · 2026-03-12T14:57:49Z

Tickets:

181018

Copilot

Pull request overview

Adds runtime-side validation for SpaceToBatch padding values to prevent invalid/overflowing padded-shape computations, and extends the template plugin reference tests with negative coverage to ensure invalid paddings throw.

Changes:

Add non-negative and overflow guards when computing padded_shape in SpaceToBatch reference evaluation.
Add new negative (exception-expected) reference tests covering huge and negative padding inputs.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
`src/core/src/op/space_to_batch.cpp`	Adds padding validation and overflow checks during `evaluate()` padded-shape computation.
`src/plugins/template/tests/functional/op_reference/space_to_batch.cpp`	Adds parameterized negative tests asserting invalid padding configurations throw `ov::Exception`.

You can also share your feedback on Copilot code review. Take the survey.

src/core/src/op/space_to_batch.cpp

maxnick · 2026-03-13T09:59:42Z

@mitruska , could you please review the core shape inference changes and tests?

mitruska

Approved as the checks are valid and the output shape logic remains the same, backward compatibility is preserved.

The only suggestion is that those checks could be moved from evaluate to shape_infer.

mitruska · 2026-03-13T11:18:34Z

src/core/shape_inference/include/space_to_batch_shape_inference.hpp

+                const auto padded_dim =
+                    data_shape[idx] + static_cast<TVal>((*pads_begin)[idx]) + static_cast<TVal>((*pads_end)[idx]);


Any reason why the validation has been not extended in the shape_infer function?

The shape_infer is called during op constructor and in SpaceToBatch::evaluate, in case of Const pads and static/bounded dimensions, it could preserve from creation of such op even before execution. Also the checks could be beneficial for plugins where the shape_infer function is shared and called for dynamic case.

Related tests to cover such case are here:

openvino/src/core/tests/type_prop/space_to_batch.cpp

Lines 73 to 85 in 44967ef

TEST(type_prop, space_to_batch_when_space_is_static) {

auto data_shape = PartialShape{{2, 5}, 100, 1024, 3};

set_shape_symbols(data_shape);

auto data = make_shared<ov::op::v0::Parameter>(element::f32, data_shape);

auto block_shape = make_shared<ov::op::v0::Constant>(element::i64, Shape{4}, vector<int64_t>{1, 12, 100, 2});

auto pads_begin = make_shared<ov::op::v0::Constant>(element::i64, Shape{4}, vector<int64_t>{0, 3, 38, 1});

auto pads_end = make_shared<ov::op::v0::Constant>(element::i64, Shape{4}, vector<int64_t>{0, 5, 38, 0});

auto space_to_batch = make_shared<op::v1::SpaceToBatch>(data, block_shape, pads_begin, pads_end);

EXPECT_EQ(

space_to_batch->get_output_partial_shape(0),

(PartialShape{{2 * 12 * 100 * 2, 5 * 12 * 100 * 2}, (100 + 3 + 5) / 12, (1024 + 38 + 38) / 100, (3 + 1) / 2}));

openvino/src/plugins/intel_cpu/tests/unit/shape_inference_test/space_to_batch_shape_inference_test.cpp

Lines 50 to 65 in 44967ef

TEST_F(SpaceToBatchV1StaticShapeInferenceTest, blocks_pads_as_constants) {

const auto data = std::make_shared<Parameter>(element::f32, PartialShape{-1, -1, -1, -1});

const auto block_shape =

std::make_shared<Constant>(element::i64, ov::Shape{4}, std::vector<int64_t>{1, 12, 100, 2});

const auto pads_begin = std::make_shared<Constant>(element::i64, ov::Shape{4}, std::vector<int64_t>{0, 3, 38, 1});

const auto pads_end = std::make_shared<Constant>(element::i64, ov::Shape{4}, std::vector<int64_t>{0, 5, 38, 0});

const auto op = make_op(data, block_shape, pads_begin, pads_end);

input_shapes = {{2, 100, 1024, 3}, {4}, {4}, {4}};

output_shapes = shape_inference(op.get(), input_shapes);

EXPECT_EQ(output_shapes[0],

(StaticShape{2 * 12 * 100 * 2, (100 + 3 + 5) / 12, (1024 + 38 + 38) / 100, (3 + 1) / 2}));

}

The shape infer logic uses Dimension operator+ which handles potential overflows

I have also considered an alternative of moving the checks to a shape infer / validate step.
But I am not sure whether all those CWE detection tools will be satisfied with such approach.
Also, theoretically we can check for a buffer overflow in scope of shape_infer call but still perform some auxiliary computation in scope of 'evaluate' which could lead to some another overflow.
I mean in general shape_infer and evaluate are supposed to perform logically bound steps, but it is impossible to guarantee this.

[CORE][REFERENCE] Add bounds checking for SpaceToBatch

e6aabd1

EgorDuplensky requested review from a team as code owners March 12, 2026 14:57

github-actions bot added category: Core OpenVINO Core (aka ngraph) category: TEMPLATE OpenVINO Template plugin labels Mar 12, 2026

EgorDuplensky assigned yury-intel Mar 12, 2026

EgorDuplensky requested a review from Copilot March 12, 2026 15:49

Copilot started reviewing on behalf of EgorDuplensky March 12, 2026 15:50 View session

Copilot AI reviewed Mar 12, 2026

View reviewed changes

src/core/src/op/space_to_batch.cpp Show resolved Hide resolved

src/core/src/op/space_to_batch.cpp Outdated Show resolved Hide resolved

Check for data_size overflow as well

a853c86

EgorDuplensky assigned maxnick Mar 12, 2026

maxnick assigned mitruska Mar 13, 2026

mitruska approved these changes Mar 13, 2026

View reviewed changes

maxnick approved these changes Mar 13, 2026

View reviewed changes

maxnick added this to the 2026.1 milestone Mar 13, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[CORE][REFERENCE] Add bounds checking for SpaceToBatch#34667

[CORE][REFERENCE] Add bounds checking for SpaceToBatch#34667
EgorDuplensky wants to merge 2 commits intoopenvinotoolkit:masterfrom
EgorDuplensky:cwe_space_to_batch

EgorDuplensky commented Mar 12, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Uh oh!

maxnick commented Mar 13, 2026

Uh oh!

mitruska left a comment

Uh oh!

mitruska Mar 13, 2026

Uh oh!

EgorDuplensky Mar 13, 2026

Uh oh!

EgorDuplensky Mar 13, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

		const auto padded_dim =
		data_shape[idx] + static_cast<TVal>((pads_begin)[idx]) + static_cast<TVal>((pads_end)[idx]);

	TEST(type_prop, space_to_batch_when_space_is_static) {
	auto data_shape = PartialShape{{2, 5}, 100, 1024, 3};
	set_shape_symbols(data_shape);
	auto data = make_shared<ov::op::v0::Parameter>(element::f32, data_shape);
	auto block_shape = make_shared<ov::op::v0::Constant>(element::i64, Shape{4}, vector<int64_t>{1, 12, 100, 2});
	auto pads_begin = make_shared<ov::op::v0::Constant>(element::i64, Shape{4}, vector<int64_t>{0, 3, 38, 1});
	auto pads_end = make_shared<ov::op::v0::Constant>(element::i64, Shape{4}, vector<int64_t>{0, 5, 38, 0});

	auto space_to_batch = make_shared<op::v1::SpaceToBatch>(data, block_shape, pads_begin, pads_end);

	EXPECT_EQ(
	space_to_batch->get_output_partial_shape(0),
	(PartialShape{{2 * 12 * 100 * 2, 5 * 12 * 100 * 2}, (100 + 3 + 5) / 12, (1024 + 38 + 38) / 100, (3 + 1) / 2}));

	TEST_F(SpaceToBatchV1StaticShapeInferenceTest, blocks_pads_as_constants) {
	const auto data = std::make_shared<Parameter>(element::f32, PartialShape{-1, -1, -1, -1});
	const auto block_shape =
	std::make_shared<Constant>(element::i64, ov::Shape{4}, std::vector<int64_t>{1, 12, 100, 2});
	const auto pads_begin = std::make_shared<Constant>(element::i64, ov::Shape{4}, std::vector<int64_t>{0, 3, 38, 1});
	const auto pads_end = std::make_shared<Constant>(element::i64, ov::Shape{4}, std::vector<int64_t>{0, 5, 38, 0});

	const auto op = make_op(data, block_shape, pads_begin, pads_end);

	input_shapes = {{2, 100, 1024, 3}, {4}, {4}, {4}};
	output_shapes = shape_inference(op.get(), input_shapes);

	EXPECT_EQ(output_shapes[0],
	(StaticShape{2 * 12 * 100 * 2, (100 + 3 + 5) / 12, (1024 + 38 + 38) / 100, (3 + 1) / 2}));
	}

Conversation

EgorDuplensky commented Mar 12, 2026

Tickets:

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

maxnick commented Mar 13, 2026

Uh oh!

mitruska left a comment

Choose a reason for hiding this comment

Uh oh!

mitruska Mar 13, 2026

Choose a reason for hiding this comment

Uh oh!

EgorDuplensky Mar 13, 2026

Choose a reason for hiding this comment

Uh oh!

EgorDuplensky Mar 13, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants