On Fri, 13 Nov 2026 05:45:05 GMT, Timofei Fedotov <[email protected]> wrote:
> convolveBI() and convolveRaster() entry points read kern[klen - 1] without > requiring a nonempty kernel array. > > In JDK 25 and earlier, new Kernel(0, 0, new float[0]) is accepted by the > constructor. If this reaches the native convolution code, kern[klen - 1] > becomes kern[-1], an out-of-bounds read. > > JDK 26 added dimension validation to the Kernel constructor (JDK-8368729), > but the native methods still read the dimensions and coefficient array > directly from the object's fields. Inconsistent values can also be supplied > through reflection. > > Both methods should reject invalid kernel dimensions and array lengths before > accessing the coefficients. > > --------- > - [x] I confirm that I make this contribution in accordance with the [OpenJDK > Interim AI Policy](https://openjdk.org/legal/ai). I agree that the reflective test does not demonstrate reachability through supported APIs on current master. It exercises the native validation directly. My reason for proposing the check is that these two JNI entry points use the kernel dimensions and coefficient array to perform native memory accesses. The added validation checks their consistency once per call, before pinning the array, and returns the existing fallback result for invalid input. ------------- PR Comment: https://git.openjdk.org/jdk/pull/33887#issuecomment-6909189426
