On Wed, 18 Nov 2026 05:18:16 GMT, Timofei Fedotov <[email protected]> wrote:
> I'd prefer to keep the existing SAFE_TO_ALLOC semantics unchanged, since not > all callers require a signed integer bound. Might be an option to have another macro for the cases where we want additional checks ? But maybe @prrace or someone else from the 2d/client group can comment (@avu , @mrserb ?) . > > Btw. what about this multiplication > > https://github.com/openjdk/jdk/blob/master/src/java.desktop/share/native/libsplashscreen/splashscreen_impl.c#L149 > > (stride * height, both int), should we guard/check or cast there too ? > > Agreed. It’s also worth pointing out that screenStride can overflow before we > even get to malloc(), so casting the allocation size alone won’t fully solve > it. I’d rather track that separately, since it’s in the shared rendering code > and not really specific to the JPEG decoder. No problem for me to track that separately. > > Some more are here > > https://github.com/openjdk/jdk/blob/8b5481d43fb8d4a922873a03070a0def806daa83/src/java.desktop/share/native/libsplashscreen/splashscreen_png.c#L115 > > > > https://github.com/openjdk/jdk/blob/8b5481d43fb8d4a922873a03070a0def806daa83/src/java.desktop/share/native/libsplashscreen/splashscreen_png.c#L154 > > > > (those at least have a SAFE_TO_ALLOC check, but considering your comments > > it might be not sufficient) > > Thanks for pointing these out. The rowbytes * height multiplication uses > unsigned operands and is guarded by SAFE_TO_ALLOC, so it does not have the > same signed overflow issue. > > However, the bitmap allocation uses signed integers and the preceding stride > calculation may also overflow. I will extend the fix to cover the > corresponding PNG allocation and cleanup paths soon. Great, thanks! ------------- PR Comment: https://git.openjdk.org/jdk/pull/33915#issuecomment-6973300260 PR Comment: https://git.openjdk.org/jdk/pull/33915#issuecomment-6973312783 PR Comment: https://git.openjdk.org/jdk/pull/33915#issuecomment-6973324773
