Fix EntryPoints for MoveMemory, CopyMemory, FillMemory, ZeroMemory

[vatsan-madhavan]

Please see https://github.com/dotnet/coreclr/issues/24008 for background.

.NET Framework has special case logic to ensure that MoveMemory, CopyMemory, FillMemory, and ZeroMemory are correctly mapped to RtlMoveMemory, RtlCopyMemory, RtlFillMemory, and RtlZeroMemory respectively.

.NET Core does not have this this special case, so the following P/Invoke definition will fail on .NET Core.

    [DllImport("kernel32.dll", EntryPoint = "CopyMemory", SetLastError = false)]
    public static extern void CopyMemory(IntPtr dest, IntPtr src, uint count);

At least two definitions in PInvoke.Kernel32 are likely to fail in .NET Core - MoveMemory, and CopyMemory. (I did not inspect to see whether FillMemory and ZeroMemory have this bug).

[AArnott]

FillMemory and ZeroMemory are macros that are defined in terms of memset rather than a native exported function from a Win32 DLL. I will fix MoveMemory and CopyMemory though.

Thanks for the report.

[hugmyndakassi]

While it is technically true that these are macros in the header, kernel32.dll has long exported RtlZeroMemory as function forwarder into a function of the same name in ntdll.dll and also provides an exported RtlFillMemory function.

As a side-note, Vanara also relies on these four functions existing in kernel32.dll.

[AArnott]

@hugmyndakassi Are RtlZeroMemory or RtlFillMemory as exported functions documented? I can't find any documentation for them outside of macros. We don't expose undocumented Windows features in this library.

[mpolicki]

I'll just leave a quick note here that I got a System.EntryPointNotFoundException: 'Unable to find an entry point named 'RtlCopyMemory' in DLL 'Kernel32'. in a unit test with a CopyMemory call. The exception went away after I set Processor Architecture for AnyCPU Projects to x64 in Visual Studio Test Explorer settings.

.NET Framework 4.7.2, PInvoke 0.7.104

[AArnott]

Thanks, @mpolicki. To be clear, it sounds like you're saying that x86 processes don't get a kernel32 export of RtlCopyMemory, but x64 processes do. Does that sound about right? If so, can you open a separate issue to track that?

[mpolicki]

@AArnott Yes, that's exactly what seems to be the case. The issue is #587.

[AArnott]

Thank you.