Interacting with native libraries in .NET Core 3.0

NativeLibrary is a new class in .NET Core 3.0 for interacting with native libraries. In this article, we'll take a closer look.

DllImport

.NET makes it simple to call functions from a native library using DllImport:

[DllImport("mylibrary")]
public static extern int foo();

This code makes available the function foo from the native library mylibrary. This function accepts no arguments and returns an int. .NET takes care of marshaling the argument types. It is possible to use managed types (like strings), which will be automagically marshaled.

When we use this function, .NET Core tries to find mylibrary. It looks in the application folder and in the system library folders. When looking, it tries variations of the name. For example, on Windows, it adds a .dll extension; on Linux, it adds an .so extension. The lookup also takes into account the current platform based on the Runtime IDentifier (RID). An application can include libraries for different runtime identifiers (organized in rid-folders), and the most appropriate library will be used.

The main limitation of DllImport is that the library name and the symbol names are fixed at compile time. In many cases, especially when you're building the library yourself and including it with the application, this limitation isn't an issue.

NativeLibrary

NativeLibrary is a static class with only a couple of methods:

void Free(IntPtr handle)
IntPtr GetExport(IntPtr handle, String name)
IntPtr Load(String libraryPath)
IntPtr Load(String libraryName, Assembly, DllImportSearchPath?)
void SetDllImportResolver(Assembly, DllImportResolver)
IntPtr TryGetExport(IntPtr handle, String name, out IntPtr address)
IntPtr TryLoad(String libraryPath, out IntPtr handle)
IntPtr TryLoad(String libraryName, Assembly, DllImportSearchPath?, out IntPtr handle)

The first thing we can do is control the library we use in the DllImport, by providing a DllImportResolver delegate for our assembly. The DllImportResolver has the following signature:

public delegate IntPtr DllImportResolver(string libraryName, Assembly assembly, DllImportSearchPath? searchPath);

Its arguments provide us with the context of the DllImport and as a return value we must provide an IntPtr for the library. We get this IntPtr using the Load methods. The Load(string) method loads the library at a specific path. The other Load method provides the default DllImport loading logic. Let’s see how we can use this information:

static class Library
{
const string MyLibrary = "mylibrary";

static Library()
{
    NativeLibrary.SetDllImportResolver(typeof(Library).Assembly, ImportResolver);
}

private static IntPtr ImportResolver(string libraryName, Assembly assembly, DllImportSearchPath? searchPath)
{
    IntPtr libHandle = IntPtr.Zero;
    if (libraryName == MyLibrary)
    {
        // Try using the system library 'libmylibrary.so.5'
        NativeLibrary.TryLoad("libmylibrary.so.5", assembly, DllImportSearchPath.System32, out libHandle);
    }
    return libHandle;
}

[DllImport(MyLibrary)]
public static extern int foo();
}

In this example, we register a DllImportResolver for the assembly. In our resolver, we try loading libmylibrary.so.5 from the system libraries. If it fails, we fall back to the default DllImport resolution by returning IntPtr.Zero. This result gives us the usability of DllImport with the flexibility of picking a specific library at runtime.

Another thing we can do using NativeLibrary is directly resolve symbols using GetExport/TryGetExport. Let's look at an example:

class Library : IDisposable
{
private readonly IntPtr _libHandle;
private readonly Func<int> _foo;
private bool _disposed;

public Library()
{
    _libHandle = NativeLibrary.Load("mylibrary", typeof(Library).Assembly, DllImportSearchPath.System32);

    if (NativeLibrary.TryGetExport(_libHandle, "foo", out IntPtr fooHandle))
    {
        _foo = Marshal.GetDelegateForFunctionPointer<Func<int>>(fooHandle);
    }
    else
    {
        _foo = () => { throw new NotSupportedException("'foo' not found"); };
    }
}

~Library()
{
    Dispose(false);
}

public int foo()
{
    ThrowIfDisposed();
    return _foo();
}

public void Dispose()
{
    Dispose(true);
    GC.SuppressFinalize(this);
}

protected virtual void Dispose(bool disposing)
{
    if (!_disposed)
    {
        _disposed = true;
        NativeLibrary.Free(_libHandle);
    }
}

private void ThrowIfDisposed()
{
    if (_disposed)
    {
        ThrowObjectDisposedException();
    }
}

private void ThrowObjectDisposedException()
    => throw new ObjectDisposedException(typeof(Library).FullName);
}

Here we've replaced the DllImport with calls to NativeLibrary.Load, TryGetExport and Marshal.GetDelegateForFunctionPointer. There's more code involved, but in return we now have full control over the library we are using and can detect and use its symbols dynamically.

Conclusion

In this article, you’ve learned about the new NativeLibrary class and how you can use it—instead of the DllImport attribute—when you need more control over the library resolution and the symbols you use.