How to Automate Testing of Medium Trust Code

Question

I would like to write automated tests that run in medium trust and fail if they require full trust.

I am writing a library where some functionality is only available in full trust scenarios and I want to verify that the code I wish to run in medium trust will work fine. If also want to know that if I change a class that requires full trust, that my tests will fail.

I have tried creating another AppDomain and loading the medium trust PolicyLevel, but I always get an error with assembly or its dependency could not be loaded while trying to run the cross AppDomain callback.

Is there a way to pull this off?

UPDATE: Based replies, here is what I have. Note that your class being tested must extend MarshalByRefObject. This is very limiting, but I don't see a way around it.

using System;
using System.Reflection;
using System.Security;
using System.Security.Permissions;
using Xunit;

namespace PartialTrustTest
{
    [Serializable]
    public class ClassUnderTest : MarshalByRefObject
    {
        public void PartialTrustSuccess()
        {
            Console.WriteLine( "partial trust success #1" );
        }

        public void PartialTrustFailure()
        {
            FieldInfo fi = typeof (Int32).GetField( "m_value", BindingFlags.Instance | BindingFlags.NonPublic );
            object value = fi.GetValue( 1 );
            Console.WriteLine( "value: {0}", value );
        }
    }

    public class Test
    {
        [Fact]
        public void MediumTrustWithExternalClass()
        {
            // ClassUnderTest must extend MarshalByRefObject
            var classUnderTest = MediumTrustContext.Create<ClassUnderTest>();

            classUnderTest.PartialTrustSuccess();
            Assert.Throws<FieldAccessException>( classUnderTest.PartialTrustFailure );
        }
    }

    internal static class MediumTrustContext
    {
        public static T Create<T>()
        {
            AppDomain appDomain = CreatePartialTrustDomain();
            var t = (T) appDomain.CreateInstanceAndUnwrap( typeof (T).Assembly.FullName, typeof (T).FullName );
            return t;
        }

        public static AppDomain CreatePartialTrustDomain()
        {
            var setup = new AppDomainSetup {ApplicationBase = AppDomain.CurrentDomain.BaseDirectory};
            var permissions = new PermissionSet( null );
            permissions.AddPermission( new SecurityPermission( SecurityPermissionFlag.Execution ) );
            permissions.AddPermission( new ReflectionPermission( ReflectionPermissionFlag.RestrictedMemberAccess ) );
            return AppDomain.CreateDomain( "Partial Trust AppDomain: " + DateTime.Now.Ticks, null, setup, permissions );
        }
    }
}

Answer 1

I just posted an article titled Partial Trust Testing with xUnit.net. It details the xUnit.net-based framework that we use on the Entity Framework team to exercise code under partial trust.

Here is an example of its usage.

public class SomeTests : MarshalByRefObject
{
    [PartialTrustFact]
    public void Partial_trust_test1()
    {
        // Runs in medium trust
    }
}

// Or...

[PartialTrustFixture]
public class MoreTests : MarshalByRefObject
{
    [Fact]
    public void Another_partial_trust_test()
    {
        // Runs in medium trust
    }
}

Answer 2

Shamelessly stolen from How to Host a Partial Trust Sandbox – #7, but reimplemented (along with a simple test case) in F# just for kicks :-)

open System
open System.Reflection
open System.Security
open System.Security.Permissions
open System.Security.Policy

type Program() =
    inherit System.MarshalByRefObject()
    member x.PartialTrustSuccess() =
        Console.WriteLine("foo")
    member x.PartialTrustFailure() =
        let field = typeof<Int32>.GetField("m_value", BindingFlags.Instance ||| BindingFlags.NonPublic)
        let value = field.GetValue(1)
        Console.WriteLine("value: {0}", value)

[<EntryPoint>]
let main _ =
    let appDomain =
        let setup = AppDomainSetup(ApplicationBase = AppDomain.CurrentDomain.BaseDirectory)
        let permissions = PermissionSet(null)
        permissions.AddPermission(SecurityPermission(SecurityPermissionFlag.Execution)) |> ignore
        permissions.AddPermission(ReflectionPermission(ReflectionPermissionFlag.RestrictedMemberAccess)) |> ignore
        AppDomain.CreateDomain("Partial Trust AppDomain", null, setup, permissions)

    let program = appDomain.CreateInstanceAndUnwrap(
                      typeof<Program>.Assembly.FullName,
                      typeof<Program>.FullName) :?> Program

    program.PartialTrustSuccess()

    try
        program.PartialTrustFailure()
        Console.Error.WriteLine("partial trust test failed")
    with
        | :? FieldAccessException -> ()

    0

And a C# version:

using System;
using System.Reflection;
using System.Security;
using System.Security.Permissions;
using System.Security.Policy;

namespace PartialTrustTest
{
    internal class Program : MarshalByRefObject
    {
        public void PartialTrustSuccess()
        {
            Console.WriteLine("partial trust success #1");
        }

        public void PartialTrustFailure()
        {
            FieldInfo fi = typeof(Int32).GetField("m_value", BindingFlags.Instance | BindingFlags.NonPublic);
            object value = fi.GetValue(1);
            Console.WriteLine("value: {0}", value);
        }

        private static AppDomain CreatePartialTrustDomain()
        {
            AppDomainSetup setup = new AppDomainSetup() { ApplicationBase = AppDomain.CurrentDomain.BaseDirectory };
            PermissionSet permissions = new PermissionSet(null);
            permissions.AddPermission(new SecurityPermission(SecurityPermissionFlag.Execution));
            permissions.AddPermission(new ReflectionPermission(ReflectionPermissionFlag.RestrictedMemberAccess));
            return AppDomain.CreateDomain("Partial Trust AppDomain", null, setup, permissions);
        }

        static void Main(string[] args)
        {
            AppDomain appDomain = CreatePartialTrustDomain();

            Program program = (Program)appDomain.CreateInstanceAndUnwrap(
                typeof(Program).Assembly.FullName,
                typeof(Program).FullName);

            program.PartialTrustSuccess();

            try
            {
                program.PartialTrustFailure();
                Console.Error.WriteLine("!!! partial trust test failed");
            }
            catch (FieldAccessException)
            {
                Console.WriteLine("partial trust success #2");
            }
        }
    }
}

 C:\temp\PartialTrustTest\bin\Debug>PartialTrustTest.exe
 partial trust success #1
 partial trust success #2

Answer 3

I've posted a three-part blog post on unit testing in Medium Trust

I spin up an alternative AppDomain in a similar fashion to some answers here, but take it further by using a MarshalByRefObject to invoke the test method in the other AppDomain, meaning your test classes do not need to implement MarshalByRefObject

Part three (containing links to the other parts) is here http://boxbinary.com/2011/10/how-to-run-a-unit-test-in-medium-trust-with-nunitpart-three-umbraco-framework-testing/

Answer 4

I have never attempted to do this, but in general, to handle assembly load failures from custom AppDomains, you can use the AppDomain.AssemblyResolve event.

Although totally unrelated, here's an example of using AppDomain.AssemblyResolve.

Answer 5

The answer depends on what your code does when someone with medium trust privilege attempts to access a full trust feature. I assume some kind of exception will be thrown.

In that case, write a unit test that runs in medium trust context, attempts to access a full trust feature, and expects the exception to be thrown. If you've never written a test like this, one common way to do it that most testing frameworks will support is this:

testMediumTrustUserWontAccessFeatureX()
{
    // set up the context of your test ...

    try
    {
        accessFullTrustFeature();
        fail("Test failed - Medium trust user can access full trust feature");
    }
    catch( SomeKindOfException e )
    {
        // Success - feature was denied to the untrusted user 
    }
}

If the exception is caught, it means your untrusted user didn't get to access the feature (and the test passes) but if the exception is never caught, the test fails (we expected an exception and didn't get it).

This is java-esque pseudo code but this pattern should work in whatever language you are using assuming it support exception handling.