Skip to content
forked from 0xd4d/dnlib

Fork of 0xd4d/dnlib to allow parsing "stark" assemblies and modules

License

Notifications You must be signed in to change notification settings

stark-lang/stark-dnlib

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

dnlib Build status NuGet

.NET module/assembly reader/writer library

Compiling

v3.0 requires VS2017 (C#7.3) or later to build it. .NET Core SDK 2.1 or later is also required. See below for breaking changes going from 2.1 to 3.0

An older v2.1 branch can be used to build with older VS versions. This branch won't get any new updates.

v3.0 breaking changes

  • VS2017, C# 7.2 is required to compile it
  • It targets .NET Framework 3.5 or later and netstandard 2.0 or later (.NET Framework 2.0 and 3.0 aren't supported)
  • *MetaData* -> *Metadata*
  • IMetaData interface is an abstract class Metadata
  • _32Bit* -> Bit32*
  • IAssemblyResolver only has a Resolve method. The other methods are still implemented by the default assembly resolver (AssemblyResolver)
  • Raw table rows, eg. RawMethodRow
    • They are immutable structs and the methods to read them have been renamed from eg. ReadMethodRow -> TryReadMethodRow
    • An indexer replaces their Read() method
    • The IRawRow interface has been removed
  • The Constant table info (TableInfo) has an extra padding byte column
  • ModuleWriterOptionsBase.Listener is obsolete, use the new event ModuleWriterOptionsBase.WriterEvent instead
  • Module writer events related to the current progress have been removed. Use the new event ModuleWriterOptionsBase.ProgressUpdated instead
  • StrongNameKey, PublicKey, PublicKeyToken are immutable classes
  • RidList is a struct
  • IBinaryReader, IImageStream have been removed and replaced with new classes
    • MemoryImageStream -> ByteArrayDataReaderFactory
      • It has two static factory methods, Create and CreateReader
    • BinaryReaderChunk -> DataReaderChunk
    • To get a reader, call CreateReader on IPEImage, DataReaderFactory, or #Blob stream
    • The reader is a struct called DataReader and it's not disposable
    • The reader has Slice methods to get another reader (replaces the older Create methods)
    • Since the reader is a struct, pass it by reference to methods if its position should be updated when the method returns
    • DataReader.Position is now a uint and not a long so expressions that were long could now be uint and possibly overflow/underflow
      • reader.Position + 0xFFFFFFFF
      • reader.Position + someRandomValue
      • var pos = reader.Position; <-- pos is a uint and not a long
    • DataReader.Position only accepts valid values and will throw (an IOException) if you set it to an invalid position
  • FileOffset is uint, used to be long
  • MethodBodyWriterBase uses ArrayWriter instead of BinaryWriter (all virtual methods)
  • ModuleWriter and NativeModuleWriter use DataWriter instead of BinaryWriter
  • The native module writer now tries to fit the new metadata, method bodies, resources and other data in the old locations. This results in smaller files. It can be disabled by creating your own NativeModuleWriterOptions
  • MetadataOptions' OtherHeaps and OtherHeapsEnd have been removed. Use CustomHeaps, MetadataHeapsAdded and PreserveHeapOrder() instead.
  • Instruction.GetLocal() returns a local if the instruction is a ldloca or ldloca.s instruction (it used to return null)
  • ModuleCreationOptions.PdbImplementation has been removed and replaced with PdbOptions
  • Renamed
    • ITokenCreator -> ITokenProvider
    • MetadataCreator -> MetadataFactory
    • ResourceDataCreator -> ResourceDataFactory
    • FullNameCreator -> FullNameFactory

Examples

All examples use C#, but since it's a .NET library, you can use any .NET language (eg. VB.NET).

See the Examples project for several examples.

Opening a .NET assembly/module

First of all, the important namespaces are dnlib.DotNet and dnlib.DotNet.Emit. dnlib.DotNet.Emit is only needed if you intend to read/write method bodies. All the examples below assume you have the appropriate using statements at the top of each source file:

    using dnlib.DotNet;
    using dnlib.DotNet.Emit;

ModuleDefMD is the class that is created when you open a .NET module. It has several Load() methods that will create a ModuleDefMD instance. If it's not a .NET module/assembly, a BadImageFormatException will be thrown.

Read a .NET module from a file:

    ModuleDefMD module = ModuleDefMD.Load(@"C:\path\to\file.exe");

Read a .NET module from a byte array:

    byte[] data = System.IO.File.ReadAllBytes(@"C:\path\of\file.dll");
    ModuleDefMD module = ModuleDefMD.Load(data);

You can also pass in a Stream instance, an address in memory (HINSTANCE) or even a System.Reflection.Module instance:

    System.Reflection.Module reflectionModule = typeof(void).Module;	// Get mscorlib.dll's module
    ModuleDefMD module = ModuleDefMD.Load(reflectionModule);

To get the assembly, use its Assembly property:

    AssemblyDef asm = module.Assembly;
    Console.WriteLine("Assembly: {0}", asm);

Saving a .NET assembly/module

Use module.Write(). It can save the assembly to a file or a Stream.

    module.Write(@"C:\saved-assembly.dll");

If it's a C++/CLI assembly, you should use NativeWrite()

    module.NativeWrite(@"C:\saved-assembly.dll");

To detect it at runtime, use this code:

    if (module.IsILOnly) {
    	// This assembly has only IL code, and no native code (eg. it's a C# or VB assembly)
    	module.Write(@"C:\saved-assembly.dll");
    }
    else {
    	// This assembly has native code (eg. C++/CLI)
    	module.NativeWrite(@"C:\saved-assembly.dll");
    }

PDB files

PDB files are read from disk by default. You can change this behaviour by creating a ModuleCreationOptions and passing it in to the code that creates a module.

To save a PDB file, create a ModuleWriterOptions / NativeModuleWriterOptions and set its WritePdb property to true. By default, it will create a PDB file with the same name as the output assembly but with a .pdb extension. You can override this by writing the PDB file name to PdbFileName or writing your own stream to PdbStream. If PdbStream is initialized, PdbFileName should also be initialized because the name of the PDB file will be written to the PE file.

    var mod = ModuleDefMD.Load(@"C:\myfile.dll");
    // ...
    var wopts = new dnlib.DotNet.Writer.ModuleWriterOptions(mod);
    wopts.WritePdb = true;
    // wopts.PdbFileName = @"C:\out2.pdb";	// Set other file name
    mod.Write(@"C:\out.dll", wopts);

dnlib supports Windows PDBs, portable PDBs and embedded portable PDBs.

Windows PDBs

It's only possible to write Windows PDBs on Windows (portable PDBs can be written on any OS). dnlib has a managed Windows PDB reader that supports all OSes.

There are two native Windows PDB reader and writer implementations, the old diasymreader.dll that ships with .NET Framework and Microsoft.DiaSymReader.Native which has been updated with more features and bug fixes.

dnlib will use Microsoft.DiaSymReader.Native if it exists and fall back to diasymreader.dll if needed. PdbReaderOptions and PdbWriterOptions can be used to disable one of them.

Microsoft.DiaSymReader.Native is a NuGet package with 32-bit and 64-bit native DLLs. You have to add a reference to this NuGet package if you want dnlib to use it. dnlib doesn't add a reference to it.

Strong name signing an assembly

Use the following code to strong name sign the assembly when saving it:

    using dnlib.DotNet.Writer;
    ...
    // Open or create an assembly
    ModuleDef mod = ModuleDefMD.Load(.....);
    
    // Create writer options
    var opts = new ModuleWriterOptions(mod);
    
    // Open or create the strong name key
    var signatureKey = new StrongNameKey(@"c:\my\file.snk");
    
    // This method will initialize the required properties
    opts.InitializeStrongNameSigning(mod, signatureKey);
    
    // Write and strong name sign the assembly
    mod.Write(@"C:\out\file.dll", opts);

Enhanced strong name signing an assembly

See this MSDN article for info on enhanced strong naming.

Enhanced strong name signing without key migration:

    using dnlib.DotNet.Writer;
    ...
    // Open or create an assembly
    ModuleDef mod = ModuleDefMD.Load(....);
    
    // Open or create the signature keys
    var signatureKey = new StrongNameKey(....);
    var signaturePubKey = new StrongNamePublicKey(....);
    
    // Create module writer options
    var opts = new ModuleWriterOptions(mod);
    
    // This method will initialize the required properties
    opts.InitializeEnhancedStrongNameSigning(mod, signatureKey, signaturePubKey);
    
    // Write and strong name sign the assembly
    mod.Write(@"C:\out\file.dll", opts);

Enhanced strong name signing with key migration:

    using dnlib.DotNet.Writer;
    ...
    // Open or create an assembly
    ModuleDef mod = ModuleDefMD.Load(....);
    
    // Open or create the identity and signature keys
    var signatureKey = new StrongNameKey(....);
    var signaturePubKey = new StrongNamePublicKey(....);
    var identityKey = new StrongNameKey(....);
    var identityPubKey = new StrongNamePublicKey(....);
    
    // Create module writer options
    var opts = new ModuleWriterOptions(mod);
    
    // This method will initialize the required properties and add
    // the required attribute to the assembly.
    opts.InitializeEnhancedStrongNameSigning(mod, signatureKey, signaturePubKey, identityKey, identityPubKey);
    
    // Write and strong name sign the assembly
    mod.Write(@"C:\out\file.dll", opts);

Exporting managed methods (DllExport)

dnlib supports exporting managed methods so the managed DLL file can be loaded by native code and then executed. .NET Framework supports this feature, but there's no guarantee that other CLRs (eg. .NET Core or Mono/Unity) support this feature.

The MethodDef class has an ExportInfo property. If it gets initialized, the method gets exported when saving the module. At most 65536 (2^16) methods can be exported. This is a PE file limitation, not a dnlib limitation.

Exported methods should not be generic.

The method's calling convention should be changed to eg. stdcall, or cdecl, by adding an optional modifier to MethodDef.MethodSig.RetType. It must be a System.Runtime.CompilerServices.CallConvCdecl, System.Runtime.CompilerServices.CallConvStdcall, System.Runtime.CompilerServices.CallConvThiscall, or a System.Runtime.CompilerServices.CallConvFastcall, eg.:

var type = method.MethodSig.RetType;
type = new CModOptSig(module.CorLibTypes.GetTypeRef("System.Runtime.CompilerServices", "CallConvCdecl"), type);
method.MethodSig.RetType = type;

Requirements:

  • The assembly platform must be x86, x64, IA-64 or ARM (ARM64 isn't supported at the moment). AnyCPU assemblies are not supported. This is as simple as changing (if needed) ModuleWriterOptions.PEHeadersOptions.Machine when saving the file. x86 files should set 32-bit required flag and clear 32-bit preferred flag in the COR20 header.
  • ModuleWriterOptions.Cor20HeaderOptions.Flags: The IL Only bit must be cleared.
  • It must be a DLL file (see ModuleWriterOptions.PEHeadersOptions.Characteristics). The file will fail to load at runtime if it's an EXE file.

Type classes

The metadata has three type tables: TypeRef, TypeDef, and TypeSpec. The classes dnlib use are called the same. These three classes all implement ITypeDefOrRef.

There's also type signature classes. The base class is TypeSig. You'll find TypeSigs in method signatures (return type and parameter types) and locals. The TypeSpec class also has a TypeSig property.

All of these types implement IType.

TypeRef is a reference to a type in (usually) another assembly.

TypeDef is a type definition and it's a type defined in some module. This class does not derive from TypeRef. :)

TypeSpec can be a generic type, an array type, etc.

TypeSig is the base class of all type signatures (found in method sigs and locals). It has a Next property that points to the next TypeSig. Eg. a Byte[] would first contain a SZArraySig, and its Next property would point to Byte signature.

CorLibTypeSig is a simple corlib type. You don't create these directly. Use eg. module.CorLibTypes.Int32 to get a System.Int32 type signature.

ValueTypeSig is used when the next class is a value type.

ClassSig is used when the next class is a reference type.

GenericInstSig is a generic instance type. It has a reference to the generic type (a TypeDef or a TypeRef) and the generic arguments.

PtrSig is a pointer sig.

ByRefSig is a by reference type.

ArraySig is a multi-dimensional array type. Most likely when you create an array, you should use SZArraySig, and not ArraySig.

SZArraySig is a single dimension, zero lower bound array. In C#, a byte[] is a SZArraySig, and not an ArraySig.

GenericVar is a generic type variable.

GenericMVar is a generic method variable.

Some examples if you're not used to the way type signatures are represented in metadata:

    ModuleDef mod = ....;
    
    // Create a byte[]
    SZArraySig array1 = new SZArraySig(mod.CorLibTypes.Byte);
    
    // Create an int[][]
    SZArraySig array2 = new SZArraySig(new SZArraySig(mod.CorLibTypes.Int32));
    
    // Create an int[,]
    ArraySig array3 = new ArraySig(mod.CorLibTypes.Int32, 2);
    
    // Create an int[*] (one-dimensional array)
    ArraySig array4 = new ArraySig(mod.CorLibTypes.Int32, 1);
    
    // Create a Stream[]. Stream is a reference class so it must be enclosed in a ClassSig.
    // If it were a value type, you would use ValueTypeSig instead.
    TypeRef stream = new TypeRefUser(mod, "System.IO", "Stream", mod.CorLibTypes.AssemblyRef);
    SZArraySig array5 = new SZArraySig(new ClassSig(stream));

Sometimes you must convert an ITypeDefOrRef (TypeRef, TypeDef, or TypeSpec) to/from a TypeSig. There's extension methods you can use:

    // array5 is defined above
    ITypeDefOrRef type1 = array5.ToTypeDefOrRef();
    TypeSig type2 = type1.ToTypeSig();

Naming conventions of metadata table classes

For most tables in the metadata, there's a corresponding dnlib class with the exact same or a similar name. Eg. the metadata has a TypeDef table, and dnlib has a TypeDef class. Some tables don't have a class because they're referenced by other classes, and that information is part of some other class. Eg. the TypeDef class contains all its properties and events, even though the TypeDef table has no property or event column.

For each of these table classes, there's an abstract base class, and two sub classes. These sub classes are named the same as the base class but ends in either MD (for classes created from the metadata) or User (for classes created by the user). Eg. TypeDef is the base class, and it has two sub classes TypeDefMD which is auto-created from metadata, and TypeRefUser which is created by the user when adding user types. Most of the XyzMD classes are internal and can't be referenced directly by the user. They're created by ModuleDefMD (which is the only public MD class). All XyzUser classes are public.

Metadata table classes

Here's a list of the most common metadata table classes

AssemblyDef is the assembly class.

AssemblyRef is an assembly reference.

EventDef is an event definition. Owned by a TypeDef.

FieldDef is a field definition. Owned by a TypeDef.

GenericParam is a generic parameter (owned by a MethodDef or a TypeDef)

MemberRef is what you create if you need a field reference or a method reference.

MethodDef is a method definition. It usually has a CilBody with CIL instructions. Owned by a TypeDef.

MethodSpec is a instantiated generic method.

ModuleDef is the base module class. When you read an existing module, a ModuleDefMD is created.

ModuleRef is a module reference.

PropertyDef is a property definition. Owned by a TypeDef.

TypeDef is a type definition. It contains a lot of interesting stuff, including methods, fields, properties, etc.

TypeRef is a type reference. Usually to a type in another assembly.

TypeSpec is a type specification, eg. an array, generic type, etc.

Method classes

The following are the method classes: MethodDef, MemberRef (method ref) and MethodSpec. They all implement IMethod.

Field classes

The following are the field classes: FieldDef and MemberRef (field ref). They both implement IField.

Comparing types, methods, fields, etc

dnlib has a SigComparer class that can compare any type with any other type. Any method with any other method, etc. It also has several pre-created IEqualityComparer<T> classes (eg. TypeEqualityComparer, FieldEqualityComparer, etc) which you can use if you intend to eg. use a type as a key in a Dictionary<TKey, TValue>.

The SigComparer class can also compare types with System.Type, methods with System.Reflection.MethodBase, etc.

It has many options you can set, see SigComparerOptions. The default options is usually good enough, though.

    // Compare two types
    TypeRef type1 = ...;
    TypeDef type2 = ...;
    if (new SigComparer().Equals(type1, type2))
    	Console.WriteLine("They're equal");

    // Use the type equality comparer
    Dictionary<IType, int> dict = new Dictionary<IType, int>(TypeEqualityComparer.Instance);
    TypeDef type1 = ...;
    dict.Add(type1, 10);

    // Compare a `TypeRef` with a `System.Type`
    TypeRef type1 = ...;
    if (new SigComparer().Equals(type1, typeof(int)))
    	Console.WriteLine("They're equal");

It has many Equals() and GetHashCode() overloads.

.NET Resources

There's three types of .NET resource, and they all derive from the common base class Resource. ModuleDef.Resources is a list of all resources the module owns.

EmbeddedResource is a resource that has data embedded in the owner module. This is the most common type of resource and it's probably what you want.

AssemblyLinkedResource is a reference to a resource in another assembly.

LinkedResource is a reference to a resource on disk.

Win32 resources

ModuleDef.Win32Resources can be null or a Win32Resources instance. You can add/remove any Win32 resource blob. dnlib doesn't try to parse these blobs.

Parsing method bodies

This is usually only needed if you have decrypted a method body. If it's a standard method body, you can use MethodBodyReader.Create(). If it's similar to a standard method body, you can derive a class from MethodBodyReaderBase and override the necessary methods.

Resolving references

TypeRef.Resolve() and MemberRef.Resolve() both use module.Context.Resolver to resolve the type, field or method. The custom attribute parser code may also resolve type references.

If you call Resolve() or read custom attributes, you should initialize module.Context to a ModuleContext. It should normally be shared between all modules you open.

    AssemblyResolver asmResolver = new AssemblyResolver();
    ModuleContext modCtx = new ModuleContext(asmResolver);
    
    // All resolved assemblies will also get this same modCtx
    asmResolver.DefaultModuleContext = modCtx;
    
    // Enable the TypeDef cache for all assemblies that are loaded
    // by the assembly resolver. Only enable it if all auto-loaded
    // assemblies are read-only.
    asmResolver.EnableTypeDefCache = true;

All assemblies that you yourself open should be added to the assembly resolver cache.

    ModuleDefMD mod = ModuleDefMD.Load(...);
    mod.Context = modCtx;	// Use the previously created (and shared) context
    // This code assumes you're using the default assembly resolver
    ((AssemblyResolver)mod.Context.AssemblyResolver).AddToCache(mod);

Resolving types, methods, etc from metadata tokens

ModuleDefMD has several ResolveXXX() methods, eg. ResolveTypeDef(), ResolveMethod(), etc.

Creating mscorlib type references

Every module has a CorLibTypes property. It has references to a few of the simplest types such as all integer types, floating point types, Object, String, etc. If you need a type that's not there, you must create it yourself, eg.:

    TypeRef consoleRef = new TypeRefUser(mod, "System", "Console", mod.CorLibTypes.AssemblyRef);

Importing runtime types, methods, fields

To import a System.Type, System.Reflection.MethodInfo, System.Reflection.FieldInfo, etc into a module, use the Importer class.

    Importer importer = new Importer(mod);
    ITypeDefOrRef consoleRef = importer.Import(typeof(System.Console));
    IMethod writeLine = importer.Import(typeof(System.Console).GetMethod("WriteLine"));

You can also use it to import types, methods etc from another ModuleDef.

All imported types, methods etc will be references to the original assembly. I.e., it won't add the imported TypeDef to the target module. It will just create a TypeRef to it.

Using decrypted methods

If ModuleDefMD.MethodDecrypter is initialized, ModuleDefMD will call it and check whether the method has been decrypted. If it has, it calls IMethodDecrypter.GetMethodBody() which you should implement. Return the new MethodBody. GetMethodBody() should usually call MethodBodyReader.Create() which does the actual parsing of the CIL code.

It's also possible to override ModuleDefMD.ReadUserString(). This method is called by the CIL parser when it finds a Ldstr instruction. If ModuleDefMD.StringDecrypter is not null, its ReadUserString() method is called with the string token. Return the decrypted string or null if it should be read from the #US heap.

Low level access to the metadata

The low level classes are in the dnlib.DotNet.MD namespace.

Open an existing .NET module/assembly and you get a ModuleDefMD. It has several properties, eg. StringsStream is the #Strings stream.

The Metadata property gives you full access to the metadata.

To get a list of all valid TypeDef rids (row IDs), use this code:

    using dnlib.DotNet.MD;
    // ...
    ModuleDefMD mod = ModuleDefMD.Load(...);
    RidList typeDefRids = mod.Metadata.GetTypeDefRidList();
    for (int i = 0; i < typeDefRids.Count; i++)
    	Console.WriteLine("rid: {0}", typeDefRids[i]);

You don't need to create a ModuleDefMD, though. See MetadataFactory.

Credits

Big thanks to Ki for writing the managed Windows PDB reader!

List of all contributors

About

Fork of 0xd4d/dnlib to allow parsing "stark" assemblies and modules

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages

  • C# 100.0%