August 28th, 2024

BinaryFormatter removed from .NET 9

Immo Landwerth
Immo Landwerth
Program Manager

Starting with .NET 9, we no longer include an implementation of BinaryFormatter in the runtime (.NET Framework remains unchanged). The APIs are still present, but their implementation always throws an exception, regardless of project type. Hence, setting the existing backwards compatibility flag is no longer sufficient to use BinaryFormatter.

In this blog post, I’ll explain why this change was made and what options you have to move forward.

TL;DR: What should I do?

You have two options to address the removal of BinaryFormatter‘s implementation:

  1. Migrate away from BinaryFormatter. We strongly recommend that you investigate options to stop using BinaryFormatter due to the associated security risks. The BinaryFormatter migration guide lists several options.
  2. Keep using BinaryFormatter. If you need to continue using BinaryFormatter in .NET 9, you need to depend on the unsupported System.Runtime.Serialization.Formatters NuGet package, which restores the unsafe legacy functionality and replaces the throwing implementation.

Note

Please note that .NET Framework is unaffected by this change and continues to include an implementation of BinaryFormatter. However, we still strongly recommend to stop using BinaryFormatter from .NET Framework too, for the same reasons.

What’s the risk in using BinaryFormatter?

Any deserializer, binary or text, that allows its input to carry information about the objects to be created is a security problem waiting to happen. There is a common weakness enumeration (CWE) that describes the issue: CWE-502 “Deserialization of Untrusted Data”. BinaryFormatter, included in the the initial release of .NET Framework in 2002, is such a deserializer. We also cover this in the BinaryFormatter security guide.

Why we removed BinaryFormatter

We strongly believe that .NET should make it easy for customers to do the right thing and hard, if not impossible, to do the wrong thing. We generally refer to this as the “pit of success”.

Shipping a technology that is widely regarded as unsafe is counter to this goal. At the same time, we also have a responsibility to ensure customers can support and move their existing code forward. We can’t just remove widely used components from a .NET release, even when communicated long in advance. We also need a migration plan and stop gap options.

This removal was not a sudden change. Due to the known risks of using BinaryFormatter, we excluded it from .NET Core 1.0. But without a clear migration path to using something safer, customer demand led to BinaryFormatter being included in .NET Core 2.0.

Since then, we have been on the path to removing BinaryFormatter, slowly turning it off by default in multiple project types but letting consumers opt-in via flags if still needed for backward compatibility:

In .NET 9 we removed all remaining in-box dependencies on BinaryFormatter and replaced the implementation with one that always throws.

Options to move forward

New code should not take a dependency on BinaryFormatter. For existing code, you should first investigate alternatives to BinaryFormatter. In case you don’t control the serializer but only perform deserialization, you can consider only reading the BinaryFormatter payload, without performing any deserialization. And if none of this works for you can bring the implementation back by depending on an (unsupported) compatibility package.

I’ll explore these options in more detail below.

Migrate-Away

You should first investigate whether you can replace BinaryFormatter with another serializer. We have four recommendations:

Since DataContractSerializer honors the same attribute and interface as BinaryFormatter (namely [Serializable] and ISerializable), it’s probably the easiest one to migrate to. If your migration goals include adopting a modern and performant serializer or a format with better cross-platform interoperability, the other options should be considered.

Read BinaryFormatter Payloads

If your code doesn’t control the serialization but only the deserialization, use the new NrbfDecoder to read BinaryFormatter payloads. This allows you to read the encoded data without any deserialization. It’s the equivalent of using a JSON/XML reader without the deserializer:

using System.Formats.Nrbf;

void Read(Stream payload)
{
    SerializationRecord rootObject = NrbfDecoder.Decode(payload);

    if (rootObject is PrimitiveTypeRecord primitiveRecord)
    {
        Console.WriteLine($"It was a primitive value: '{primitiveRecord.Value}'");
    }
    else if (rootObject is ClassRecord classRecord)
    {
        Console.WriteLine($"It was a class record of '{classRecord.TypeName.AssemblyQualifiedName}' type name.");
    }
    else if (rootObject is SZArrayRecord<byte> arrayOfBytes)
    {
        Console.WriteLine($"It was an array of `{arrayOfBytes.Length}`-many bytes.");
    }
}

For more details, check out the Nrbf documentation.

BinaryFormatter compatibility package

If you have explored the options and determined you can’t migrate away from BinaryFormatter, you can also install the unsupported System.Runtime.Serialization.Formatters NuGet package and set the compatibility switch to true:

<PropertyGroup>
  <TargetFramework>net9.0</TargetFramework>
  <EnableUnsafeBinaryFormatterSerialization>true</EnableUnsafeBinaryFormatterSerialization>
</PropertyGroup>

<ItemGroup>
  <PackageReference Include="System.Runtime.Serialization.Formatters" Version="9.0.0" />
</ItemGroup>

The package replaces the in-box implementation of BinaryFormatter with a functioning one, including its vulnerabilities and risks. It’s meant as a stopgap if you can’t wait with migrating to .NET 9 while not having replaced the usages of BinaryFormatter yet.

Since the BinaryFormatter API still exists and this package only replaces the in-box implementation you only need to reference it from application projects. Existing code that is compiled against BinaryFormatter will continue to work.

Caution

The compatibility package is not supported and unsafe. We strongly recommend against taking a dependency on this package and to instead migrate away from BinaryFormatter.

Summary

Since the start of .NET Core we have been on a path of deprecating BinaryFormatter, due to its security risks.

Starting with .NET 9, we no longer ship an implementation with the runtime. We recommend that you migrate away from BinaryFormatter. If that doesn’t work for you can either start reading the binary payloads without deserializing or you can take a dependency on the unsupported compatibility package.

Category
.NET.NET CoreSecurity

Author

Immo Landwerth
Immo Landwerth
Program Manager

Immo Landwerth is a program manager on the .NET Framework team at Microsoft. He specializes in API design, the base class libraries (BCL), and .NET Standard. He works on base class libraries which represents the core types of the .NET platform, such as string and int but also includes collections and IO. He's involved with portable class libraries and works on shipping more framework components in an out-of-band fashion via NuGet.

21 comments

Leave a comment

Newest
Newest
Popular
Oldest
  • Tyler 1 day ago

    We had replaced our use of BinaryFormatter with JSON serialization using STJ but only as a stopgap. I believe we will be adopting MessagePack given the huge difference in payload size.

  • Steve Tabler 1 week ago

    This is why I keep my software developments centered on .NET 3.5

    I have begun to use XML Serialization on many projects, but that’s so I can better debug things, and occasionally hand-edit my XML File over some issue, which I can’t conveniently do with a binary serialized file. However, the XML Serialization is extremely in efficient: my new editor, which keeps track of many text and document attributes, storage for a 1K test-file can easily exceed 64-K with all the attributes being XML-serialized.

    Eliminating binary serialization is just yet another mistake in C# management in a long string of C# management.

  • Michael Taylor 1 week ago

    For our code the greatest impact this has had is with custom exceptions. Since the beginning of .NET the guidelines for custom exceptions have always been to mark it Serializable (which still works), implement a protected constructor that accepts Serializationinfo and StreamingContext and override GetObjectData so your exception data can be properly serialized across boundaries.

    With the removal of the formatter then these should now be removed from custom exception types. However that is a breaking change if you are a library writer, your exception class isn’t sealed and somebody created a derived class. While you could remove GetObjectData and it derived classes would work, the protected constructor cannot be removed. This leaves you with obsolete code you cannot remove, that won’t ever be called (in theory) and is probably broken anyway.

    • Joe Beans

      Exceptions have been my pain point too. I use IPC to call code in other processes and BinaryFormatter has been the go-to method of propagating exceptions back to the caller process. I don’t really know what to replace it with.

    • David Lee 1 week ago · Edited

      I don’t think you need to actually remove the code. Those APIs should still be present and not cause compile-time issues. They only throw when actually used at run time.

      Keeping the code also allows you to serve library consumers who actually have to install the legacy fallback package and use this technology.

      • Michael Taylor 1 week ago

        They do cause a compile time issue. They cause compiler warnings because they are marked obsolete. If you compile warnings as errors then it doesn’t work. The fix is to annotate your own code as well but now you also have code marked as obsolete that you really can’t get rid of. As I mentioned, it complicates things.

  • Irakli Lomidze 1 week ago · Edited

    That is a wrong, very wrong decision. What about GRPC or other serialization scenarios?
    Depreciation of BinarySerializaton will affect the performance of our solutions.
    Stop doing that; mark it with a high-level warning and keep stuff that worked fine.

    • Immo LandwerthMicrosoft employee Author 1 week ago

      It seems there is some confusion around what this post is saying:

      1. What is deprecated is a specific binary serializer, namely BinaryFormatter, which is part of the .NET remoting infrastructure.
      2. Serialization into a binary format of course still makes sense and this post recommends two serializers: MessagePack and Protobuf.NET (the latter is also part of the .NET story for GRPC).
      3. BinaryFormatter is indeed marked with a high-level warning via obsoletion as well as the fact that by default it will throw. However, if you insist, BinaryFormatter is still available via the package.
    • Michael Taylor 1 week ago

      Binary serialization IS NOT deprecated. Nowhere in this post does it say that. In fact it identifies 2 binary serialization formats that you can use. What is removed is BinaryFormatter which was the original binary serialization format that .NET used for things like remoting and whatnot. It has not been recommended for use for quite a while because: a) it is a security vulnerability, and b) there are better options.

      You can use binary serialization in your code, and you likely already are. You just don’t want to use BinaryFormatter as the serialization protocol anymore.

      What about GRPC or other serialization scenarios?

      gRPC uses protobuf which is one of the links in the post to use as an alternative format. Other serialization scenarios will require you to eval your code. If you are communicating with a legacy NF project that uses BinaryFormatter on its side then you’ll need to pull in the new NuGet package for now. If you have control over the format then switch to a binary serializer that was mentioned in the post or another one that you have found.

    • David Lee 1 week ago · Edited

      BinaryFormatter here means a specific class that serializes to the insecure NRBF format. It’s not about deprecating all binary serialization. In fact, protobuf, employed by gRPC you mentioned, is even recommended as a migration path in the article.

    • Huo Yaoyuan 1 week ago

      As discussed in many threads, binary serializer is NOT performant.
      Other binary binary serialization formats are unrelated to the .NET binary serialization format, which is fundamentally dangerous.

      • Immo LandwerthMicrosoft employee Author 1 week ago · Edited

        @Irakli Lomidize

        How a String parser could have the same performance as memcpy ?

        BinaryFormatter is not equivalent to a memcpy. It has to walk the object graph, has to keep track of unique instances, handle cycles, call via interfaces, and use reflection to read attributes/fields in order to determine the state that needs to be written out. Deserialization is even more expensive, it has to load and resolve types, use reflection to call constructors, set fields, and so on.

        If performance is important then BinaryFormatter isn’t the best choice. System.Text.Json support serialization via a static code generator which has likely a better performance profile because it doesn’t need reflection. However, that doesn’t mean the payload is smaller, which is where binary formats usually shine.

        The post recommends two binary serializers, both of which perform better than `BinaryFormatter`, namley MessagePack and protobuf-net.

      • Irakli Lomidze 1 week ago

        How a String parser could have the same performance as memcpy ?

  • Daniel Smith 1 week ago

    What’s the implications for WinForms? Don’t a lot of the components like ImageList etc. use BinaryFormatter behind the scenes?

  • Angius 1 week ago

    Oh, wow, gotta mark this date in my calendar! On this day, the .NET team actually decided to deprecate something! 🤣

    • Immo LandwerthMicrosoft employee Author 1 week ago

      We do deprecate / obsolete APIs a lot more in .NET Core, mostly because we have better tooling support and a better side-by-side story. What is rare for us is making breaking changes to the extent of actually removing components from the core libraries. There are very few cases where such a massive disruption is worth the monumental effort for the ecosystem. We believe binary formatter is one of those very rare instances because there is decades of security research that shows how much trouble this thing causes.

Feedback