/rasn

A Safe #[no_std] ASN.1 Codec Framework

Primary LanguageRustOtherNOASSERTION

Rasn

crates.io Help Wanted Lines Of Code Documentation

Welcome to rasn (pronounced "raisin"), a safe #[no_std] ASN.1 codec framework. That enables you to safely create, share, and handle ASN.1 data types from and to different encoding rules. If you are unfamiliar with ASN.1 and encoding formats like BER/DER, I would recommend reading "A Warm Welcome to ASN.1 and DER" by Let's Encrypt as a quick introduction before continuing. In short it is an "Interface Description Language" (and data model) with a set of encoding formats (called rules) for that model. It was originally designed in the late 1980s and is used throughout the industry especially in telecommunications and cryptography.

Features

Abstract Codec Data Model

There are quite a few existing ASN.1 related Rust crates already, however they are currently specific to a single format or even a single standard, this makes it hard to share and re-use standards that are specified in ASN.1. Now with rasn's abstract model you can build and share ASN.1 data types as crates that work with any encoder or decoder regardless of the underlying encoding rules, whether it's BER, CER, DER, or your own custom encoding.

#[no_std] Support

Rasn is entirely #[no_std], so you can share the same ASN.1 implementation on any Rust target platform that can support alloc.

Rich Data Types

Rasn currently has support for nearly all of ASN.1's data types. rasn uses popular community libraries such as bitvec, bytes, and chrono for some of its data types as well as providing a couple of its own. Check out the types module for what's currently available.

Safe BER, CER, and DER Codecs

Included with the framework is a implementation of the X.690 standard also known as the Basic Encoding Rules, Canonical Encoding Rules, and Distinguished Encoding Rules codecs. The encoder and decoder have been written in 100% safe Rust and fuzzed with American Fuzzy Lop Plus Plus to ensure that the decoder correctly handles random input, and if valid that the encoder can correctly re-encode that value.

RFC implementations

Rasn also provides implementations for a number of IETF RFCs using the rasn framework for use out of the box. These crates provide strongly typed definitions for the necessary data types. Like rasn they are #[no_std], as well as being transport layer and encoding rule agnostic.

Powerful Derive Macros

Easily model your structs and enums with derive equivalents of all of the traits. These macros provide a automatic implementation that ensures your model is a valid ASN.1 type at compile-time. To explain that though, first we have to explain…

How It Works

The codec API has been designed for ease of use, safety, and being hard to misuse. The most common mistakes are around handling the length and ensuring it's correctly encoded and decoded. In rasn this is completely abstracted away letting you focus on the abstract model. Let's look at what decoding a simple custom SEQUENCE type looks like.

Person ::= SEQUENCE {
  age INTEGER,
  name UTF8String
}

Which we want to map to the following equivalent Rust code.

struct Person {
    age: rasn::types::Integer,
    name: String, // or rasn::types::Utf8String
}

Implementing The Traits

When modelling an ASN.1 data type, there are three traits we'll need to implement. Decode and Encode for converting to and from encoding rules, and the shared AsnType trait; which defines some associated data needed to be given to the encoder and decoder. Currently the only thing we have define is the tag to use to identify our type.

# struct Person;
use rasn::{AsnType, Tag};

impl AsnType for Person {
    // Default tag for sequences.
    const TAG: Tag = Tag::SEQUENCE;
}

Next is the Decode and Encode traits. These are mirrors of each other and both have one provided method (decode/encode) and one required method (decode_with_tag/encode_with_tag). Since in ASN.1 nearly every type can be implicitly tagged allowing anyone to override the tag associated with the type, having *_with_tag as a required method requires the implementer to correctly handle this case, and the provided methods simply calls *_with_tag with the type's associated AsnType::TAG. Let's look at what the codec implementation of Person looks like.

# struct Person { name: Utf8String, age: Integer }
# impl rasn::AsnType for Person { const TAG: Tag = Tag::SEQUENCE; }
use rasn::{Decode, Decoder, Encode, Encoder, Tag, types::{Integer, Utf8String}};

impl Decode for Person {
    fn decode_with_tag<D: Decoder>(decoder: &mut D, tag: Tag) -> Result<Self, D::Error> {
        // Accepts a closure that decodes the contents of the sequence.
        decoder.decode_sequence(tag, |decoder| {
            let age = Integer::decode(decoder)?;
            let name = Utf8String::decode(decoder)?;
            Ok(Self { age, name })
        })
    }
}

impl Encode for Person {
    fn encode_with_tag<E: Encoder>(&self, encoder: &mut E, tag: Tag) -> Result<(), E::Error> {
        // Accepts a closure that encodes the contents of the sequence.
        encoder.encode_sequence(tag, |encoder| {
            self.age.encode(encoder)?;
            self.name.encode(encoder)?;
            Ok(())
        })?;

        Ok(())
    }
}

That's it! We've just created a new ASN.1 that be encoded and decoded to BER, CER, and DER; and nowhere did we have to check the tag, the length, or whether the string was primitive or constructed encoded. All those nasty encoding rules details are completely abstracted away so your type only has handle how to map to and from ASN.1's data model.

With all the actual conversion code isolated to the codec implementations you can know that your model is always safe to use. The API has also been designed to prevent you from making common logic errors that can lead to invalid encoding. For example; if we look back our Encode implementation, and what if we forgot to use the encoder we were given in encode_sequence and tired to use the parent instead?

error[E0501]: cannot borrow `*encoder` as mutable because previous closure requires unique access
   --> tests/derive.rs:122:9
    |
122 |           encoder.encode_sequence(tag, |sequence| {
    |           ^       ---------------      ---------- closure construction occurs here
    |           |       |
    |  _________|       first borrow later used by call
    | |
123 | |             self.age.encode(encoder)?;
    | |                             ------- first borrow occurs due to use of `encoder` in closure
124 | |             self.name.encode(sequence)?;
125 | |             Ok(())
126 | |         })?;
    | |__________^ second borrow occurs here

error[E0500]: closure requires unique access to `encoder` but it is already borrowed
   --> tests/derive.rs:122:38
    |
122 |         encoder.encode_sequence(tag, |sequence| {
    |         ------- ---------------      ^^^^^^^^^^ closure construction occurs here
    |         |       |
    |         |       first borrow later used by call
    |         borrow occurs here
123 |             self.age.encode(encoder)?;
    |                             ------- second borrow occurs due to use of `encoder` in closure

Our code fails to compile! Which, in this case is great, there's no chance that our contents will accidentally be encoded in the wrong sequence because we forgot to change the name of a variable. These ownership semantics also mean that an Encoder can't accidentally encode the contents of a sequence multiple times in their implementation. Let's see how we can try to take this even further.

Compile-Safe ASN.1 With Macros

So far we've shown how rasn's API takes steps to be safe and protect from accidentally creating an invalid model. However, it's often hard to cover everything in an imperative API. Something that is important to understand about ASN.1 that isn't obvious in the above examples is that; in ASN.1, all types can be identified by a tag (essentially two numbers e.g. INTEGER's tag is 0, 2). Field and variant names are not transmitted in most encoding rules, so this tag is also used to identify fields or variants in a SEQUENCE or CHOICE. This means that every that in a ASN.1 struct or enum every field and variant must have a distinct tag for the whole type to be considered valid. For example ; If we changed age in Person to be a String like below it would be invalid ASN.1 even though it compiles and runs correctly, we have to either use a different type or override age's tag to be distinct from name's. When implementing the AsnType trait yourself this requirement must checked by manually, however as we'll see you generally won't need to do that.

Included with rasn is a set of derive macros that enable you to have your ASN.1 model implementation implemented declaratively. The Encode and Decode macros will essentially auto-generate the implementations we showed earlier, but the real magic is the AsnType derive macro. Thanks to the static-assertations crate and recent developments in const fn; the AsnType derive will not only generate your AsnType implementation, it will also generate a check that asserts that every field or variant has a distinct tag at compile-time. This means now if for some reason we made a change to one of the types in person, we don't have re-check that our model is still valid, the compiler takes care of that for us.

// Invalid
#[derive(rasn::AsnType)]
struct Person {
    age: Option<String>,
    name: Option<String>,
}

We'll now get the following error trying to compile the above definition.

error[E0080]: evaluation of constant value failed
   --> tests/derive.rs:146:10
    |
146 | #[derive(rasn::AsnType)]
    |          ^^^^^^^^^^^^^ the evaluated program panicked at 'Person's fields is not a valid order of ASN.1 tags, ensure that your field's tags and OPTIONAL
s are correct.', tests/derive.rs:146:10
    |
    = note: this error originates in the macro `$crate::panic::panic_2015` (in Nightly builds, run with -Z macro-backtrace for more info)

Validating your model at compile-time enables you to work on ASN.1 code without fear that you're unintentionally changing something in the background. I bet you're wondering now though, how we are supposed to have a struct with two strings for fields? The answer is thankfully pretty simple, you just add #[rasn(tag)] attribute to override the tags of one or more of the types. However we can actually go further, because in ASN.1 there's the concept of having AUTOMATIC TAGS which essentially tells your ASN.1 compiler to automatically generate distinct tags for your ASN.1 definition. Now with rasn you can do that in Rust! Applying #[rasn(automatic_tags)] to the container automatically generate tags will apply the same automatic tagging transformation you'd expect from an ASN.1 compiler.

use rasn::AsnType;

// Valid
#[derive(AsnType)]
struct Person {
    #[rasn(tag(context, 0))] // or just #[rasn(tag(0))]
    age: Option<String>,
    name: Option<String>,
}

// Also valid
#[derive(AsnType)]
#[rasn(automatic_tags)]
struct Person2 {
    age: Option<String>,
    name: Option<String>,
}

What's Missing & What's Next?

While rasn is starting out relatively full featured, there are still plenty of missing features and types. The main limitation that exists currently, is the lack of support for constraints. In ASN.1 you can arbitrarily put constraints on types (e.g. array always of length 4, or a number between 1–10). There are encoding rules that take advantage of these constraints to save space, and you can't currently communicate those constraints to a codec in a safe abstract fashion. So this means for example you wouldn't be able to implement a Packed Encoding Rules (PER) codec that handles constraints correctly. I want to eventually support constraints and PER directly, but it still requires a lot of design work to be thought through first.

Disclaimer

The software is provided "as is" and the authors disclaim all warranties with regard to this software including all implied warranties of merchant-ability and fitness. In no event shall the authors be liable for any special, direct, indirect, or consequential damages or any damages whatsoever resulting from loss of use, data or profits, whether in an action of contract, negligence or other tortuous action, arising out of or in connection with the use or performance of this software.