JavaCAN

This README is for the latest, possibly unreleased, version. For the documentation on the 2.x releases, check the releases/2.x branch.

Bindings for SocketCAN's CAN_RAW, CAN_BCM and CAN_ISOTP sockets with full support for blocking and non-blocking IO. Non-blocking IO is possible using the epoll module, that provides an API very similar to Java's Selector API.

Implementing Java's SelectableChannel API is not possible with EPoll and SocketCAN due to various hardcoded assumptions in the JDK.

Is this project actively maintained?

Yes! While feature development tends to happen in bursts around specific topics, issues and pull requests will always be answered. Dependencies will be updated occasionally. Releases are made mostly based on demand. The master branch is generally in a releasable state, if you need something released, just ask for it.

What works?

• Creating and binding CAN_RAW, CAN_BCM, CAN_ISOTP and CAN_J1939 sockets
• Sending and receiving standard CAN and CAN-FD frames with and without EFF
• Getting and setting all supported socket options
• Event-driven networking using an IOSelector
• Fairly robust test coverage

What is missing?

• Support for other CAN protocols (e.g. CAN_MCNET)
• CAN XL
• A netty integration (see #20)
• BSD Support
• io_uring Support

Pull requests are welcome!

Related Projects

• obd4s: A Scala library for OBD-II communication with vehicles.
• VirtualECU: An ECU simulator to test OBD-II clients against.
• Apache PLC4X: Apache PLC4X brings support for various PLC systems. JavaCAN serves as the transport layer for CANopen and other CAN related protocols.

Supported Operating Systems

This project is a wrapper around SocketCAN, which is a Linux kernel module that implements CAN communication. As such, only Linux can be supported. For this reason, the custom Selector will also only use epoll (Linux API for event-driven IO), as support for other OS' is not possible anyway.

Supported Architectures

The project uses dockcross to cross-compile its native components for various Linux supported platforms.

Currently, the full build process includes the following architectures:

• x86_32
• x86_64
• armv6
• armv7
• armv7a
• armv7l (musl libc, linked statically)
• aarch64
• riscv32
• riscv64

The implementation can handle word sizes up to 64 bit and is byte order aware. If you need another architecture, feel free to ask for it! Alternatively read how to build another architecture down below.

Android

Additionally, the following architectures are included specifically for Android:

• android-arm
• android-arm64
• android-x86_64
• android-x86_32

How to use

CAN_RAW, CAN_BCM and CAN_ISOTP channels

1. Compile yourself or get a compiled release from Maven Central: Core, EPoll (Check Versions -> Browse for published artifacts)
2. Install the native components into your LD_LIBRARY_PATH or configure the appropriate Java properties (See next section)
3. Create a channel by calling one of the CanChannels.new...Channel() methods
4. Create a NetworkDevice using its static lookup(String) method
5. Bind the channel to an interface using the bind(CanDevice) method

Usage example can be found in the unit tests or in the related projects mentioned above.

Remember: JavaCAN is a fairly thin wrapper around Linux syscalls. Even though some aspects of the low-level C API are hidden, most Java APIs in this library will at some point call into a (usually similarly named) C API and as such inherits all of its properties. For example RawCanChannel.close() translates to a call to close() on the underlying file descriptor, so their behaviour should be identical. So if the behaviour of a certain API is unclear, a look into the man pages of related Linux syscalls might help. Feel free to still request additional documentation in the issues on GitHub!

Native components

The library relies on several native (JNI) components. By default, these components are either loaded from the standard library path (LD_LIBRARY_PATH / java.library.path) or are extracted from the classpath into a temporary folder.

There are a few approaches to get the correct native libraries loaded:

1. Installing the libraries into the library path (the LD_LIBRARY_PATH environment variable or the java.library.path property)
2. Configuring the javacan.native.javacan-<module>.path property to tell JavaCAN the exact file system path where the native component is located
3. Configuring the javacan.native.javacan-<module>.classpath property to tell JavaCAN the exact location on the classpath where the native component is located
4. Adding one of the architecture-specific jar files into the classpath (either at compile time or runtime)

For applications that are intended to run on a single architecture or that build architecture-specific versions already, the simplest solution is to bundle the provided architecture-specific jar files matching the build architecture.

For applications supporting multiple architectures at once I'd recommend dynamically adding the architecture-specific jar file at runtime or to repackage the available native libraries and dynamically configuring the javacan.native.javacan-<module>.path properties in the CLI or before any JavaCAN classes are loaded.

The value for the <module> placeholder used throughout this section is core and if the EPoll support is used, an additional option with epoll for <module> is necessary.

Architecture Detection

While JavaCAN 2.x bundled the native components in its main artifacts, starting with the 3.x release series the native components are instead provided as separate jar files (classified by their architecture). This provides full control over which library is loaded, especially on architectures on which the JVM doesn't provide enough information for a reliable architecture detection. Programs using JavaCAN usually depend on specific architectures and thus can pull just the relevant components and nothing more.

For applications like test tools that don't really care about program size and don't need to support every possible architecture, starting with JavaCAN 3.3 *-arch-detect modules are provided. These modules bundle all prebuilt architectures and provide a function to automatically load the correct variant based on the os.arch system property.

For example for the javacan-core module you would instead depend on the javacan-core-arch-detect module and then, somewhere early in your program, invoke JavaCANAutoDetect.initialize(). This will configure the necessary system property based on the architecture detection and eagerly initialize JavaCAN.

Troubleshooting

In case you have issues, have a look at the troubleshooting document.

How to build

Prerequisites

For local compilation:

• GCC (or compatible)
• Java >= 11

For cross compilation:

• Podman or docker and permissions to run containers
• Java >= 11

For tests:

• A fairly recent Linux kernel with CAN support
• The can-isotp kernel module loaded (Kernel 5.10 with CONFIG_CAN_ISOTP enabled or the out-of-tree module)
• can-utils installed in the PATH
• A CAN interface named "vcan0"
• Java >= 11

For usage:

• A fairly recent Linux kernel with CAN support
• For ISOTP channels, the can-isotp kernel module loaded (Kernel 5.10 with CONFIG_CAN_ISOTP enabled or the out-of-tree module)
• Java >= 8
• A few kilobytes of disk space to extract the native components

Building

All architectures can be built in parallel with using the compileNativeAll task:

./gradlew compileNativeAll

Alternatively there exist packageNativeFor* tasks for all the bundled architectures. A special task is packageNativeForHost, which will use the compiler toolchain on your machine instead of dockcross. This is primarily intended for unit tests, but can also be used to build on systems that are not supported by dockcross.

All architectures allow their dockcross image to be overridden by properties like dockcross.image.<architecture>. The linking mode (dynamic or static) can also be overridden by properties like dockcross.link.<architecture>.

Using the property javacan.extra-archs additional architectures can be defined as a comma-separated list. The values will be used as the jar classifier. These additional architectures will also require manually setting configuring the dockcross image and linking mode using the previously mentioned properties.