io_splice - zero-copy pipe I/O for Linux and Ruby

The splice family of Linux system calls can transfer data between file descriptors without the need to copy data into userspace. Instead of a userspace buffer, they rely on an ordinary Unix pipe as a kernel-level buffer.

Features

  • Efficient zero-copy I/O avoids data copies into userspace, reducing garbage for the Ruby garbage collector as a side effect.

  • More flexible than sendfile, may be used to enable copies between arbitrary file descriptors (assuming kernel support), not just file-to-socket (or file-to-anything in newer Linux).

  • Thread-safe blocking operations under Ruby 1.9, releases GVL if blocking operations are used.

  • Almost drop-in replacement for IO.copy_stream: IO::Splice.copy_stream

  • Safely usable with non-blocking I/O frameworks (unlike IO.copy_stream) when combined with the IO::Splice::F_NONBLOCK flag.

  • Fully-documented library API and examples

  • Adds helpful IO#pipe_size and IO#pipe_size= accessor methods for resizing the pipe buffer in Linux 2.6.35 or later.

Install

Operating system support for the splice(2), tee(2) and vmsplice(2) system calls is required. Currently, only Linux 2.6.17 or later with a modern C library (glibc 2.5 or later) support these system calls. Using the latest stable Linux kernel is HIGHLY recommended as there have been numerous bugs in the early releases of these system calls.

If you’re using a packaged Ruby distribution, make sure you have a C compiler and the matching Ruby development libraries and headers.

If you use RubyGems:

gem install io_splice

Otherwise grab the latest tarball from:

bogomips.org/ruby_io_splice/files/

Unpack it, and run “ruby setup.rb”

Usage

Our API matches the C API closely, see the RDoc for full API documentation and the Linux manpages for more details.

All of these system calls are fairly new have seen limited usage anywhere, even outside of the Ruby world.

If you encounter problems (privilege escalation, memory leaks (in the kernel), poor performance, corrupt data, etc..), try upgrading the Linux kernel to the latest stable version.

These system calls have a lot of potential, and will hopefully be standardized and available in non-Linux kernels some day.

Development

You can get the latest source via git from the following locations:

git://bogomips.org/ruby_io_splice.git
git://repo.or.cz/ruby_io_splice.git (mirror)

You may browse the code from the web and download the latest snapshot tarballs here:

Inline patches (from “git format-patch”) to the mailing list are preferred because they allow code review and comments in the reply to the patch.

We will adhere to mostly the same conventions for patch submissions as git itself. See the Documentation/SubmittingPatches document distributed with git on on patch submission guidelines to follow. Just don’t email the git mailing list or maintainer with io_splice patches.

Contact

All feedback (bug reports, user/development discussion, patches, pull requests) go to the mailing list: [email protected]

Mailing List Archives

In addition to the rsync-able archives provided by librelist.com/, we are also mirrored to Gmane and maintain our own mbox mirrors downloadable via HTTP.