Utility for installing binaries from source with a single command


License
Other
Install
pip install whack==0.7.0

Documentation

whack: compile and run relocatable Linux programs

Whack allows Linux programs such as nginx and the Apache HTTP Server to be installed with a single command. For instance, to install nginx to ~/apps/nginx:

whack install git+https://github.com/mwilliamson/whack-package-nginx.git ~/apps/nginx

Most Linux binaries aren't relocatable, meaning that they're compiled for a specific path on your filesystem. This means that if you want to install exactly the same program to a different path, you'll need to recompile the entire program. Whack allows you to create relocatable versions of these programs. On the first installation, the program is compiled and copied to the target directory. On subsequent installations, a cached version of the application is copied to the target directory.

Installation

Before you can use Whack, you need to install a utility called whack-run. You can download whack-run from GitHub:

$ curl -L https://github.com/mwilliamson/whack-run/archive/1.0.0.tar.gz > whack-run-1.0.0.tar.gz
$ tar xzf whack-run-1.0.0.tar.gz
$ cd whack-run-1.0.0
$ make

And as root:

# make install

This installs the binary whack-run to /usr/local/bin. Once whack-run has been installed, you can install Whack as an ordinary Python package:

pip install whack

Usage

whack install PACKAGE_SOURCE DESTINATION

For instance, to install nginx under ~/apps/nginx:

whack install git+https://github.com/mwilliamson/whack-package-nginx.git ~/apps/nginx

nginx can then be run with the command:

~/apps/sbin/nginx

Package sources can be git or hg repositories (prefix the repository URL with git+ and hg+ respectively), tarballs fetched over HTTP (detected by the prefixes http:// or https://`), or local paths (detected by one of the prefixes ``/, ./, or ../).

You can pass build parameters using the argument --add-parameter KEY=VALUE, or with its short alias -p KEY=VALUE. The build parameters that can be set depend on the package. For instance, to install a specific version of nginx:

whack install git+https://github.com/mwilliamson/whack-package-nginx.git ~/apps/nginx \
    -p nginx_version=1.2.7

If a build parameter isn't set, a package will usually have a sensible default.

Creating package sources

A package source describes how to go from nothing to an installed instance of a given program. The output directory containing the installed program is referred to as a package.

The below gives a fairly thorough description of how a package is built, but it will probably more sense once you take a look at a concrete example. The nginx source package is a good example since it's relatively straightforward.

There are normally at least three files in each package source:

  • whack/whack.json: a JSON file describing the package source
  • whack/downloads: an executable file that outputs required downloads to stdout
  • whack/build: an executable file that is executed to build the package

whack/whack.json

whack/whack.json should be a JSON object containing the following properties:

  • name: the name of the package, such as nginx. It should only contain lowercase letters, numbers, and hyphens.
  • sourcePaths (optional): the paths in the source package that are required to build the package. Defaults to ["whack"].
  • defaultParameters (optional): an object containing the default build parameters for the package.

Build parameters

When executing whack/downloads and whack/build, any build parameters are passed as environment variables. Build parameters are set according to the defaults in whack/whack.json. The build parameters explicitly passed by the user are then added, overriding any default parameters. The name of each build parameter is converted to uppercase in the environment. For instance, the build parameter "version" is available as the environment variable "VERSION".

whack/downloads

Before building the package, whack/downloads is executed with the output to stdout being captured. The output should be a list of URLs that are required to build the package, separated by new lines. Downloads are cached, so if you have an URL where the contents might change (for instance, the latest tarball for a program), you can either:

  • Try to find a URL for that specific version
  • Download the file manually in the build step

Building the package

The following steps are executed to build a package:

  • Read whack/whack.json to get a package description.
  • Set the values of the build parameters based on the defaults set in whack/whack.json and the user-provided parameters.
  • Create a temporary directory, called the build directory.
  • Copy the directories and files in the source package specified by sourcePaths into the build directory.
  • Execute whack/downloads with the build parameters set as environment variables, and download the files into the build directory.
  • Execute whack/build:
  • The build parameters are set as environment variables
  • The current working directory is set to the build directory
  • The target directory for the package is passed as a command line argument

When the package is built, any executable files should be placed in either .bin or .sbin directories, instead of bin and sbin. When the package is installed by Whack, bin and sbin will contain thin wrappers that set up the filesystem correctly, and then delegate to the equivalent executables in .bin and .sbin. See the section "How does Whack work?" for more details.

Examples of package sources:

How does Whack work?

Many Linux applications can be compiled and installed by running the following commands, or similar:

$ ./configure
$ make
$ make install

This usually installs the application under /usr/local. However, sometimes we want to install isolated instances of an application without being root. For instance, if we're developing a web application that uses Apache, it's helpful to have an isolated installation of Apache. We can change the installation prefix when running ./configure:

$ ./configure --prefix=/home/user/projects/web-app/apache
$ make
$ make install

While this works, it requires us to re-compile the application whenever we want to install it in a different location. Depending on the application, compilation can take a quite a while.

Whack solves this problem by using unshare and mount to change the filesystem for a single process. Each application is compiled with its prefix set to /usr/local/whack. Before running the binary for an application, Whack uses the unshare syscall to create a private mount namespace. This means that any mount calls only have visible effects within that process. We then mount the directory that the application was installed in onto /usr/local/whack, and exec the binary.

For instance, say we've installed nginx to ~/web-app/nginx by running

whack install git+https://github.com/mwilliamson/whack-package-nginx.git \
    ~/web-app/nginx

The actual nginx binary can be found in ~/web-app/nginx/.sbin (note that the binary is in a directory called .sbin, not sbin). If we try to run ~/web-app/nginx/.sbin/nginx directly, we'll get an error:

$ ~/web-app/nginx/.sbin/nginx
nginx: [alert] could not open error log file: open() "/usr/local/whack/logs/error.log" failed (2: No such file or directory)
2013/02/18 11:25:17 [emerg] 11586#0: open() "/usr/local/whack/conf/nginx.conf" failed (2: No such file or directory)

nginx expects to be installed under /usr/local/whack, but it's actually installed under ~/web-app/nginx. To run nginx successfully, we need to use whack-run:

$ whack-run ~/web-app/nginx ~/web-app/nginx/.sbin/nginx

When using whack-run, the following happens:

  1. whack-run calls unshare(CLONE_NEWNS), creating a private mount namespace.
  2. whack-run mounts ~/web-app/nginx onto /usr/local/whack. Since we called unshare earlier, this mount is only visible to this process.
  3. whack-run drops its user and group privileges. whack-run is installed with the setuid bit set so it can call unshare and mount.
  4. whack-run calls exec with the arguments it was passed i.e. exec ~/web-app/nginx/.sbin/nginx

Using whack-run to run nginx is a bit tedious. However, we can call ~/web-app/nginx/sbin/nginx directly (instead of ~/web-app/nginx/.sbin/nginx), which will call whack-run with appropriate arguments.

Although whack-run has the setuid bit set, it only uses root privileges to call unshare and mount. After that, user and group privileges are dropped.