[virt-tools-list] [PATCH 1/3] Add detection for LXC + tests

[Daniel P. Berrange]

On Thu, Apr 11, 2013 at 02:51:53PM +0200, Marc Fournier wrote:
> LXC detection is not as trivial as one would expect. As it's based on
> cgroups, and cgroups can be used for other things than LXC containers,
> we must be careful not to match non-LXC cgroup usage.
> Furthermore, there are several LXC container implementations, which
> leave different marks, so we have to be sure to test for each of them.

Just look for 'container' in /proc/1/environ. Anything else is
legacy / deprecated.

> diff --git a/virt-what.in b/virt-what.in
> index 297956d..c2dad16 100644
> --- a/virt-what.in
> +++ b/virt-what.in
> @@ -135,6 +135,58 @@ if [ -d $root/proc/vz -a ! -d $root/proc/bc ]; then
>      echo openvz
>  fi
>  
> +# Check for LXC containers
> +# Added by Marc Fournier
> +# Based on a similar script written in ruby by Krzysztof Wilczynski
> +
> +# LXC containers are based on cgroups. So first check if cgroups are present
> +# in the running kernel.
> +if [ -e $root/proc/self/cgroup ] &&
> +  # But if the last field of the cgroup colon-delimited file is "/", we are
> +  # outside of a container. So only proceed further if this field has more
> +  # characters.
> +  grep -Eq '^[0-9]+:.+:/.+$' $root/proc/self/cgroup; then
> +
> +  # This is still not enough, as cgroups can be used for other things than LXC
> +  # containers. We base our final checks on the value/presence of various
> +  # environment variables passed to the "init" process.
> +  for envvar in $(cat $root/proc/1/environ | tr '\000' '\n'); do
> +
> +    case $(echo $envvar | cut -f1 -d'=') in
> +      # native lxc tools and virsh will set the "container" environment
> +      # variable to either "lxc", or "libvirt-lxc", or "libvirt", depending on
> +      # the version.
> +      container)
> +        container_type=$(echo $envvar | while IFS='=' read k v; do echo $v; done)
> +        if echo $container_type | grep -Eq '^(lxc|libvirt-lxc|libvirt)$'; then
> +          echo lxc
> +          break
> +        fi
> +      ;;
> +
> +      # older native lxc tools don't set the "container" environment variable,
> +      # so the only way to detect them is via the "_" environment variable
> +      # (which is the command used to start the container).
> +      _)
> +        external_command=$(echo $envvar | while IFS='=' read k v; do echo $v; done)
> +        if echo $external_command | grep -iq 'lxc'; then
> +          echo lxc
> +          break
> +        fi
> +      ;;
> +
> +      # recent libvirt-based LXC containers will have 2 unambiguous environment
> +      # variables set. Checking for their presence should be sufficient.
> +      LIBVIRT_LXC_UUID|LIBVIRT_LXC_NAME)
> +        echo lxc
> +        break
> +      ;;
> +
> +    esac
> +  done
> +
> +fi

IMHO this is uneccesarily complex. Just check for existence of
a 'container' environment varibles in /proc/1/environ, which is
the agreed marker for a container[1], and ignore legacy code
which didn't set that. The cgroups check doesn't add any value
here either.

Daniel

[1] http://www.freedesktop.org/wiki/Software/systemd/ContainerInterface
-- 
|: http://berrange.com      -o-    http://www.flickr.com/photos/dberrange/ :|
|: http://libvirt.org              -o-             http://virt-manager.org :|
|: http://autobuild.org       -o-         http://search.cpan.org/~danberr/ :|
|: http://entangle-photo.org       -o-       http://live.gnome.org/gtk-vnc :|