#!/bin/bash
#
# chkconfig: 345 20 80
# description: Create BTRFS subvolumes for LXC reference images.
#
# Thierry Parmentelat <thierry.parmentelat@inria.fr>
# Copyright (C) 2012 INRIA
#

# not needed -- Source function library
#. /etc/init.d/functions

# a f>=18 root context won't have /bin nor /sbin in its path, which is where
# chkconfig is located on f<=14 images
export PATH=$PATH:/bin:/sbin

# This is where sliceimage(s) rpms store their reference images
sliceimage_dir=/vservers/.vref
sliceimage_stubs=/vservers/.vstub
lxc_dir=/vservers/.lvref
tmp_dir=/vservers/.ltmp

# Check if we are in the build environment
function check_node () {
    mount | grep -q 'planetlab-vservers' || exit 0
    [ -d $sliceimage_dir ] || { echo "No sliceimage installed" ; exit 1; }
}

function start () {

    echo "lxc-sliceimage: --- STARTING $(date) ---"

    check_node
    
    mkdir -p $lxc_dir

    slicefamilies=$(cd $sliceimage_dir ; ls )

    for slicefamily in $slicefamilies; do
	echo ==================== $slicefamily
	# initialize if needed
	[ -d $lxc_dir/$slicefamily ] || { 
	    echo "creating btrfs subvolume" ; 
	    btrfs subvolume create $lxc_dir/$slicefamily ;
	}
        # xxx what is that ?
	#btrfs subvolume create $lxc_dir/lxc-squeeze-x86_64
        # copy the slice image into the btrfs ?
#	echo "============================== mirroring start $(date)"
	echo "$sliceimage_dir/$slicefamily/ onto btrfs subvolume $lxc_dir/$slicefamily/"
	# mention --delete for rpm updates
	# we want to keep only our own additions, that will be re-created in the remaining 
	# of this script anyway
	rsync -a --delete $sliceimage_dir/$slicefamily/ $lxc_dir/$slicefamily/ 
#	echo "============================== mirroring done $(date)"
	# adapting slice images to the node virtualization mode can unfortunately
	# not be done at build-time since we do not know that yet
	patch_lvref $lxc_dir/$slicefamily
    done

    # create ref images from stubs
    unfold_system_slices

    echo "lxc-sliceimage: --- DONE $(date) ---"
}

function status () {
    echo -n "Checking node .. "
    check_node
    echo OK
    echo "From installed sliceimage variants"
    ls $sliceimage_dir
    echo "Exported to lxc"
    ls $lxc_dir
}

# prefer to use e.g.
# sed -i $rootfs/etc/sudoers
# rather than
# chroot $rootfs sed -i /etc/sudoers
# because the latter has sed issue weird messages like
# sed: warning: failed to get security context of ...
function patch_lvref () {
    rootfs=$1; shift

    echo "patch_lvref : handling $rootfs"
    
    # Allow a user who logins via ssh to sudo
    sed -i 's/^Defaults\ *requiretty/\#&/' $rootfs/etc/sudoers
    # Allow to login at virsh console. loginuid.so doen't work in the absence of auditd
    # which cannot run inside a container.
    sed -i 's/^.*loginuid.so.*$/\#&/' $rootfs/etc/pam.d/login
    # Enable eth0 on bootup
    cat <<EOF > $rootfs/etc/sysconfig/network-scripts/ifcfg-eth0
DEVICE=eth0
BOOTPROTO=dhcp
ONBOOT=yes
EOF
    # Tweak sshd configuration - not all slice images have sshd preinstalled
    if [ -f $rootfs/etc/ssh/sshd_config ] ; then
	sed -e 's/^UsePAM\ *yes/\#&/' \
	    -e 's/^GSSAPIAuthentication.*$/GSSAPIAuthentication no/' \
	    -e 's/^PasswordAuthentication.*$/PasswordAuthentication no/' \
	    -i $rootfs/etc/ssh/sshd_config
    fi
    # Allow root to login at virsh console
    echo "pts/0" >> $rootfs/etc/securetty
    # our image does not have NetworkManager, only network, and it is off by default
    chroot $rootfs chkconfig network on

    # this will run only if the rootfs seems to be systemd-based
    patch_lvref_systemd $rootfs
}

# beware that patch_lvref_systemd will return early on non systemd-powered images
# inspired from https://gist.github.com/1142202
ETC=/etc/systemd/system
LIB=/lib/systemd/system

function patch_lvref_systemd () {
    rootfs=$1; shift

    absetc=$rootfs/$ETC
    abslib=$rootfs/$LIB

    # this applies to systemd only
    [ -d $abslib ] || return

    # stay away from trouble, avoid chroot'ing for straightforward stuff
    echo "patch_lvref_systemd: handling $rootfs"
    # sliceimage comes with graphical.target as default
    ln -sf $LIB/multi-user.target $absetc/default.target
    # sysinit.target seems to stick on boot, so disable it. However, we need
    # systemd-tmpfiles-setup.service that was started by the dependency of
    # sysinit.target to boot up correctly, so start it instead.
    cp $abslib/basic.target $absetc/basic.target
    sed -e 's/sysinit.target/systemd-tmpfiles-setup.service/' -i $absetc/basic.target
    # Stop starting sysinit.target. Symlinking one to /dev/null is a standard way
    # to disable a target (or a service and others).
    ln -sf /dev/null $absetc/sysinit.target
    # is also a cause of stuck on boot
    ln -sf /dev/null $absetc/udev-settle.service
    # It prevents systemd-tmpfiles-setup.service from starting
    ln -sf /dev/null $absetc/fedora-readonly.service
    # Libvirt lxc provides only tty1
    rm -f $absetc/getty.target.wants/getty\@tty{2,3,4,5,6}.service
    # It launches sulogin on console(tty1) but it interferes getty@tty1
    ln -sf /dev/null $absetc/console-shell.service
    # Workarounds for libvirt 0.9.4. Without this, getty@tty1 doen't launch
    # because a trigger event on tty1 doesn't happen.
    cp $abslib/getty\@.service $absetc/getty\@.service
    sed -e 's/^BindTo/\#&/' -i $absetc/getty\@.service
    ln -sf $ETC/getty\@.service $absetc/getty.target.wants/getty\@tty1.service

}

function unfold_system_slices () {
    for clonedstamp in $sliceimage_stubs/*/*.cloned; do
	unfold_system_slice_from_cloned $clonedstamp
    done
}

# untested - how does this behave when updating the slice reference image ?
# OTOH system slices probably already run at update-time 
# so it's kind of too late anyways
function unfold_system_slice_from_cloned () {
    clonedstamp=$1; shift
    # e.g. NAME=planetflow
    NAME=$(basename $clonedstamp .cloned)
    DIR=$(dirname $clonedstamp)
    # e.g. SLICEFAMILY=planetlab-f8-i386
    SLICEFAMILY=$(cat $clonedstamp)
    # deduce the actual name used in .vref by replacing the first part of slice-family 
    # (pldistro) with the slice name
    # e.g. VREFNAME=planetflow-f8-i386
    VREFNAME=$(echo $SLICEFAMILY | sed -e "s,^[^-]*-,$NAME-,")

    VREFPATH="$lxc_dir/$VREFNAME"

    # do not redo existing vref's unless force is mentioned
    [ -z "$FORCE" -a -d "$VREFPATH" ] && {
	echo "unfold_system_slice_from_cloned : $VREFPATH looks fine" ;
	return ;
    }
    [ -z "$FORCE" ] && message=Building
    [ -n "$FORCE" ] && message=Force-building

    # Copy base reference image
    echo -n $"$message slice image for $NAME in $VREFNAME: "

    # build the systemslice from the one it was originally cloned from
    FAMILYREF="$lxc_dir/$SLICEFAMILY"
    if [ ! -d "$FAMILYREF" ] ; then
	echo system slice from $clonedstamp - could not find reference $FAMILYREF - skipped
	return
    fi
    
    # safe side
    mkdir -p "$lxc_dir"
    # cleanup reference image if already exists
    [ -d "$VREFPATH" ] && btrfs subvolume delete "$VREFPATH"
    # extra safe
    [ -d "$VREFPATH" ] && rm -rf "$VREFPATH"

    # clone subvolume
    btrfs subvolume snapshot $FAMILYREF $VREFPATH
    
    # merge the stub with the reference to get the system slice
    (cd "$DIR/$NAME"/ && find . | cpio -m -d -u -p "$VREFPATH"/) || \
	{ echo "Could not apply stub $DIR/$NAME - skipping $clonedstamp" ; btrfs subvolume delete "$VREFPATH" ; return ; }

    # Clean RPM state
    rm -f "$VREFPATH/var/lib/rpm/__db"*

    echo Done
}

# run with the --tty option to see results, otherwise get that logged
case "$1" in
    --tty) shift;;
    *) exec >> /var/log/lxc-sliceimage 2>&1 ;;
esac

case "$1" in
    start|restart|reload)	start ; exit 0 ;;
    status)			status ; exit 0 ;;
    stop)			exit 0 ;;
    *)  echo $"Usage: $0 [--tty] {start|stop|status}"
        exit 1
        ;;
esac