#!/bin/sh
#  Copyright, 2012 Dustin Kirkland <kirkland@ubuntu.com>
#  Copyright, 2012 Scott Moser <smoser@ubuntu.com>
#  Copyright, 2012 Axel Heider
#
#  Based on scripts from
#    Sebastian P.
#    Nicholas A. Schembri State College PA USA
#    Axel Heider
#    Dustin Kirkland
#    Scott Moser
#
#
#  This program is free software: you can redistribute it and/or modify
#  it under the terms of the GNU General Public License as published by
#  the Free Software Foundation, either version 3 of the License, or
#  (at your option) any later version.
#
#  This program is distributed in the hope that it will be useful,
#  but WITHOUT ANY WARRANTY; without even the implied warranty of
#  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
#  GNU General Public License for more details.
#
#    You should have received a copy of the GNU General Public License
#    along with this program.  If not, see
#    <http://www.gnu.org/licenses/>.

case "$1" in
	# no pre-reqs
	prereqs) echo ""; exit 0;;
esac

. /scripts/functions

PATH=/usr/sbin:/usr/bin:/sbin:/bin
MYTAG="overlayroot"

TEMP_D="${TMPDIR:-/tmp}/${0##*/}.configs"
VARIABLES="overlayroot overlayroot_cfgdisk"

# generic settings
# ${ROOT} and ${rootmnt} are predefined by caller of this script. Note that
# the root fs ${rootmnt} it mounted readonly on the initrams, which fits
# nicely for our purposes.
root_rw=/media/root-rw
root_ro=/media/root-ro
ROOTMNT=${rootmnt} # use global name to indicate created outside this
OVERLAYROOT_DEBUG=0
# PERSIST_DIR will persist after pivot-root.  It is used for log file
# and for the mktemp-made password file in crypt.
PERSIST_DIR=/run/initramfs
if [ ! -d "$PERSIST_DIR" -a -d /dev/.initramfs ]; then
	PERSIST_DIR="/dev/.initramfs"
fi
if [ ! -d "$PERSIST_DIR" ]; then
	mkdir -p "$PERSIST_DIR" ||
		echo "WARNING: $MYTAG: failed to create ${PERSIST_DIR}" 1>&2
fi
LOG_FILE="${PERSIST_DIR}/${MYTAG}.log"

log() {
	"log_${1}_msg" "$MYTAG: $2";
	_debug "[$1]:" "$2"
}
log_fail() { log failure "$*"; }
log_success() { log success "$*"; }
log_warn() { log warning "$*"; }
fail() {
	[ $# -eq 0 ] || log_fail "$@";
	exit 0; # why do we exit success?
}
cleanup() {
	[ -d "${TEMP_D}" ] && rm -Rf "${TEMP_D}"
}
debug() {
	_debug "$@"
	[ "${OVERLAYROOT_DEBUG:-0}" = "0" ] && return
	echo "$MYTAG:" "$@"
}
_debug() {
	if [ "${DEBUG_BUSTED:-0}" ]; then
		{ echo "$@" >> "$LOG_FILE"; } 2>/dev/null ||
			{ DEBUG_BUSTED=1; log_warn "debug is busted"; }
	fi
}

safe_string() {
	local prev="$1" allowed="$2" cur=""
	[ -n "$prev" ] || return 1
	while cur="${prev#[${allowed}]}"; do
		[ -z "$cur" ] && return 0
		[ "$cur" = "$prev" ] && break
		prev="$cur"
	done
	return 1
}

parse_string() {
	# parse a key/value string like:
	# name=mapper,pass=foo,fstype=ext4,mkfs=1
	# set variables under namespace 'ns'.
	#  _RET_name=mapper
	#  _RET_pass=foo
	#  _RET_fstype=ext4
	# set _RET to the list of variables found
	local input="${1}" delim="${2:-,}" ns="${3:-_RET_}"
	local oifs="$IFS" tok="" keys="" key="" val=""

	set -f; IFS="$delim"; set -- $input; IFS="$oifs"; set +f;
	_RET=""
	for tok in "$@"; do
		key="${tok%%=*}"
		val="${tok#${key}}"
		val=${val#=}
		safe_string "$key" "0-9a-zA-Z_" ||
			{ debug "$key not a safe variable name"; return 1; }
		eval "${ns}${key}"='${val}' || return 1
		keys="${keys} ${ns}${key}"
	done
	_RET=${keys# }
	return
}
get_varval() { eval _RET='${'$1'}'; }
write_kernel_cmdline_cfg() {
	local cfg="$1" desc="${2:-kernel cmdline}"
	local cmdline="" var=""
	read cmdline < /proc/cmdline || return 1
	: > "${cfg}" || return
	set -f
	{
	echo "desc='${desc}'"
	for tok in $cmdline; do
		for var in $VARIABLES; do
			if [ "$tok" = "$var" ]; then
				log_warn "kernel param without value '${tok}'";
				continue;
			elif [ "${tok}" = "${var}=" ]; then
				echo "${var}=''";
			elif [ "${tok#${var}=}" != "${tok}" ]; then
				echo "${var}='${tok#${var}=}'"
			fi
		done
	done
	} >> "$cfg"
	set +f
}

wait_for_dev() {
	local dev="$1" timeout="${2:-0}"
	[ -b "$dev" ] && return 0
	[ "$timeout" = "0" ] && return 1
	# wait-for-root writes fstype to stdout, redirect to null
	wait-for-root "$dev" "$timeout" >/dev/null
}
crypto_setup() {
	local fstype="ext4" pass="" mapname="secure" mkfs="1" dev=""
	local timeout=0
	local entropy_sources="/proc/sys/kernel/random/boot_id /proc/sys/kernel/random/uuid /dev/urandom"
	local seed="" rootseed="$ROOTMNT/var/lib/urandom/random-seed"
	if [ ! -f "$rootseed" -a -f "$ROOTMNT/var/lib/systemd/random-seed" ]; then
		rootseed="$ROOTMNT/var/lib/systemd/random-seed"
	fi
	# Seed the psuedo random number generator with available seeds
	for seed in "/.random-seed" "$rootseed"; do
		[ -f "${seed}" ] ||
			{ debug "missing rng seed [${seed}]"; continue; }
		cat "${seed}" > /dev/urandom ||
			debug "failed seeding /dev/urandom from $seed"
	done
	# this does necessary crypto setup and sets _RET
	# to the appropriate block device (ie /dev/mapper/secure)

	# mkfs (default is 1):
	#  0: never create filesystem
	#  1: if pass is given and mount fails, create a new one
	#     if no pass given, create new
	#  2: if pass is given and mount fails, fail
	#     if no pass given, create new
	local options="$1"
	parse_string "${options}" ||
		{ log_fail "failed parsing '${options}'"; return 1; }

	fstype=${_RET_fstype:-${fstype}}
	pass=${_RET_pass:-${pass}}
	mapname=${_RET_mapname:-${mapname}}
	mkfs=${_RET_mkfs:-${mkfs}}
	dev=${_RET_dev:-${dev}}
	timeout=${_RET_timeout:-${timeout}}

	[ -n "$dev" ] ||
		{ log_fail "dev= argument not provided in '${options}'"; return 1; }

	short2dev "$dev" ||
		{ log_fail "failed to convert $dev to a device"; return 1; }
	dev="${_RET}"

	debug "fstype=${fstype} pass=${pass} mapname=${mapname}"
	debug "mkfs=${mkfs} dev=${dev} timeout=${timeout}"

	wait_for_dev "$dev" "$timeout" || {
		log_fail "crypt dev device $dev does not exist after ${timeout}s";
		return 1;
	}

	if [ -n "$pass" ]; then
		printf "%s" "$pass" |
			cryptsetup luksOpen "$dev" "$mapname" --key-file -
		if [ $? -eq 0 ]; then
			local tdev="/dev/mapper/$mapname"
			log_warn "reusing existing luks device at $dev"
			wait_for_dev "$tdev" 20 ||
				{ log_fail "$tdev did not appear"; return 1; }
			_RET_DEVICE="/dev/mapper/$mapname"
			return 0
		fi
		if [ "$mkfs" != "1" ]; then
			log_fail "luksOpen failed on $dev with mkfs=$mkfs";
			return 1;
		fi
		log_warn "re-opening $dev failed with mkfs=$mkfs will create new"
	else
		[ "$mkfs" = "0" ] &&
			{ log_fail "mkfs=0, but no password provided"; return 1; }
		entropy_sources="$entropy_sources $dev"
		local tmpf="" pass_file="${PERSIST_DIR}/${MYTAG}.passwd"
		tmpf=$(mktemp "${PERSIST_DIR}/${MYTAG}.XXXXXX") ||
			{ log_fail "failed creation of password file"; return 1; }
		stat -L /dev/* /proc/* /sys/* >"$tmpf" 2>&1 ||
			{ log_warn "could not seed with stat entropy [$entropy_sources]"; }
		head -c 4096 $entropy_sources >> "$tmpf" ||
			{ log_fail "failed reading entropy [$entropy_sources]"; return 1; }
		pass=$(sha512sum "$tmpf") ||
			{ log_fail "failed generation of password"; return 1; }
		pass=${pass%% *}
		printf "%s" "${pass}" > "$tmpf" ||
			{ log_fail "failed to record password to '$tmpf'"; return 1; }
		mv "$tmpf" "${pass_file}" ||
			{ log_fail "failed rename '$tmpf' to '${pass_file}'"; return 1; }
		log_debug "stored password in ${pass_file}"
	fi

	log_warn "setting up new luks device at $dev"
	# clear backing device
	wipefs -a "$dev" ||
		{ log_fail "failed to wipe $dev"; return 1; }
	printf "%s" "$pass" | cryptsetup luksFormat "$dev" --key-file - ||
		{ log_fail "luksFormat $dev failed"; return 1; }
	printf "%s" "$pass" |
		cryptsetup luksOpen "$dev" "$mapname" --key-file - ||
		{ log_fail "luksOpen $dev failed"; return 1; }
	mke2fs -t "${fstype}" "/dev/mapper/${mapname}" || {
		log_fail "failed to mkfs -t $fstype on map $mapname";
		return 1;
	}

	_RET_DEVICE="/dev/mapper/$mapname"
	return 0
}

dev_setup() {
	local options="$1" dev="" timeout=0 path="/"
	# options supported:
	#    dev=device,timeout=X,path=/
	parse_string "${options}" ||
		{ log_fail "failed parsing '${options}'"; return 1; }

	dev=${_RET_dev:-${dev}}
	timeout=${_RET_timeout:-${timeout}}

	[ -n "$dev" ] ||
		{ log_fail "dev= argument not provided in '${options}'"; return 1; }

	short2dev "$dev" ||
		{ log_fail "failed to convert $dev to a device"; return 1; }
	dev="${_RET}"

	debug "dev=${dev} timeout=${timeout}"

	wait_for_dev "$dev" "$timeout"
	_RET_DEVICE="$dev"
}

clean_path() {
	# remove '//' in a path.
	local p="$1" tmp=""
	while [ "${p#*//}" != "$p" ]; do
		tmp=${p#*//}
		p="${p%%//*}/${tmp}"
	done
	_RET="$p"
}

get_workdir() {
	local root_rw="$1" dir_prefix="$2" file="$3"
	file=${file%/}
	if [ "$file" = "/" -o "$file" = "" ]; then
		file="_"
	fi
	clean_path "${root_rw}/${dir_prefix%/}-workdir/${file}"
}

gen_fstab() {
	# gen_fstab(mp) - generate a fabricated /etc/fstab for mount point $mp
	local mp="$1" m_spec="" m_opts="" m_fstype=""
	local spec file vfstype opts pass freq

	# just remove any trailing /
	[ "$mp" != "/" ] && mp=${mp%/}
	while read spec file vfstype opts pass freq; do
		[ "$file" = "$mp" ] && {
			m_spec="$spec"
			m_opts="$opts"
			m_fstype="$vfstype"
			break
		}
	done </proc/mounts
	[ -z "$m_spec" ] && {
		log_warn "did not find root mount point $mp in /proc/mounts"
		m_opts="/dev/root"
		m_opts="defaults"
		m_fstype="auto"
	}

	echo "# fabricated by overlayroot, rootfs did not contain /etc/fstab"
	echo "${m_spec} / ${m_fstype} ${m_opts} 0 0 "
}

overlayrootify_fstab() {
	# overlayrootify_fstab(input, root_ro, root_rw, dir_prefix, recurse, swap)
	# read input fstab file, write an overlayroot version to stdout
	# also returns (_RET) a list of directories that will need to be made
	local input="$1" root_ro="${2:-/media/root-ro}"
	local root_rw="${3:-/media/root-rw}" dir_prefix="${4:-/}"
	local recurse=${5:-1} swap=${6:-0} fstype=${7:-overlayfs}
	local hash="#" oline="" ospec="" upper="" dirs="" copy_opts="" copy_opt=""
	local spec file vfstype opts pass freq line ro_line
	local workdir="" use_orig="" relfile="" needs_workdir=false noauto=""
	
	[ -f "$input" ] || return 1

    cat <<EOF
#
#  This fstab is for overlayroot. The real one can be found at
#  ${root_ro}/etc/fstab
#  The original entry for '/' and other mounts have been updated to be placed
#  under $root_ro.
#  To permanently modify this (or any other file), you should change-root into
#  a writable view of the underlying filesystem using:
#      sudo overlayroot-chroot
#
EOF

	needs_workdir && needs_workdir=true || needs_workdir=false
	while read spec file vfstype opts pass freq; do
		[ "$file" = "/" ] && noauto="noauto" || noauto=""
		line="$spec $file $vfstype $opts $pass $freq"
		case ",$opts," in
			*,ro,*) ro_opts="$opts";;
			*) ro_opts="ro,${opts}";;
		esac
		ro_line="$spec ${root_ro}$file $vfstype"
		ro_line="${ro_line} ${ro_opts}${noauto:+,${noauto}} $pass $freq"

		use_orig=""
		if [ "${spec#${hash}}" != "$spec" ]; then
			use_orig="comment"
		elif [ -z "$freq" ]; then
			use_orig="malformed-line"
		else
			case "$vfstype" in
				vfat|fat) use_orig="fs-unsupported";;
				proc|sys|tmpfs|dev|udev|debugfs) use_orig="fs-virtual";;
			esac
		fi

		if [ -n "$use_orig" ]; then
			if [ "$use_orig" != "comment" ]; then
				echo "$line # $MYTAG:$use_orig"
			else
				echo "$line"
			fi
		elif [ "$vfstype" = "swap" ]; then
			if [ "$swap" = "0" ]; then
				# comment out swap lines
				echo "#$MYTAG:swap=${swap}#${line}"
			elif [ "${spec#/}" != "${spec}" ] &&
			     [ "${spec#/dev/}" = "${spec}" ]; then
				# comment out swap files (spec starts with / and not in /dev)
				echo "#$MYTAG:swapfile#${line}"
			else
				echo "${line}"
			fi
		else
			ospec="${root_ro}${file}"
			copy_opts=""
			for copy_opt in nobootwait noauto nofail; do
				case ",$opts," in
					*,${copy_opt},*)
						copy_opts="${copy_opts},${copy_opt}";;
				esac
			done

			clean_path "${root_rw}/${dir_prefix}${file}"
			upper="$_RET"

			oline="${ospec} ${file} $fstype "
			clean_path "${root_ro}${file}"
			oline="${oline}lowerdir=$_RET"
			oline="${oline},upperdir=${upper}${copy_opts}"
			if [ "$fstype" = "overlayfs" -o "$fstype" = "overlay" ] &&
				${needs_workdir}; then
				get_workdir "$root_rw" "$dir_prefix" "$file"
				workdir="$_RET"
				oline="${oline},workdir=$workdir"
				dirs="${dirs} $workdir"
			fi
 			oline="${oline} $pass $freq"

			if [ "$recurse" != "0" -o "$file" = "/" ]; then
				if [ "$file" = "/" ]; then
					# this can confuse systemd (LP: #1723183)
					echo "#$ro_line"
				else
					echo "$ro_line"
				fi
				echo "$oline"
				dirs="${dirs} ${upper}"
			else
				echo "$line"
			fi
		fi
	done < "$input"
	_RET=${dirs# }
}

short2dev() {
	# turn 'LABEL=' or 'UUID=' into a device path
	# also support /dev/* and 'vdb' or 'xvda'
	local input="$1" oifs="$IFS" dev newdev s
	case "$input" in
		LABEL=*)
			dev="${input#LABEL=}"
			case "${dev}" in
				*/*) dev="$(echo "${dev}" | sed 's,/,\\x2f,g')";;
			esac
			dev="/dev/disk/by-label/${dev}"
			;;
		UUID=*) dev="/dev/disk/by-uuid/${input#UUID=}" ;;
		/dev/*) dev="${input}";;
		*) dev="/dev/${input}";;
	esac
	_RET=$dev
}

get_cfg() {
	# get_cfg(device, file, cfg, timeout=0)
	# copy the file $cfg off $device to $file, waiting $timeout for
	# $device to appear
	local dev="$1" file="$2" cfg="${3:-overlayroot.conf}" timeout="${4:-0}"
	local cfgdev="" success=0 didmnt=false mp="" pre="get_cfg($dev):"
	[ -z "$dev" ] && return 1

	if [ "$timeout" != "0" ]; then
		wait_for_dev "$dev" "$timeout" ||
			{ debug "$pre did not appear in $timeout"; return 1; }
	else
		udevadm settle
	fi

	short2dev "$dev" && cfgdev="$_RET" && [ -b "$cfgdev" ] ||
		{ debug "$pre not present"; return 1; }

	if mp="${TEMP_D}/mp" && mkdir "$mp" &&
		mount -o ro "$cfgdev" "$mp"; then
		if [ -f "$mp/$cfg" ]; then
			cp "$mp/$cfg" "$file" && success=1 ||
				debug "$pre copy failed"
		else
			debug "$pre '$file' file not found"
		fi
		umount "$mp"
	else
		debug "$pre mount failed"
	fi
	[ -d "$mp" ] || rmdir "$mp"
	[ $success -eq 1 ] || return 1
	_RET="$dev/$cfg"
}

parse_cfg() {
	# parse_cfg($file,$desc,$vars)
	# this reasonably safely sources the "config" file $file
	# and then declares the variables listed to stdout
	#
	# known issues:
	# * sourced file could re-define 'echo'
	# * just fails if a value has a ' in it
	[ -f "$1" ] || return 0
	tick="'" sh -c '
		__tick=$tick
		__file=$1; __desc=$2;
		__unset__="__unset__"
		shift 2
		__vars="$*"
		readonly __tick __file __desc __vars
		. "${__file}"
		if [ $? -ne 0 ]; then
			echo "failed source \"${__file}\" ($__desc)" 1>&2
			return 1
		fi
		set -e
		echo "desc=${__tick}${__desc}${__tick}"
		for var in ${__vars}; do
			eval val=\${${var}-${__unset__}} ||
				{ echo "eval of $var failed"; exit 1; }
			[ "${val#*${__tick}}" = "${val}" ] || {
				echo "variable \"$var\" had single tick in it. fail";
				exit 1;
			}
			[ "${val}" = "${__unset__}" ] ||
				echo "$var=${__tick}${val}${__tick}"
		done
	' -- "$@"
}

readcfgd() {
	local cfg_d="$1"
	# this kind of stinks, each VARIABLE goes into global scope
	local or="" or_cfgdisk=""
	_RET_desc_oroot=""
	_RET_desc_cfgdisk=""
	set +f
	for f in "${cfg_d}"/*; do
		. "$f" || fail "failed reading $f"
		[ "$or" != "${overlayroot}" ] &&
			_RET_desc_oroot="$desc"
		[ "${or_cfgdisk}" != "${overlayroot_cfgdisk}" ] &&
			_RET_desc_cfgdisk="$desc"
		or=${overlayroot}
		or_cfgdisk=${overlayroot_cfgdisk}
	done
	set -f
}

fix_upstart_overlayfs() {
	# inotify events on an overlayfs underlay do not propogate through
	# a newly created overlay.  This causes job creation and deletion issues
	# for upstart, which uses inotify on /etc/init. So ensure that the overlay
	# explicitly has a /etc/init (LP: #1213925)
	local root="$1"
	local initctl="$root/sbin/initctl"
	local eifile="/etc/init/.overlayfs-upstart-helper"

	[ -e "$initctl" -o -L "$initctl" ] || return 0
	[ -d "$root/etc/init" ] || return 0
	echo "#upstart needs help for overlayfs (LP: #1213925)." \
		> "$root/$eifile" ||
		{ log_fail "failed to write $root/$eifile"; return 1; }
	debug "created $eifile under $root (LP: #1213925)"
}

needs_workdir() {
	local uname_r="${UNAME_R}"
	if [ -z "$uname_r" ]; then
		uname_r=$(uname -r) && UNAME_R="$uname_r" ||
			{ log_warn "failed to read uname!"; return 2; }
	fi
	# 3.18+ require 'workdir=' option.
	case "${uname_r}" in
		2*|3.1[01234567]*|3.[0-9].*) return 1;;
		*) return 0;;
	esac
}


search_fs_driver() {
	# search_fs_driver(driver1, driver2...)
	# return the first supported filesystem driver in the list.
	# if no drivers are provided, search default list.
	# assumes that name in /proc/filesystems and module name are the same.
	local m="" driver="" tab='	'
	for m in "$@"; do
		grep -q "${tab}$m$" /proc/filesystems && driver="$m" &&
			{ debug "/proc/filesystems support for '$m'"; break; }
	done
	if [ -z "$driver" ]; then
		for m in "$@"; do
			if modprobe -qb "$m"; then
				if grep -q "${tab}$m$" /proc/filesystems; then
					debug "loaded module '$m'";
					driver="$m"
					break;
				else
					log_warn "loaded '$m' module but no '$m' in /proc/filesystems"
				fi
			fi
		done
	fi
	[ -n "$driver" ] ||
		{ debug "Unable to find driver/module. searched: $*"; return 1; }
	_RET="$driver"
}

cfgd="${TEMP_D}/configs"
mkdir -p "${cfgd}" || fail "failed to create tempdir"
trap cleanup EXIT

# collect the different config locations into a file
# write individual config files in $cfgd that contain
{
echo "desc='builtin'"
echo "overlayroot=''"
echo "overlayroot_cfgdisk='disabled'"
} > "$cfgd/00-builtin"

write_kernel_cmdline_cfg "${cfgd}/90-kernel" ||
	fail "failed to read kernel command line!"

parse_cfg "/conf/conf.d/overlayroot" "initramfs config" \
	"$VARIABLES" > "${cfgd}/10-initramfs" ||
	fail "failed parsing initramfs config"

parse_cfg "${ROOTMNT}/etc/overlayroot.conf" "$ROOT/etc/overlayroot.conf" \
	"$VARIABLES" > "${cfgd}/20-root-config" ||
	fail "failed parsing root config"

parse_cfg "${ROOTMNT}/etc/overlayroot.local.conf" \
	"$ROOT/etc/overlayroot.local.conf" \
	"$VARIABLES" > "${cfgd}/30-root-local-config" ||
	fail "failed parsing root config"

# now read the trusted configs, to see if overlayroot_cfgdisk is set
readcfgd "$cfgd" ||
	fail "reading configs failed"
used_desc="${_RET_desc_oroot}"

[ -n "${_RET_desc_cfgdisk}" ] &&
	debug "${_RET_desc_cfgdisk} set cfgdisk='${overlayroot_cfgdisk}'"

# if one of the trusted configs above gave us set the overlayroot_cfgdisk
# variable, then look for such a device and read a config from it.
cfgdisk=0
if [ "${overlayroot_cfgdisk:-disabled}" != "disabled" ] &&
	get_cfg "${overlayroot_cfgdisk}" "${TEMP_D}/cfgdisk"; then
	desc=${_RET}
	parse_cfg "${TEMP_D}/cfgdisk" "${desc}" "$VARIABLES" \
		> "${cfgd}/80-cfgdisk" ||
		fail "reading config from ${desc} failed"
	used_desc="${_RET_desc_oroot}"

	# now read the parsed configs again.
	readcfgd "$cfgd" ||
		fail "reading configs failed"

	[ -n "${_RET_desc_cfgdisk}" ] &&
		debug "${_RET_desc_cfgdisk} set cfgdisk=${overlayroot_cfgdisk}"
	[ -n "${_RET_desc_oroot}" ] &&
		debug "${_RET_desc_oroot} set overlayroot=${overlayroot}"
fi

opts=""
cfgmsg="${used_desc:+ (per ${used_desc})}"
case "${overlayroot:-disabled}" in
	tmpfs|tmpfs:*)
		mode="tmpfs"
		opts="${overlayroot#tmpfs}";
		opts=${opts#:}
		;;
	/dev/*|device:*)
		case "$overlayroot" in
			/dev/*) opts="dev=${overlayroot}";;
			*) opts="${overlayroot#device:}";;
		esac
		dev_setup "${opts}" ||
			fail "failed setup overlay for ${overlayroot} [$opts]${cfgmsg}"
		mode="device"
		device="$_RET_DEVICE"
		;;
	crypt:*)
		mode="crypt"
		opts=${overlayroot#crypt:}
		crypto_setup "${opts}" ||
			fail "failed setup crypt for ${overlayroot}${cfgmsg}"
		device="$_RET_DEVICE"
		;;
	disabled)
		debug "overlayroot disabled${cfgmsg}"
		exit 0;;
	*)
		fail "invalid value for overlayroot: ${overlayroot}${cfgmsg}"
		exit 0;;
esac

parse_string "$opts" "," _RET_common_
swap=${_RET_common_swap:-0}
recurse=${_RET_common_recurse:-1}
OVERLAYROOT_DEBUG=${_RET_common_debug:-${OVERLAYROOT_DEBUG}}
dir_prefix=${_RET_common_dir:-"/overlay"}
overlayroot_driver=${_RET_common_driver}

read cmdline </proc/cmdline
cmdline=" $cmdline "
[ "${cmdline#* ro }" != "$cmdline" ] && cmdline_ro=true || cmdline_ro=false

if [ "${overlayroot_driver:-auto}" = "auto" ]; then
	search_fs_driver overlay overlayfs ||
		fail "Unable to find a driver. searched: overlay overlayfs"
	overlayroot_driver="$_RET"
else
	search_fs_driver ${overlayroot_driver} ||
		fail "Unable to find a driver named '${overlayroot_driver}'"
	overlayroot_driver="$_RET"
fi

debug "swap=$swap recurse=$recurse debug=$OVERLAYROOT_DEBUG dir=$dir_prefix"
debug "device=$device mode=$mode driver=${overlayroot_driver}"

[ "$swap" = "0" -o "$swap" = "1" ] ||
	fail "invalid setting for swap: $swap. must be '0' or '1'"
[ "$recurse" = "0" -o "$recurse" = "1" ] ||
	fail "invalid setting for recurse: $recurse. must be '0' or '1'"

log_warn "configuring overlayroot with driver=${overlayroot_driver} mode=$mode opts='$opts' per $used_desc"

# settings based on overlayroot_driver
workdir=""
case "${overlayroot_driver}" in
	overlayfs|overlay)
		mount_type="${overlayroot_driver}"
		mount_opts="-o lowerdir=${root_ro},upperdir=${root_rw}/${dir_prefix}"
		if needs_workdir; then
			get_workdir "$root_rw" "$dir_prefix" "/"
			workdir="$_RET"
			mount_opts="${mount_opts},workdir=$workdir"
		fi
		clean_path "${mount_opts} overlayroot ${ROOTMNT}"
		mount_opts="$_RET"
		;;
	aufs)
		mount_type="aufs"
		mount_opts="-o dirs=${root_rw}/${dir_prefix}:${root_ro}=ro aufs-root ${ROOTMNT}"
		;;
	*)
		log_fail "invalid overlayroot driver: ${overlayroot_driver}"
		panic "$MYTAG"
		;;
esac

# make the mount point on the init root fs ${root_rw}
mkdir -p "${root_rw}" ||
	fail "failed to create ${root_rw}"

# make the mount point on the init root fs ${root_ro}
mkdir -p "${root_ro}" ||
	fail "failed to create ${root_ro}"

# mount the backing device to $root_rw
if [ "$mode" = "tmpfs" ]; then
	# mount a tmpfs using the device name tmpfs-root
	mount -t tmpfs tmpfs-root "${root_rw}" ||
		fail "failed to create tmpfs"
else
	# dev or crypto
	mount "$device" "${root_rw}" ||
		fail "failed mount backing device $device"
fi

mkdir -p "${root_rw}/${dir_prefix}" ||
	fail "failed to create ${dir_prefix} on ${device}"

[ -z "$workdir" ] || mkdir -p "$workdir" ||
	fail "failed to create workdir '$workdir' on ${device}"

# root is mounted on ${ROOTMNT}, move it to ${ROOT_RO}.
mount --move "${ROOTMNT}" "${root_ro}" ||
	fail "failed to move root away from ${ROOTMNT} to ${root_ro}"

# there is nothing left at ${ROOTMNT} now. So for any error we get we should
# either do recovery to restore ${ROOTMNT} for drop to a initramfs shell using
# "panic". Otherwise the boot process is very likely to fail with even more
# errors and leave the system in a wired state.

# mount virtual fs ${ROOTMNT} with rw-fs ${root_rw} on top or ro-fs ${root_ro}.
debug mount -t "$mount_type" $mount_opts
mount -t "$mount_type" $mount_opts
if [ $? -ne 0 ]; then
	log_fail "failed to create new ro/rw layered ${ROOTMNT}"
	log_fail "mount command was: mount -t $mount_type $mount_opts"
	# do recovery and try restoring the mount for ${ROOTMNT}
	mount --move ${root_ro} ${ROOTMNT}
	if [ $? -ne 0 ]; then
		# thats bad, drop to shell to let the user try fixing this
		log_fail "RECOVERY_ERROR: failed to move $root_ro back to ${ROOTMNT}"
		panic "$MYTAG"
	fi
	exit 0
fi

# now the real root fs is on ${root_ro} of the init file system, our
# layered root fs is set up at ${ROOTMNT}. So we can write anywhere in
# {ROOTMNT} and the changes will end up in ${root_rw} while ${root_ro} it
# not touched. However ${root_ro} and ${root_rw} are on the initramfs root
# fs, which will be removed an replaced by ${ROOTMNT}. Thus we must move
# ${root_ro} and ${root_rw} to the rootfs visible later, ie.
# ${ROOTMNT}${root_ro} and ${ROOTMNT}${root_ro}.  Since the layered ro/rw
# is already up, these changes also end up on ${root_rw} while ${root_ro}
# is not touched.

# move mount from ${root_ro} to ${ROOTMNT}${root_ro}
mkdir -p "${ROOTMNT}/${root_ro}"
mount --move ${root_ro} "${ROOTMNT}${root_ro}" ||
	fail "failed to move ${root_ro} to ${ROOTMNT}${root_ro}"

# move mount from ${root_rw} to ${ROOTMNT}${root_rw}
[ -d ${ROOTMNT}${root_rw} ] || mkdir -p ${ROOTMNT}${root_rw}
mount --move "${root_rw}" "${ROOTMNT}${root_rw}" ||
	fail "failed to move ${root_rw} to ${ROOTMNT}${root_rw}"

# technically, everything is set up nicely now. Since ${ROOTMNT} had beend
# mounted read-only on the initfamfs already, ${ROOTMNT}${root_ro} is it,
# too.  Now we init process could run - but unfortunately, we may have to
# prepare some more things here.
# Basically, there are two ways to deal with the read-only root fs. If the
# system is made aware of this, things can be simplified a lot.  If it is
# not, things need to be done to our best knowledge.
#
# So we assume here, the system does not really know about our read-only
# root fs.
#
# Let's deal with /etc/fstab first. It usually contains an entry for the
# root fs, which is no longer valid now. We have to remove it and add our
# new ${root_ro} entry.
# Remember we are still on the initramfs root fs here, so we have to work
# on ${ROOTMNT}/etc/fstab. The original fstab is
# ${ROOTMNT}${root_ro}/etc/fstab.
source_fstab="$ROOTMNT/${root_ro}/etc/fstab"
if [ ! -f "$source_fstab" ] && [ "$cmdline_ro" = "true" ]; then
	debug "rootfs did not contain /etc/fstab, creating" \
		"based on kernel cmdline"
	source_fstab="${TEMP_D}/fstab"
	gen_fstab "${ROOTMNT}${root_ro}" > "$source_fstab" ||
		log_fail "gen_fstab failed for $ROOTMNT/$root_ro"
fi

if [ -f "$source_fstab" ]; then
	overlayrootify_fstab "$source_fstab" "$root_ro" \
		"$root_rw" "$dir_prefix" "$recurse" "$swap" "${overlayroot_driver}" \
		> "${ROOTMNT}/etc/fstab" ||
		log_fail "failed to modify /etc/fstab"
else
	debug "rootfs did not contain /etc/fstab."
fi

# we have to make the directories in ${root_rw} because if they do not
# exist, then the 'upper=' argument to overlayfs will fail.
for d in ${_RET}; do
	mkdir -p "${ROOTMNT}/$d"
done

if [ "${overlayroot_driver}" = "overlayfs" ]; then
	fix_upstart_overlayfs "$ROOTMNT" ||
		log_fail "failed to fix upstart for overlayfs"
fi

# if / is supposed to be mounted read-only (cmdline with 'ro')
# then mount our overlayfs as read-only just to be more normal
if [ "${cmdline_ro}" = "true" ]; then
	mount -o remount,ro "$ROOTMNT" ||
		log_fail "failed to remount overlayroot read-only"
	debug "mounted $ROOTMNT read-only per kernel cmdline"
fi

msg="configured root with '$overlayroot' using ${overlayroot_driver} per"
msg="$msg ${used_desc}"
log_success "$msg"

exit 0
# vi: ts=4 noexpandtab