Alpine Linux Porting (一点九?)

最近睡太少身体开始亮红灯Orz
来逐步解剖一下Alpine initramfs的init script当作知识储备：

/bin/busybox mkdir -p /usr/bin /usr/sbin /proc /sys /dev $sysroot \
        /media/cdrom /media/usb /tmp /run/cryptsetup

# Spread out busybox symlinks and make them available without full path
/bin/busybox --install -s
export PATH=/usr/bin:/bin:/usr/sbin:/sbin

# Make sure /dev/null is a device node. If /dev/null does not exist yet, the command
# mounting the devtmpfs will create it implicitly as an file with the "2>" redirection.
# The -c check is required to deal with initramfs with pre-seeded device nodes without
# error message.
[ -c /dev/null ] || mknod -m 666 /dev/null c 1 3

mount -t sysfs -o noexec,nosuid,nodev sysfs /sys
mount -t devtmpfs -o exec,nosuid,mode=0755,size=2M devtmpfs /dev 2>/dev/null \
	|| mount -t tmpfs -o exec,nosuid,mode=0755,size=2M tmpfs /dev

# Make sure /dev/kmsg is a device node. Writing to /dev/kmsg allows the use of the
# earlyprintk kernel option to monitor early init progress. As above, the -c check
# prevents an error if the device node has already been seeded.
[ -c /dev/kmsg ] || mknod -m 660 /dev/kmsg c 1 11

mount -t proc -o noexec,nosuid,nodev proc /proc
# pty device nodes (later system will need it)
[ -c /dev/ptmx ] || mknod -m 666 /dev/ptmx c 5 2
[ -d /dev/pts ] || mkdir -m 755 /dev/pts
mount -t devpts -o gid=5,mode=0620,noexec,nosuid devpts /dev/pts

# shared memory area (later system will need it)
[ -d /dev/shm ] || mkdir /dev/shm
mount -t tmpfs -o nodev,nosuid,noexec shm /dev/shm

上面起头的逻辑相对的非常简单，busybox把该要的directory建好，然後自我安装。
接下来把基础的sysfs跟devtmpfs拉起来，以供後续的mounting跟housekeeping使用。
接着建好pty的char dev，这样以後才有办法作pty的population。

# read the kernel options. we need surve things like:
#  acpi_osi="!Windows 2006" xen-pciback.hide=(01:00.0)
set -- $(cat /proc/cmdline)

myopts="alpine_dev autodetect autoraid chart cryptroot cryptdm cryptheader cryptoffset
	cryptdiscards cryptkey debug_init dma init init_args keep_apk_new modules ovl_dev
	pkgs quiet root_size root usbdelay ip alpine_repo apkovl alpine_start splash
	blacklist overlaytmpfs rootfstype rootflags nbd resume s390x_net dasd ssh_key
	BOOTIF zfcp"

for opt; do
	case "$opt" in
	s|single|1)
		SINGLEMODE=yes
		continue
		;;
	console=*)
		opt="${opt#*=}"
		KOPT_consoles="${opt%%,*} $KOPT_consoles"
		continue
		;;
	esac

	for i in $myopts; do
		case "$opt" in
		$i=*)	eval "KOPT_${i}"='${opt#*=}';;
		$i)	eval "KOPT_${i}=yes";;
		no$i)	eval "KOPT_${i}=no";;
		esac
	done
done

如注解所说，上面是init script正在把kernel boot arg爬下来、转换成它自己的syntax (KOPT_xxx)，後面要根据这些option做事。

echo "Alpine Init $VERSION" > /dev/kmsg
[ "$KOPT_quiet" = yes ] || echo "Alpine Init $VERSION"

# enable debugging if requested
[ -n "$KOPT_debug_init" ] && set -x

# set default values
: ${KOPT_init:=/sbin/init}

# pick first keymap if found
for map in /etc/keymap/*; do
	if [ -f "$map" ]; then
		ebegin "Setting keymap ${map##*/}"
		zcat "$map" | loadkmap
		eend
		break
	fi
done

# start bootcharting if wanted
if [ "$KOPT_chart" = yes ]; then
	ebegin "Starting bootchart logging"
	/sbin/bootchartd start-initfs "$sysroot"
	eend 0
fi

# The following values are supported:
#   alpine_repo=auto         -- default, search for .boot_repository
#   alpine_repo=http://...   -- network repository
ALPINE_REPO=${KOPT_alpine_repo}
[ "$ALPINE_REPO" = "auto" ] && ALPINE_REPO=

# hide kernel messages
[ "$KOPT_quiet" = yes ] && dmesg -n 1

# optional blacklist
for i in ${KOPT_blacklist//,/ }; do
	echo "blacklist $i" >> /etc/modprobe.d/boot-opt-blacklist.conf
done

# determine if we are going to need networking
if [ -n "$KOPT_ip" ] || [ -n "$KOPT_nbd" ] || \
	is_url "$KOPT_apkovl" || is_url "$ALPINE_REPO"; then

	do_networking=true
else
	do_networking=false
fi

Alpine init的debug log正式出现，开始处理initramfs的option。
其中最重要的是 alpine_repo ，Alpine原始状态下，会使用该option指向的 repo mirror，去抓取最新（或指定版本）的package下来组他真正的rootfs。
然而如果它指向的是一个网路URI 例如：
http://alpine.ccns.ncku.edu.tw/alpine/edge/main/
他就会需要作网路设定的逻辑，本文後面会提及。

# make sure we load zfs module if root=ZFS=...
rootfstype=${KOPT_rootfstype}
if [ -z "$rootfstype" ]; then
	case "$KOPT_root" in
	ZFS=*) rootfstype=zfs ;;
	esac
fi

# load available drivers to get access to modloop media
ebegin "Loading boot drivers"

modprobe -a $(echo "$KOPT_modules $rootfstype" | tr ',' ' ' ) loop squashfs 2> /dev/null
if [ -f /etc/modules ] ; then
	sed 's/\#.*//g' < /etc/modules |
	while read module args; do
		modprobe -q $module $args
	done
fi
eend 0

if [ -n "$KOPT_cryptroot" ]; then
	cryptopts="-c ${KOPT_cryptroot}"
	if [ "$KOPT_cryptdiscards" = "yes" ]; then
		cryptopts="$cryptopts -D"
	fi
	if [ -n "$KOPT_cryptdm" ]; then
		cryptopts="$cryptopts -m ${KOPT_cryptdm}"
	fi
	if [ -n "$KOPT_cryptheader" ]; then
		cryptopts="$cryptopts -H ${KOPT_cryptheader}"
	fi
	if [ -n "$KOPT_cryptoffset" ]; then
		cryptopts="$cryptopts -o ${KOPT_cryptoffset}"
	fi
	if [ "$KOPT_cryptkey" = "yes" ]; then
		cryptopts="$cryptopts -k /crypto_keyfile.bin"
	elif [ -n "$KOPT_cryptkey" ]; then
		cryptopts="$cryptopts -k ${KOPT_cryptkey}"
	fi
fi

if [ -n "$KOPT_nbd" ]; then
	# TODO: Might fail because nlplug-findfs hasn't plugged eth0 yet
	configure_ip
	setup_nbd || echo "Failed to setup nbd device."
fi

# zpool reports /dev/zfs missing if it can't read /etc/mtab
ln -s /proc/mounts /etc/mtab

# check if root=... was set
if [ -n "$KOPT_root" ]; then
	if [ "$SINGLEMODE" = "yes" ]; then
		echo "Entering single mode. Type 'exit' to continue booting."
		sh
	fi

	ebegin "Mounting root"
	nlplug-findfs $cryptopts -p /sbin/mdev ${KOPT_debug_init:+-d} \
		$KOPT_root

	if echo "$KOPT_modules $rootfstype" | grep -qw btrfs; then
		/sbin/btrfs device scan >/dev/null || \
			echo "Failed to scan devices for btrfs filesystem."
	fi

	if [ -n "$KOPT_resume" ]; then
		echo "Resume from disk"
		if [ -e /sys/power/resume ]; then
			printf "%d:%d" $(stat -Lc "0x%t 0x%T" "$KOPT_resume") >/sys/power/resume
		else
			echo "resume: no hibernation support found"
		fi
	fi

	if [ "$KOPT_overlaytmpfs" = "yes" ]; then
		mkdir -p /media/root-ro /media/root-rw $sysroot/media/root-ro \
			$sysroot/media/root-rw
		mount -o ro $KOPT_root /media/root-ro
		mount -t tmpfs root-tmpfs /media/root-rw
		mkdir -p /media/root-rw/work /media/root-rw/root
		mount -t overlay -o lowerdir=/media/root-ro,upperdir=/media/root-rw/root,workdir=/media/root-rw/work overlayfs $sysroot
	else
		if [ "$rootfstype" = "zfs" ]; then
			prepare_zfs_root
		fi
		mount ${rootfstype:+-t} ${rootfstype} \
			-o ${KOPT_rootflags:-ro} \
			${KOPT_root#ZFS=} $sysroot
	fi

	eend $?
	cat /proc/mounts | while read DEV DIR TYPE OPTS ; do
		if [ "$DIR" != "/" -a "$DIR" != "$sysroot" -a -d "$DIR" ]; then
			mkdir -p $sysroot/$DIR
			mount -o move $DIR $sysroot/$DIR
		fi
	done
	sync
	exec /bin/busybox switch_root $sysroot $chart_init "$KOPT_init" $KOPT_init_args
	echo "initramfs emergency recovery shell launched"
	exec /bin/busybox sh
fi

然而如果bootarg有指定rootdev的 root=/path/to/rootdev，那便是Alpine的世界观里有permanant installation的状态，那就会在处理完rootfs type（ZFS/有没有加密过/是否有透过overlayfs去打造读写分离...）後，便会直接mount好、并且switch_root过去。

而这边有个Alpine重要的辅助tool在上篇有提及，那就是 nlplug-findfs ，下篇将会深入来看他的选项与逻辑，这边先想成威力加强版的mount就好。

if $do_networking; then
	repoopts="-n"
else
	repoopts="-b $repofile"
fi

# locate boot media and mount it
ebegin "Mounting boot media"
nlplug-findfs $cryptopts -p /sbin/mdev ${KOPT_debug_init:+-d} \
	${KOPT_usbdelay:+-t $(( $KOPT_usbdelay * 1000 ))} \
	$repoopts -a /tmp/apkovls
eend $?

# Setup network interfaces
if $do_networking; then
        configure_ip
fi

# early console?
if [ "$SINGLEMODE" = "yes" ]; then
	echo "Entering single mode. Type 'exit' to continue booting."
	sh
fi

# mount tmpfs sysroot
rootflags="mode=0755"
if [ -n "$KOPT_root_size" ]; then
	echo "WARNING: the boot option root_size is deprecated. Use rootflags instead"
	rootflags="$rootflags,size=$KOPT_root_size"
fi
if [ -n "$KOPT_rootflags" ]; then
	rootflags="$rootflags,$KOPT_rootflags"
fi

mount -t tmpfs -o $rootflags tmpfs $sysroot

if [ -z "$KOPT_apkovl" ]; then
	# Not manually set, use the apkovl found by nlplug
	if [ -e /tmp/apkovls ]; then
		ovl=$(head -n 1 /tmp/apkovls)
	fi
elif is_url "$KOPT_apkovl"; then
	# Fetch apkovl via network
	MACHINE_UUID=$(cat /sys/class/dmi/id/product_uuid 2>/dev/null)
	url="${KOPT_apkovl/{MAC\}/$MAC_ADDRESS}"
	url="${url/{UUID\}/$MACHINE_UUID}"
	ovl=/tmp/${url##*/}
	wget -O "$ovl" "$url" || ovl=
else
	ovl="$KOPT_apkovl"
fi

# parse pkgs=pkg1,pkg2
if [ -n "$KOPT_pkgs" ]; then
	pkgs=$(echo "$KOPT_pkgs" | tr ',' ' ' )
fi

# load apkovl or set up a minimal system
if [ -f "$ovl" ]; then
	ebegin "Loading user settings from $ovl"
	# create apk db and needed /dev/null and /tmp first
	apk add --root $sysroot --initdb --quiet

	unpack_apkovl "$ovl" $sysroot
	eend $? $errstr || ovlfiles=
	# hack, incase /root/.ssh was included in apkovl
	[ -d "$sysroot/root" ] && chmod 700 "$sysroot/root"
	pkgs="$pkgs $(cat $sysroot/etc/apk/world 2>/dev/null)"
fi

好的，这边就提到Alpine很重要的开机世界观：〝Boot media〞，前面有说对於Alpine预设而言，它会从alpine_repo去捞package来组rootfs，而alpine_repo的来源有很多，网路是最简单的，但是如选项所示，他其实也可以是其他blockdev，所以他才会有个usb wait的选项让nlplug-findfs去等USB dev能够被正确识别(这边牵扯到kerenel device hotplug的uevent机制、有点复杂，下一篇谈nlpluf-findfs时会想办法一并带过）

然而一个重点是可以看到configure_ip这个，该shell function会展开成：

# Default (when configure_ip is called without setting ip=):
#   ip=dhcp
#
configure_ip() {
	[ -n "$MAC_ADDRESS" ] && return

	local IFS=':'
	set -- ${KOPT_ip:-dhcp}
	unset IFS

	local client_ip="$1"
	local gw_ip="$3"
	local netmask="$4"
	local device="$6"
	local autoconf="$7"
	local dns1="$8"
	local dns2="$9"

	case "$client_ip" in
		off|none) return;;
		dhcp) autoconf="dhcp";;
	esac

	[ -n "$device" ] || device=$(ip_choose_if)

	if [ -z "$device" ]; then
		echo "ERROR: IP requested but no network device was found"
		return 1
	fi

	if [ "$autoconf" = "dhcp" ]; then
		# automatic configuration
		if [ ! -e /usr/share/udhcpc/default.script ]; then
			echo "ERROR: DHCP requested but not present in initrd"
			return 1
		fi
		ebegin "Obtaining IP via DHCP ($device)"
		ifconfig "$device" 0.0.0.0
		udhcpc -i "$device" -f -q
		eend $?
	else
		# manual configuration
		[ -n "$client_ip" -a -n "$netmask" ] || return
		ebegin "Setting IP ($device)"
		if ifconfig "$device" "$client_ip" netmask "$netmask"; then
			[ -z "$gw_ip" ] || ip route add 0.0.0.0/0 via "$gw_ip" dev "$device"
		fi
		eend $?
	fi

	# Never executes if variables are empty
	for i in $dns1 $dns2; do
		echo "nameserver $i" >> /etc/resolv.conf
	done

	MAC_ADDRESS=$(cat /sys/class/net/$device/address)
}

从逻辑来看，如果以QEMU + SLIRP (usernet)的方式，最简单就是bootarg直接 ip=dhcp 让它自动抓。

if [ -f "$sysroot/etc/.default_boot_services" -o ! -f "$ovl" ]; then
	# add some boot services by default
	rc_add devfs sysinit
	rc_add dmesg sysinit
	rc_add mdev sysinit
	rc_add hwdrivers sysinit
	rc_add modloop sysinit

	rc_add modules boot
	rc_add sysctl boot
	rc_add hostname boot
	rc_add bootmisc boot
	rc_add syslog boot

	rc_add mount-ro shutdown
	rc_add killprocs shutdown
	rc_add savecache shutdown

	rc_add firstboot default

	rm -f "$sysroot/etc/.default_boot_services"
fi

这边也是一个重点，他在建OpenRC要的init files，rc_add会展开成：

#  add a boot service to $sysroot
rc_add() {
	mkdir -p $sysroot/etc/runlevels/$2
	ln -sf /etc/init.d/$1 $sysroot/etc/runlevels/$2/$1
}

而其中 /etc/init.d 对我来说是个有点阴阳魔界的状态，到现在我还在对initramfs log来查他的来源，到底在Alpine aports的哪个package会组件出来。

if [ "$KOPT_splash" != "no" ]; then
	echo "IMG_ALIGN=CM" > /tmp/fbsplash.cfg
	for fbdev in /dev/fb[0-9]; do
		[ -e "$fbdev" ] || break
		num="${fbdev#/dev/fb}"
		for img in /media/*/fbsplash$num.ppm; do
			[ -e "$img" ] || break
			config="${img%.*}.cfg"
			[ -e "$config" ] || config=/tmp/fbsplash.cfg
			fbsplash -s "$img" -d "$fbdev" -i "$config"
			break
		done
	done
	for fbsplash in /media/*/fbsplash.ppm; do
		[ -e "$fbsplash" ] && break
	done
fi

if [ -n "$fbsplash" ] && [ -e "$fbsplash" ]; then
	ebegin "Starting bootsplash"
	mkfifo $sysroot/$splashfile
	config="${fbsplash%.*}.cfg"
	[ -e "$config" ] || config=/tmp/fbsplash.cfg
	setsid fbsplash -T 16 -s "$fbsplash" -i $config -f $sysroot/$splashfile &
	eend 0
else
	KOPT_splash="no"
fi

if [ -f $sysroot/etc/fstab ]; then
	has_fstab=1
	fstab=$sysroot/etc/fstab

	# let user override tmpfs size in fstab in apkovl
	mountopts=$(awk '$2 == "/" && $3 == "tmpfs" { print $4 }' $sysroot/etc/fstab)
	if [ -n "$mountopts" ]; then
		mount -o remount,$mountopts $sysroot
	fi
	# move the ALPINE_MNT if ALPINE_DEV is specified in users fstab
	# this is so a generated /etc/apk/repositories will use correct
	# mount dir
	relocate_mount "$sysroot"/etc/fstab
elif [ -f /etc/fstab ]; then
	relocate_mount /etc/fstab
fi

# hack so we get openrc
pkgs="$pkgs alpine-base"

# copy keys so apk finds them. apk looks for stuff relative --root
mkdir -p $sysroot/etc/apk/keys/
cp -a /etc/apk/keys $sysroot/etc/apk

# generate apk repositories file. needs to be done after relocation
find_boot_repositories > $repofile

# silently fix apk arch in case the apkovl does not match
if [ -r "$sysroot"/etc/apk/arch ]; then
	apk_arch="$(apk --print-arch)"
	if [ -n "$apk_arch" ]; then
		echo "$apk_arch" > "$sysroot"/etc/apk/arch
	fi
fi

# generate repo opts for apk
for i in $(cat $repofile); do
	repo_opt="$repo_opt --repository $i"
done

# install new root
ebegin "Installing packages to root filesystem"

if [ "$KOPT_chart" = yes ]; then
	pkgs="$pkgs acct"
fi

# add openssh
if [ -n "$KOPT_ssh_key" ]; then
	pkgs="$pkgs openssh"
	rc_add sshd default
fi

# use swclock if no RTC is found
if rtc_exists || [ "$(uname -m)" = "s390x" ]; then
	rc_add hwclock boot
else
	rc_add swclock boot
fi

# enable support for modloop verification
if [ -f /var/cache/misc/*modloop*.SIGN.RSA.*.pub ]; then
	mkdir -p "$sysroot"/var/cache/misc
	cp /var/cache/misc/*modloop*.SIGN.RSA.*.pub "$sysroot"/var/cache/misc
	pkgs="$pkgs openssl"
fi

apkflags="--initramfs-diskless-boot --progress"
if [ -z "$MAC_ADDRESS" ]; then
	apkflags="$apkflags --no-network"
else
	apkflags="$apkflags --update-cache"
fi

if [ "$KOPT_quiet" = yes ]; then
	apkflags="$apkflags --quiet"
fi

if [ "$KOPT_keep_apk_new" != yes ]; then
	apkflags="$apkflags --clean-protected"
	[ -n "$ovlfiles" ] && apkflags="$apkflags --overlay-from-stdin"
fi
mkdir -p $sysroot/sys $sysroot/proc $sysroot/dev
mount -o bind /sys $sysroot/sys
mount -o bind /proc $sysroot/proc
mount -o bind /dev $sysroot/dev
if [ -n "$ovlfiles" ]; then
	apk add --root $sysroot $repo_opt $apkflags $pkgs <$ovlfiles
else
	apk add --root $sysroot $repo_opt $apkflags $pkgs
fi
umount $sysroot/sys $sysroot/proc $sysroot/dev
eend $?

# unmount ovl mount if needed
if [ -n "$ovl_unmount" ]; then
	umount $ovl_unmount 2>/dev/null
fi

# remount according default fstab from package
if [ -z "$has_fstab" ] && [ -f "$sysroot"/etc/fstab ]; then
	relocate_mount "$sysroot"/etc/fstab
fi

# generate repositories if none exists. this needs to be done after relocation
if ! [ -f "$sysroot"/etc/apk/repositories ]; then
	find_boot_repositories > "$sysroot"/etc/apk/repositories
fi

# respect mount options in fstab for ALPINE_MNT (e.g if user wants rw)
if [ -f "$sysroot"/etc/fstab ]; then
	opts=$(awk "\$2 == \"$ALPINE_MNT\" {print \$4}" $sysroot/etc/fstab)
	if [ -n "$opts" ]; then
		mount -o remount,$opts "$ALPINE_MNT"
	fi
fi

这边就是重头戏了，透过Alpine的套件管理系统来把rootfs要的东西，一个一个地装进 ALPINE_MNT中，如果使用者有下 rw，会帮忙remount成可读可写。

# fix inittab if alternative console
setup_inittab_console

# copy alpine release info
#if ! [ -f "$sysroot"/etc/alpine-release ] && [ -f $ALPINE_MNT/.alpine-release ]; then
#	cp $ALPINE_MNT/.alpine-release $sysroot/
#	ln -sf /.alpine-release $sysroot/etc/alpine-release
#fi

! [ -f "$sysroot"/etc/resolv.conf ] && [ -f /etc/resolv.conf ] && \
  cp /etc/resolv.conf "$sysroot"/etc

# setup bootchart for switch_root
chart_init=""
if [ "$KOPT_chart" = yes ]; then
	/sbin/bootchartd stop-initfs "$sysroot"
	chart_init="/sbin/bootchartd start-rootfs"
fi

if [ ! -x "${sysroot}${KOPT_init}" ]; then
	[ "$KOPT_splash" != "no" ] && echo exit > $sysroot/$splashfile
	echo "$KOPT_init not found in new root. Launching emergency recovery shell"
	echo "Type exit to continue boot."
	/bin/busybox sh
fi

# switch over to new root
cat /proc/mounts | while read DEV DIR TYPE OPTS ; do
	if [ "$DIR" != "/" -a "$DIR" != "$sysroot" -a -d "$DIR" ]; then
		mkdir -p $sysroot/$DIR
		mount -o move $DIR $sysroot/$DIR
	fi
done
sync

[ "$KOPT_splash" = "init" ] && echo exit > $sysroot/$splashfile
echo ""
exec /bin/busybox switch_root $sysroot $chart_init "$KOPT_init" $KOPT_init_args

[ "$KOPT_splash" != "no" ] && echo exit > $sysroot/$splashfile
echo "initramfs emergency recovery shell launched"
exec /bin/busybox sh
reboot

接下来就是最终的house-keeping了，inittab/rootfs有哪些东西要override掉这是最後一个可以干涉的环节，如果没有、那就会呼叫switch_root，去执行rootfs里面的OpenRC起来作真正的系统设定与daemon启动。