set -eE -o pipefail
trap 'echo "$0:$LINENO:error: \"$BASH_COMMAND\" returned $?" >&2' ERR
+# for calling outside of FAI, first
+# source /a/bin/fai-wrapper
+# then to set classes with: fai-setclass OPT1...
+# which sets CLASS_OPT1=true...
+#
+# OPTIONS:
+#
+# environment variables:
+#
+# HOSTNAME: if frodo, we exclude 2 devices from the /boot array, which
+# the bios does not see. if demohost, we set the luks password to just
+# 'x'.
+#
+# SPECIAL_DISK: For use outside of fai. A base disk name like
+# /dev/sdk. If set, we just cryptsetup and partition this one disk then
+# exit. This is useful for partitioning a disk in preparation to replace
+# a failed or failing disk from a raid10 array.
+#
+# classes:
+#
+# REPARTITION: forces repartitioning even if we detect the proper amount
+# of partitions already exist.
+#
+# ROTATIONAL: forces to install onto hdds instead of sdds. normally sdds
+# are chosen if they exist.
+#
+# PARTITION_PROMPT: command line prompt before partitioning
+#
+# RAID0: forces raid0 filesystem. Normally with 4+ devices, we use
+# raid10.
+
+if [[ $SPECIAL_DISK ]]; then
+ export CLASS_REPARTITION=true
+fi
+
# # fai's setup-storage won't do btrfs on luks,
# # so we do it ourself :)
+# inspiration taken from files in fai-setup-storage package
+
+# if we are not running in fai, skiptask won't be defined, so carry on.
+skiptask partition || ! type skiptask
+
-skiptask partition || ! type skiptask # for running not in fai
#### begin configuration
rootn=1
swapn=2
bootn=3
-bios_grubn=4
-boot_mib=1500
+# ext partition so grub can write persistent variables,
+# so it can do a one time boot. grub can't write to
+# btrfs or any cow fs because it's more
+# more complicated to do and they don't want to.
+grub_extn=4
+# bios boot partition,
+# https://wiki.archlinux.org/index.php/GRUB
+bios_grubn=5
+lastn=$bios_grubn
+# this is larger than needed for several /boot subvols,
+# becuase I keep a minimal debian install on it, for
+# recovery needs, and for doing pxe-kexec.
+boot_mib=10000
##### end configuration
-add-part() {
+add-part() { # add partition suffix to $dev
local d ret
if [[ $# == 1 ]]; then
d=$dev
bootdev() { add-part $@ $bootn; }
rootdev() { add-part $@ $rootn; }
swapdev() { add-part $@ $swapn; }
+grub_extdev() { add-part $@ $grub_extn; }
bios_grubdev() { add-part $@ $bios_grubn; }
crypt-dev() { echo /dev/mapper/crypt_dev_${1##*/}; }
swap-cryptdev() { crypt-dev $(swapdev $@); }
root-cryptname() { crypt-name $(rootdev $@); }
swap-cryptname() { crypt-name $(swapdev $@); }
+devbyid() {
+ local f
+ for f in $FAI/distro-install-common/devbyid \
+ /a/bin/fai/fai/config/distro-install-common/devbyid; do
+ if [[ -e $f ]]; then $f "$@"; break; fi
+ done
+}
##### end function defs
partition=false # change to true to force a full wipe
fi
-lastn=$bios_grubn
hdds=()
done
# install all ssds, or if there are none, all hdds
-if (( ${#ssds[@]} > 0 )); then
+if ! ifclass ROTATIONAL && (( ${#ssds[@]} > 0 )); then
short_devs=( ${ssds[@]} )
else
short_devs=( ${hdds[@]} )
# check if the partitions exist have the right filesystems
#blkid="$(blkid -s TYPE)"
for dev in ${short_devs[@]}; do
- ! $partition || break
+ if $partition; then break; fi
y=$(readlink -f $dev)
- x=($y[0-9])
- [[ ${#x[@]} == "${lastn}" ]] || partition=true
+ arr=($y[0-9])
+ [[ ${#arr[@]} == "${lastn}" ]] || partition=true
for (( i=1; i <= lastn; i++ )); do
[[ -e ${dev}$i ]] || partition=true
done
if $partition && ifclass PARTITION_PROMPT; then
echo "Press any key except ctrl-c to continue and partition these drives:"
- echo " ${short_devs[@]}"
- read
+ echo " ${short_devs[*]}"
+ read -r
fi
devs=()
done
-
+first=false
boot_devs=()
for dev in ${devs[@]}; do
if ifclass frodo; then
else
boot_devs+=(`bootdev`)
fi
+ if [[ $boot_devs && $first ]]; then
+ first_grub_extdev=`grub_extdev`
+ first=false
+ fi
done
+if ifclass RAID0 || (( ${#boot_devs[@]} < 4 )); then
+ raid_level=0
+else
+ raid_level=10
+ # need double the space if we are raid 10, and then
+ # might as well give some extra.
+ boot_mib=$((boot_mib * 3))
+fi
+
+
+
if [[ ! $DISTRO ]]; then
- if ifclass STRETCH64; then
+ if ifclass VOL_STRETCH_BOOTSTRAP; then
+ DISTRO=debianstretch_bootstrap
+ elif ifclass VOL_STRETCH; then
+ DISTRO=debianstretch
+ elif ifclass VOL_TESTING; then
DISTRO=debiantesting
- elif ifclass STABLE; then
- DISTRO=debianstable
- elif ifclass XENIAL64; then
+ elif ifclass VOL_XENIAL; then
DISTRO=ubuntuxenial
+ elif ifclass VOL_BELENOS; then
+ DISTRO=trisquelbelenos
+ elif ifclass VOL_FLIDAS; then
+ DISTRO=trisquelflidas
else
echo "PARTITIONER ERROR: no distro class/var set" >&2
exit 1
fi
fi
-
-
-case ${#boot_devs[@]} in
- # need double the space if we are raid 10, and then
- # might as well give some extra overhead.
- [4-9]*|[1-3]?*) boot_mib=$((boot_mib * 3)) ;;
-esac
+first_boot_dev=${boot_devs[0]}
bpart() { # btrfs a partition
- case ${#@} in
- [1-3]) mkfs.btrfs -f $@ ;;
- [4-9]*|[1-3]?*) mkfs.btrfs -f -m raid10 -d raid10 $@ ;;
+ case $raid_level in
+ 0) mkfs.btrfs -f $@ ;;
+ 10) mkfs.btrfs -f -m raid10 -d raid10 $@ ;;
esac
}
-first_boot_dev=${boot_devs[0]}
# keyfiles generated like:
# head -c 2048 /dev/urandom | od | s dd of=/q/root/luks/host-demohost
exit 1
fi
-if ifclass tp; then
- lukspw=$(cat $luks_dir/traci)
-else
- lukspw=$(cat $luks_dir/ian)
-fi
+lukspw=$(cat $luks_dir/iank)
+# # ian: disabled while I use the tp host.
+# if ifclass tp; then
+# lukspw=$(cat $luks_dir/traci)
+# fi
if ifclass demohost; then
lukspw=x
fi
first_root_crypt=$(root-cryptdev ${devs[0]})
-bios_grubn=4
# 1.5 x based on https://access.redhat.com/documentation/en-US/Red_Hat_Enterprise_Linux/7/html/Installation_Guide/sect-disk-partitioning-setup-x86.html#sect-custom-partitioning-x86
swap_mib=$(( $(grep ^MemTotal: /proc/meminfo | \
awk '{print $2}') * 3/(${#devs[@]} * 2 ) / 1024 ))
done
shopt -s nullglob
if $partition; then
- for dev in ${devs[@]}; do
- for x in $dev[0-9]; do
- count_down=10
- # wipefs has failed, manual run works, google suggests timing issue
- while ! wipefs -a $x; do
- sleep 2
- count_down=$((count_down - 1))
- (( count_down > 0 )) || exit 1
+
+ ### begin wipefs
+ if [[ ! $SPECIAL_DISK ]]; then
+ for dev in ${devs[@]}; do
+ # if we repartition to the same as an old partition,
+ # we don't want any old fses hanging around.
+ for (( i=1; i <= lastn; i++ )); do
+ x=$(add-part $dev $i)
+ [[ -e $x ]] || continue
+ count_down=10
+ # wipefs has failed, manual run works, google suggests timing issue
+ while ! wipefs -a $x; do
+ sleep 2
+ count_down=$((count_down - 1))
+ (( count_down > 0 )) || exit 1
+ done
done
done
- done
+ fi
+ ### end wipefs
+
for dev in ${devs[@]}; do
+ if [[ $SPECIAL_DISK ]]; then
+ dev=$(devbyid $SPECIAL_DISK)
+ fi
+
# parted will round up the disk size. Do -1 so we can have
# fully 1MiB unit partitions for easy resizing of the last partition.
# Otherwise we would pass in -0 for the end argument for the last partition.
#
- # parted print error output is expected. example:
+ # Note: parted print error output is expected. example:
# Error: /dev/vda: unrecognised disk label
disk_mib=$(( $(parted -m $dev unit MiB print | \
sed -nr "s#^/dev/[^:]+:([0-9]+).*#\1#p") - 1))
swap_end=$(( root_end + swap_mib))
parted -s $dev mklabel gpt
- # gpt ubuntu cloud image uses ~4. fai uses 1 MiB.
- # I read something in the parted manual saying cheap flash media
- # likes to start at 4.
# MiB because parted complains about alignment otherwise.
pcmd="parted -a optimal -s -- $dev"
- $pcmd mkpart primary "ext3" 4MiB ${root_end}MiB
- $pcmd mkpart primary "linux-swap" ${root_end}MiB ${swap_end}MiB
+ $pcmd mkpart primary ext3 12MiB ${root_end}MiB
+ # without naming, systemd gives us misc errors like:
+ # dev-disk-by\x2dpartlabel-primary.device: Dev dev-disk-by\x2dpartlabel-primary.device appeared twice
+ $pcmd name $rootn root
+ # normally a swap is type "linux-swap", but this is encrypted swap. using that
+ # label will confuse systemd.
+ $pcmd mkpart primary "" ${root_end}MiB ${swap_end}MiB
+ $pcmd name $swapn swap
$pcmd mkpart primary "" ${swap_end}MiB ${disk_mib}MiB
+ $pcmd name $bootn boot
+ # i only need a few k, but googling min size,
+ # I found someone saying that gparted required
+ # required at least 8 because of their hard drive cylinder size.
+ # And 8 is still very tiny.
+ $pcmd mkpart primary "ext2" 4MiB 12MiB
+ $pcmd name $grub_extn grubext
+ # gpt ubuntu cloud image uses ~4 mb for this partition. fai uses 1 MiB.
+ # so, I use 3, whatever.
+ # note: parted manual saying cheap flash media
+ # should to start at 4.
$pcmd mkpart primary "" 1MiB 4MiB
+ $pcmd name $bios_grubn biosgrub
$pcmd set $bios_grubn bios_grub on
$pcmd set $bootn boot on # generally not needed on modern systems
# the mkfs failed before on a vm, which prompted me to add
# sleep .1
- # then failed on a physical machine
+ # then it failed again on a physical machine
# with:
# Device /dev/disk/by-id/foo doesn't exist or access denied,
- # so here we wait.
+ # so I added a wait until it existed.
+ # Then I added the mkfs.ext2, which claimed to succeed,
+ # but then couldn't be found upon reboot. In that case we didn't
+ # wait at all. So I've added a 3 second minimum wait.
+ sleep 3
secs=0
while [[ ! -e `rootdev` ]] && (( secs < 10 )); do
sleep 1
secs=$((secs +1))
done
+ # Holds just a single file, rarely written, so
+ # use ext2, like was often used for the /boot partition.
+ # This exists because grub can only persist data to a non-cow fs.
+ # And we use persisting a var in grub to do a one time boot.
+ # We could pass the data on the kernel command line and persist it
+ # to grubenv after booting, but that relies on the boot always succeeding.
+ # This is just a bit more robust, and it could work for booting
+ # into ipxe which can't persist data, if we ever got that working.
+ mkfs.ext2 `grub_extdev`
yes YES | cryptsetup luksFormat `rootdev` $luks_dir/host-$HOSTNAME \
-c aes-cbc-essiv:sha256 -s 256 || [[ $? == 141 ]]
yes "$lukspw" | \
cryptsetup luksOpen `rootdev` `root-cryptname` \
--key-file $luks_dir/host-$HOSTNAME
+
+ if [[ $SPECIAL_DISK ]]; then
+ exit 0
+ fi
done
- ls -la /dev/btrfs-control
+ ls -la /dev/btrfs-control # this was probably for debugging...
sleep 1
bpart $(for dev in ${devs[@]}; do root-cryptdev; done)
bpart ${boot_devs[@]}
else
for dev in ${devs[@]}; do
+ mkfs.ext2 `grub_extdev`
cryptsetup luksOpen `rootdev` `root-cryptname` \
--key-file $luks_dir/host-$HOSTNAME
done
sleep 1
fi
-mount -o subvolid=0 $first_root_crypt /mnt
-# systemd creates subvolumes we want to delete.
-s=($(btrfs subvolume list --sort=-path /mnt |
- sed -rn "s#^.*path\s*(root_$DISTRO/\S+)\s*\$#\1#p"))
-for subvol in ${s[@]}; do btrfs subvolume delete /mnt/$subvol; done
-btrfs subvolume set-default 0 /mnt
-[[ ! -e /mnt/root_$DISTRO ]] || btrfs subvolume delete /mnt/root_$DISTRO
-## create subvols ##
-cd /mnt
-btrfs subvolume create root_$DISTRO
-[[ -e q ]] || btrfs subvolume create q
-chown root:1000 q
-
-mkdir -p /mnt/root_$DISTRO/boot
-for x in q/a q/i; do
- mkdir -p $x
- chown 1000:1000 $x
- chmod 755 $x
-done
-# could set default like this, but no reason to.
-# btrfs subvolume set-default \
- # $(btrfs subvolume list . | grep "root_$DISTRO$" | awk '{print $2}') .
-chattr -Rf +C root_$DISTRO
-cd /
-umount /mnt
+if [[ $DISTRO != debianstretch_bootstrap ]]; then
+ # bootstrap distro doesn't use separate encrypted root.
+ mount -o subvolid=0 $first_root_crypt /mnt
+ # systemd creates subvolumes we want to delete.
+ s=($(btrfs subvolume list --sort=-path /mnt |
+ sed -rn "s#^.*path\s*(root_$DISTRO/\S+)\s*\$#\1#p"))
+ for subvol in ${s[@]}; do btrfs subvolume delete /mnt/$subvol; done
+ btrfs subvolume set-default 0 /mnt
+ [[ ! -e /mnt/root_$DISTRO ]] || btrfs subvolume delete /mnt/root_$DISTRO
+
+ ## create subvols ##
+ cd /mnt
+
+ btrfs subvolume create root_$DISTRO
+
+ mkdir -p /mnt/root_$DISTRO/boot
+ # could set default subvol like this, but no reason to.
+ # btrfs subvolume set-default \
+ # $(btrfs subvolume list . | grep "root_$DISTRO$" | awk '{print $2}') .
+
+ # no cow on the root filesystem. it's setup is fully scripted,
+ # if it's messed up, we will just recreated it,
+ # and we can get better perf with this.
+ # I can't remember exactly why, but this is preferable to mounting with
+ # -o nodatacow, I think because subvolumes inherit that.
+ chattr -Rf +C root_$DISTRO
+ cd /
+ umount /mnt
+fi
+
mount -o subvolid=0 $first_boot_dev /mnt
cd /mnt
-btrfs subvolume set-default 0 /mnt
-[[ ! -e /mnt/boot_$DISTRO ]] || btrfs subvolume delete /mnt/boot_$DISTRO
-btrfs subvolume create boot_$DISTRO
+btrfs subvolume set-default 0 /mnt # already default, just ensuring it.
+
+# for libreboot systems. grub2 only reads from subvolid=0
+mkdir -p /mnt/grub2
+cp $FAI/distro-install-common/libreboot_grub.cfg /mnt/grub2
+
+if [[ $DISTRO == debianstretch_bootstrap ]]; then
+ # this is just convenience for the libreboot_grub config
+ # so we can glob the other ones easier.
+ boot_vol=$DISTRO
+else
+ boot_vol=boot_$DISTRO
+fi
+[[ ! -e /mnt/$boot_vol ]] || btrfs subvolume delete /mnt/$boot_vol
+btrfs subvolume create $boot_vol
cd /
umount /mnt
## end create subvols ##
+dev=${boot_devs[0]}
+mount $first_grub_extdev /mnt
+grub-editenv /mnt/grubenv set did_fai_check=true
+grub-editenv /mnt/grubenv set last_boot=/$boot_vol
+umount /mnt
-# note, the mount point /a seems to get automatically created somewhere
-cat > /tmp/fai/fstab <<EOF
-$first_root_crypt / btrfs noatime,subvol=root_$DISTRO 0 0
-$first_root_crypt /q btrfs noatime,subvol=q 0 0
-/q/a /a none bind 0 0
-$first_boot_dev /boot btrfs noatime,subvol=boot_$DISTRO 0 0
+if [[ $DISTRO == debianstretch_bootstrap ]]; then
+ cat > /tmp/fai/fstab <<EOF
+$first_boot_dev / btrfs noatime,subvol=$boot_vol 0 0
EOF
-
-
-# I will avoid using uuid in the future.
-if ifclass treetowl; then
- cat >> /tmp/fai/fstab <<'EOF'
-$first_root_crypt /i btrfs noatime,subvol=i 0 0
-UUID=3f7b31cd-f299-40b4-a86b-7604282e2715 /i btrfs noatime 0 2
+ cat >/tmp/fai/disk_var.sh <<EOF
+BOOT_DEVICE="${short_devs[@]}"
+ROOT_PARTITION=$first_boot_dev
EOF
-fi
-
-swaps=()
-for dev in ${devs[@]}; do
- swaps+=(`swap-cryptname`)
- cat >>/tmp/fai/crypttab <<EOF
+else
+ # note, fai creates the mountpoints listed here
+ cat > /tmp/fai/fstab <<EOF
+$first_root_crypt / btrfs noatime,subvol=root_$DISTRO 0 0
+$first_root_crypt /mnt/root btrfs noatime,subvolid=0 0 0
+$first_boot_dev /boot btrfs noatime,subvol=$boot_vol 0 0
+EOF
+ swaps=()
+ for dev in ${devs[@]}; do
+ swaps+=(`swap-cryptname`)
+ cat >>/tmp/fai/crypttab <<EOF
`root-cryptname` `rootdev` none keyscript=/root/keyscript,discard,luks
`swap-cryptname` `swapdev` /dev/urandom swap,cipher=aes-xts-plain64,size=256,hash=ripemd160
EOF
- cat >> /tmp/fai/fstab <<EOF
+ cat >> /tmp/fai/fstab <<EOF
`swap-cryptdev` none swap sw 0 0
EOF
-done
-
-# fai would do this:
-#BOOT_DEVICE=\${BOOT_DEVICE:-"${devs[0]}"}
+ done
-# swaplist seems to do nothing.
+ # fai would do this:
+ #BOOT_DEVICE=\${BOOT_DEVICE:-"${devs[0]}"}
-cat >/tmp/fai/disk_var.sh <<EOF
+ # note: swaplist seems to do nothing.
+ cat >/tmp/fai/disk_var.sh <<EOF
+BOOT_DEVICE="${short_devs[@]}"
+BOOT_PARTITION=\${BOOT_PARTITION:-$first_boot_dev}
# ROOT_PARTITIONS is added by me, used in arch setup.
ROOT_PARTITIONS="${root_devs[@]}"
ROOT_PARTITION=\${ROOT_PARTITION:-$first_root_crypt}
-BOOT_PARTITION=\${BOOT_PARTITION:-$first_boot_dev}
-BOOT_DEVICE="${short_devs[@]}"
SWAPLIST=\${SWAPLIST:-"${swaps[@]}"}
EOF
+fi