# along with this program; if not, write to the Free Software
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
+PS4='+ $LINENO '
set -eE -o pipefail
trap 'echo "$0:$LINENO:error: \"$BASH_COMMAND\" returned $?" >&2' ERR
+if [[ $EUID != 0 ]]; then
+ echo "$0: error: need to run as root" >&2
+ exit 1
+fi
+
+# for calling outside of FAI:
+# fai-redep
+#
+# source /b/fai/fai-wrapper
+# - set any appropriate classes with: fai-setclass OPT1... which sets CLASS_OPT1=true...
+# or run eval-fai-classfile FILE.
+# - Set a VOL_DISTROVER (if not doing mkroot2) eg:
+# fai-setclass VOL_NABIA
+#
+# OPTIONS:
+#
+# mkroot2: for running outside of fai and setting up the root2/boot2 luks and btrfs
+#
+# 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.
+#
+# NOWIPE: use existing subvolumes if they 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.
+# RAID1: forces raid1 filesystem.
+
+mkroot2=false
+if [[ $1 ]]; then
+ case $1 in
+ mkroot2)
+ mkroot2=true
+ ;;
+ *)
+ echo "$0: error: unsupported arg: $1" >&2
+ exit 1
+ ;;
+ esac
+fi
+
+
+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
+
+if ! type -p devbyid; then
+ for d in $FAI/distro-install-common \
+ /a/bin/fai/fai/config/distro-install-common $FAI $PWD; do
+ [[ -d $d ]] || continue
+ if [[ -e $d/devbyid ]]; then
+ devbyid=$d/devbyid
+ devbyid() { $devbyid "$@"; }
+ break
+ fi
+ done
+ if [[ ! $devbyid ]]; then
+ echo "$0: error: failed to find devbyid script" >&2
+ exit 1
+ fi
+fi
+
-skiptask partition || ! type skiptask # for running not in fai
#### begin configuration
+# this is the ordering of the /dev/sdaX, but
+# the ordering of the partition layout goes like this:
+# bios_grub
+# grub_ext
+# efi
+# root
+# swap
+# boot
+
rootn=1
-swapn=2
-bootn=3
+root2n=2
+swapn=3
+bootn=4
+boot2n=5
+efin=6
# ext partition so grub can write persistent variables,
# so it can do a one time boot. grub can't write to
# btrfs or any cow fs because it's more
# more complicated to do and they don't want to.
-grub_extn=4
+grub_extn=7
# bios boot partition,
# https://wiki.archlinux.org/index.php/GRUB
-bios_grubn=5
+bios_grubn=8
+even_bign=9
lastn=$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 partition suffix to $dev
- local d ret
- if [[ $# == 1 ]]; then
- d=$dev
- part=$1
- else
- d=$1
- part=$2
- fi
- if [[ $d == /dev/disk/by-id/* ]]; then
- ret=$d-part$part
- else
- ret=$d$part
- fi
- echo $ret
+ local d part
+ if [[ $# == 1 ]]; then
+ d=$dev
+ part=$1
+ else
+ d=$1
+ part=$2
+ fi
+ echo $d-part$part
}
-bootdev() { add-part $@ $bootn; }
rootdev() { add-part $@ $rootn; }
swapdev() { add-part $@ $swapn; }
+bootdev() { add-part $@ $bootn; }
+boot2dev() { add-part $@ $boot2n; }
+efidev() { add-part $@ $efin; }
grub_extdev() { add-part $@ $grub_extn; }
bios_grubdev() { add-part $@ $bios_grubn; }
+even_bigdev() { add-part $@ $even_bign; }
crypt-dev() { echo /dev/mapper/crypt_dev_${1##*/}; }
crypt-name() { echo 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 $f "$@"; fi
- done
+dev-mib() {
+ local d=${1:-$dev}
+ echo $(( $(parted -m $d unit MiB print | \
+ sed -nr "s#^/dev/[^:]+:([0-9]+).*#\1#p") - 1))
+}
+
+luks-setup() {
+ local luksdev="$1"
+ # when we move to newer than trisquel 9, we can remove
+ # --type luks1. We can also check on cryptsetup --help | less /compil
+ # to see about the other settings. Default in debian 9 is luks2.
+ # You can convert from luks2 to luks 1 by adding a temporary key:
+ # cryptsetup luksAddKey --pbkdf pbkdf2
+ # then remove the new format keys with cryptsetup luksRemoveKey
+ # then cryptsetup convert DEV --type luks1, then readd old keys and remove temp.
+ yes YES | cryptsetup luksFormat $luksdev $luks_file || [[ $? == 141 ]]
+ yes "$lukspw" | \
+ cryptsetup luksAddKey --key-file $luks_file \
+ $luksdev || [[ $? == 141 ]]
+ # background: Keyfile and password are treated just
+ # like 2 ways to input a passphrase, so we don't actually need to have
+ # different contents of keyfile and passphrase, but it makes some
+ # security sense to a really big randomly generated passphrase
+ # as much as possible, so we have both.
+ #
+ # This would remove the keyfile.
+ # yes 'test' | cryptsetup luksRemoveKey /dev/... \
+ # /key/file || [[ $? == 141 ]]
+ cryptsetup luksOpen $luksdev $(crypt-name $luksdev) --key-file $luks_file
}
##### end function defs
-if ifclass REPARTITION;then
- partition=true # force a full wipe
+if ifclass REPARTITION; then
+ partition=true # force a full wipe
else
- partition=false # change to true to force a full wipe
+ partition=false # change to true to force a full wipe
+fi
+if ifclass NOWIPE; then
+ wipe=false
+else
+ wipe=true
fi
+if (($(nproc) > 2)); then
+ mopts=,compress=zstd
+fi
+declare -A disk_excludes
+if ! $mkroot2 && ! ifclass USE_MOUNTED; then
+ ## ignore disks that are mounted, eg when running from fai-cd
+ while read -r l; do
+ eval "$l"
+ if [[ ! $PKNAME ]]; then
+ PKNAME="$KNAME"
+ fi
+ if [[ $MOUNTPOINT ]]; then
+ disk_excludes[$PKNAME]=true
+ fi
+ done < <(lsblk -nP -o KNAME,MOUNTPOINT,PKNAME)
+fi
hdds=()
ssds=()
-cd /sys/block
-for disk in [sv]d[a-z]; do
- case $(cat $disk/queue/rotational) in
- 0) ssds+=(/dev/$disk) ;;
- 1) hdds+=(/dev/$disk) ;;
- *) echo "$0: error: unknown /sys/block/$disk/queue/rotational: \
+# this excludes usb. note: i may encounter some other type in the future.
+for disk in $(lsblk -do name,tran -n | awk '$2 ~ "^(sata|nvme)$" { print $1 }'); do
+ if [[ ${disk_excludes[$disk]} ]]; then
+ continue
+ fi
+ case $(cat /sys/block/$disk/queue/rotational) in
+ 0) ssds+=(/dev/$disk) ;;
+ 1) hdds+=(/dev/$disk) ;;
+ *) echo "$0: error: unknown /sys/block/$disk/queue/rotational: \
$(cat $disk/queue/rotational)"; exit 1 ;;
- esac
+ esac
done
-# install all ssds, or if there are none, all hdds
+# install all ssds, or if there are none, all hdds.
+# Note, usb flash disks are seen as rotational, which is
+# very odd, but convenient for ignoring them here.
+# TODO: find a reliable way to ignore them.
if ! ifclass ROTATIONAL && (( ${#ssds[@]} > 0 )); then
- short_devs=( ${ssds[@]} )
+ short_devs=( ${ssds[@]} )
else
- short_devs=( ${hdds[@]} )
+ short_devs=( ${hdds[@]} )
fi
# check if the partitions exist have the right filesystems
#blkid="$(blkid -s TYPE)"
for dev in ${short_devs[@]}; do
- if $partition; then break; fi
- y=$(readlink -f $dev)
- arr=($y[0-9])
- [[ ${#arr[@]} == "${lastn}" ]] || partition=true
- for (( i=1; i <= lastn; i++ )); do
- [[ -e ${dev}$i ]] || partition=true
- done
- # On one system, blkid is missing some partitions.
- # maybe we need a flag, like FUZZY_BLKID or something, so we
- # can check that at least some exist.
- # for x in "`rootdev`: TYPE=\"crypto_LUKS\"" "`bootdev`: TYPE=\"btrfs\""; do
- # echo "$blkid" | grep -Fx "$x" &>/dev/null || partition=true
- # done
+ if $partition; then break; fi
+ y=$(readlink -f $dev)
+ arr=($y?*)
+ if (( ${#arr[@]} < lastn )); then
+ partition=true
+ fi
+ # On one system, blkid is missing some partitions.
+ # maybe we need a flag, like FUZZY_BLKID or something, so we
+ # can check that at least some exist.
+ # for x in "`rootdev`: TYPE=\"crypto_LUKS\"" "`bootdev`: TYPE=\"btrfs\""; do
+ # echo "$blkid" | grep -Fx "$x" &>/dev/null || partition=true
+ # done
done
if $partition && ifclass PARTITION_PROMPT; then
- echo "Press any key except ctrl-c to continue and partition these drives:"
- echo " ${short_devs[*]}"
- read -r
+ echo "Press any key except ctrl-c to continue and partition these drives:"
+ echo " ${short_devs[*]}"
+ read -r
fi
devs=()
shopt -s extglob
for short_dev in ${short_devs[@]}; do
- devs+=($(devbyid $short_dev))
+ devs+=($(devbyid $short_dev))
done
+if [[ ! ${devs[0]} ]]; then
+ echo "$0: error: failed to detect devs" >&2
+ exit 1
+fi
-
-first=false
+boot_space=0
+first=true
boot_devs=()
+boot2_devs=()
for dev in ${devs[@]}; do
- if ifclass frodo; then
- # I ran into a machine where the bios doesn't know about some disks,
- # so 1st stage of grub also doesn't know about them.
- # Also, grub does not support mounting degraded btrfs as far as
- # I can tell with some googling.
- # From within an arch install env, I could detect them by noting
- # their partitions were mixed with the next disk in /dev/disk/by-path,
- # and I have mixed model disks, and I could see the 8 models which showed
- # up in the bios, and thus see which 2 models were missing.
- # hdparm -I /dev/sdh will give model info in linux.
- # However, in fai on jessie, /dev/disk/by-path dir doesn't exist,
- # and I don't see another way, so I'm hardcoding them.
- # We still put grub on them and partition them the same, for uniformity
- # and in case they get moved to a system that can recognize them,
- # we just exclude them from the boot filesystem.
- cd /dev/disk/by-id/
- bad_disk=false
- for id in ata-TOSHIBA_MD04ACA500_8539K4TQFS9A \
- ata-TOSHIBA_MD04ACA500_Y5IFK6IJFS9A; do
- if [[ $(readlink -f $id) == "$(readlink -f $dev)" ]]; then
- bad_disk=true
- break
- fi
- done
- $bad_disk || boot_devs+=(`bootdev`)
- else
- boot_devs+=(`bootdev`)
- fi
- if [[ $boot_devs && $first ]]; then
- first_grub_extdev=`grub_extdev`
- first=false
+ if ifclass frodo; then
+ # I ran into a machine where the bios doesn't know about some disks,
+ # so 1st stage of grub also doesn't know about them.
+ # Also, grub does not support mounting degraded btrfs as far as
+ # I can tell with some googling.
+ # From within an arch install env, I could detect them by noting
+ # their partitions were mixed with the next disk in /dev/disk/by-path,
+ # and I have mixed model disks, and I could see the 8 models which showed
+ # up in the bios, and thus see which 2 models were missing.
+ # hdparm -I /dev/sdh will give model info in linux.
+ # However, in fai on jessie, /dev/disk/by-path dir doesn't exist,
+ # and I don't see another way, so I'm hardcoding them.
+ # We still put grub on them and partition them the same, for uniformity
+ # and in case they get moved to a system that can recognize them,
+ # we just exclude them from the boot filesystem.
+ cd /dev/disk/by-id/
+ bad_disk=false
+ for id in ata-TOSHIBA_MD04ACA500_8539K4TQFS9A \
+ ata-TOSHIBA_MD04ACA500_Y5IFK6IJFS9A; do
+ if [[ $(readlink -f $id) == "$(readlink -f $dev)" ]]; then
+ bad_disk=true
+ break
+ fi
+ done
+ if ! $bad_disk; then
+ boot_devs+=($(bootdev))
+ boot2_devs+=($(boot2dev))
fi
+ else
+ boot_space=$(( boot_space + $(parted -m $dev unit MiB print | \
+ sed -nr "s#^/dev/[^:]+:([0-9]+).*#\1#p") - 1))
+ boot_devs+=($(bootdev))
+ boot2_devs+=($(boot2dev))
+ fi
+ if $first && [[ $boot_devs ]]; then
+ first_efi=$(efidev)
+ first_grub_extdev=$(grub_extdev)
+ first=false
+ fi
done
+first_boot_dev=${boot_devs[0]}
+
+even_raid=false
+if ifclass RAID0 || (( ${#boot_devs[@]} == 1 )); then
+ raid_level=0
+elif ifclass RAID1 || (( ${#boot_devs[@]} <= 3 )); then
+ if (( ${#boot_devs[@]} == 2 )); then
+ even_raid=true
+ fi
+ raid_level=1
+else
+ raid_level=10
+fi
-if ifclass RAID0 || (( ${#boot_devs[@]} < 4 )); then
- raid_level=0
+### Begin calculate boot partition space
+# due to raid duplication
+case $raid_level in
+ 1*) boot_space=$(( boot_space / 2 )) ;;
+esac
+build_mib=0
+browse_mib=0
+whonix_mib=0
+if (( boot_space > 60000 )); then
+ # 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
+ root2_mib=200000
+ boot2_mib=2000
+ build_mib=30000
+ browse_mib=10000
+ whonix_mib=15000
+elif (( boot_space > 30000 )); then
+ boot_mib=$(( 5000 + (boot_space - 30000) / 2 ))
+ root2_mib=100
+ boot2_mib=100
else
- raid_level=10
- # need double the space if we are raid 10, and then
- # might as well give some extra.
- boot_mib=$((boot_mib * 3))
+ # Small vms don't have room for /boot recovery. With 3 kernels
+ # installed, i'm using 132M on t8, so this seems like plenty of
+ # room. note: rhel 8 recomments 1g for /boot.
+ boot_mib=500
+ root2_mib=100
+ boot2_mib=100
fi
+case $raid_level in
+ 1*)
+ boot_mib=$(( boot_mib * 2 ))
+ boot2_mib=$(( boot2_mib * 2 ))
+ root2_mib=$(( root2_mib * 2 ))
+ build_mib=$(( build_mib * 2 ))
+ browse_mib=$(( browse_mib * 2 ))
+ whonix_mib=$(( whonix_mib * 2 ))
+ ;;
+esac
+### end calculate boot partition space
if [[ ! $DISTRO ]]; then
- if ifclass VOL_STABLE_BOOTSTRAP; then
- DISTRO=debianstable_bootstrap
- elif ifclass VOL_STRETCH; then
- DISTRO=debiantesting
- elif ifclass VOL_STABLE; then
- DISTRO=debianstable
- 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
+ if ifclass VOL_BULLSEYE_BOOTSTRAP; then
+ DISTRO=debianbullseye_bootstrap
+ elif ifclass VOL_STRETCH; then
+ DISTRO=debianstretch
+ elif ifclass VOL_BUSTER; then
+ DISTRO=debianbuster
+ elif ifclass VOL_BULLSEYE; then
+ DISTRO=debianbullseye
+ elif ifclass VOL_TESTING; then
+ DISTRO=debiantesting
+ elif ifclass VOL_XENIAL; then
+ DISTRO=ubuntuxenial
+ elif ifclass VOL_BIONIC; then
+ DISTRO=ubuntubionic
+ elif ifclass VOL_FOCAL; then
+ DISTRO=ubuntufocal
+ elif ifclass VOL_FLIDAS; then
+ DISTRO=trisquelflidas
+ elif ifclass VOL_ETIONA; then
+ DISTRO=trisqueletiona
+ elif ifclass VOL_NABIA; then
+ DISTRO=trisquelnabia
+ elif $mkroot2; then
+ :
+ else
+ echo "PARTITIONER ERROR: no distro class/var set" >&2
+ exit 1
+ fi
fi
-first_boot_dev=${boot_devs[0]}
bpart() { # btrfs a partition
- case $raid_level in
- 0) mkfs.btrfs -f $@ ;;
- 10) mkfs.btrfs -f -m raid10 -d raid10 $@ ;;
- esac
+ case $raid_level in
+ 0) mkfs.btrfs -f $@ ;;
+ 1) mkfs.btrfs -f -m raid1 -d raid1 $@ ;;
+ 10) mkfs.btrfs -f -m raid10 -d raid10 $@ ;;
+ esac
}
-# keyfiles generated like:
-# head -c 2048 /dev/urandom | od | s dd of=/q/root/luks/host-demohost
-luks_dir=${LUKS_DIR:-/var/lib/fai/config/distro-install-common/luks}
-
-if [[ ! -e $luks_dir/host-$HOSTNAME ]]; then
- echo "$0: error: no key for hostname at $luks_dir/host-$HOSTNAME" >&2
+if [[ ! $luks_dir ]]; then
+ # see README for docs about how to create these
+ luks_dir=$FAI/distro-install-common/luks
+ if [[ ! -d $luks_dir ]]; then
+ luks_dir=/q/root/luks
+ fi
+ if [[ ! -d $luks_dir ]]; then
+ echo "$0: error: no luks_dir found" >&2
exit 1
+ fi
fi
-if ifclass tp; then
- lukspw=$(cat $luks_dir/traci)
-else
- lukspw=$(cat $luks_dir/iank)
+luks_file=$luks_dir/host-$HOSTNAME
+if [[ ! -e $luks_file ]]; then
+ hostkeys=($luks_dir/host-*)
+ # if there is only one key, we might be deploying somewhere
+ # where dhcp doesnt give us a proper hostname, so use that.
+ if [[ ${#hostkeys[@]} == 1 && -e ${hostkeys[0]} ]]; then
+ luks_file=${hostkeys[0]}
+ else
+ echo "$0: error: no key for hostname at $luks_file" >&2
+ exit 1
+ fi
fi
+
+# # note, corresponding changes in /b/ds/keyscript-{on,off}
if ifclass demohost; then
- lukspw=x
+ lukspw=x
+elif [[ -e $luks_dir/$HOSTNAME ]]; then
+ lukspw=$(cat $luks_dir/$HOSTNAME)
+else
+ lukspw=$(cat $luks_dir/iank)
fi
# 1.5 x based on https://access.redhat.com/documentation/en-US/Red_Hat_Enterprise_Linux/7/html/Installation_Guide/sect-disk-partitioning-setup-x86.html#sect-custom-partitioning-x86
swap_mib=$(( $(grep ^MemTotal: /proc/meminfo | \
- awk '{print $2}') * 3/(${#devs[@]} * 2 ) / 1024 ))
+ awk '{print $2}') * 3/(${#devs[@]} * 2 ) / 1024 ))
mkdir -p /tmp/fai
root_devs=()
for dev in ${devs[@]}; do
- root_devs+=(`rootdev`)
+ root_devs+=($(rootdev))
done
shopt -s nullglob
+
+# We write to these files instead of just /etc/fstab, /etc/crypttab,
+# because these are filesystems created after our current root, and so
+# this allows us to update other root filesystems too.
+rm -f /mnt/root/root2-{fs,crypt}tab
+if $mkroot2; then
+ if $partition; then
+ echo $0: error: found partition=true but have mkroot2 arg
+ exit 1
+ fi
+ for dev in ${devs[@]}; do
+ luks_file=$luks_dir/host-amy
+ lukspw=$(cat $luks_dir/amy)
+ luks-setup $(root2dev)
+ cat >>/mnt/root/root2-crypttab <<EOF
+$(root2-cryptname) $(root2dev) $luks_file discard,luks,initramfs
+EOF
+ done
+ bpart $(for dev in ${devs[@]}; do root2-cryptdev; done)
+ bpart ${boot2_devs[@]}
+ mkdir -p /mnt/root2 /mnt/boot2
+ cat >>/mnt/root/root2-fstab <<EOF
+$(root2-cryptdev ${devs[0]}) /mnt/root2 btrfs nofail,x-systemd.device-timeout=30s,x-systemd.mount-timeout=30s,noatime,subvolid=0$mopts 0 0
+${boot2_devs[0]} /mnt/boot2 btrfs nofail,x-systemd.device-timeout=30s,x-systemd.mount-timeout=30s,noatime,subvolid=0 0 0
+EOF
+ exit 0
+fi
+
+
if $partition; then
+ ### 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
+ # 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 $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
- for dev in ${devs[@]}; do
- # parted will round up the disk size. Do -1 so we can have
- # fully 1MiB unit partitions for easy resizing of the last partition.
- # Otherwise we would pass in -0 for the end argument for the last partition.
- #
- # parted print error output is expected. example:
- # Error: /dev/vda: unrecognised disk label
- disk_mib=$(( $(parted -m $dev unit MiB print | \
- sed -nr "s#^/dev/[^:]+:([0-9]+).*#\1#p") - 1))
- root_end=$(( disk_mib - swap_mib - boot_mib / ${#boot_devs[@]} ))
- swap_end=$(( root_end + swap_mib))
-
- parted -s $dev mklabel gpt
- # MiB because parted complains about alignment otherwise.
- pcmd="parted -a optimal -s -- $dev"
- $pcmd mkpart primary "ext3" 12MiB ${root_end}MiB
- $pcmd mkpart primary "linux-swap" ${root_end}MiB ${swap_end}MiB
- $pcmd mkpart primary "" ${swap_end}MiB ${disk_mib}MiB
- # 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
- # 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 set $bios_grubn bios_grub on
- $pcmd set $bootn boot on # generally not needed on modern systems
- # the mkfs failed before on a vm, which prompted me to add
- # sleep .1
- # then it failed again on a physical machine
- # with:
- # Device /dev/disk/by-id/foo doesn't exist or access denied,
- # so I added a wait until it existed.
- # Then I added the mkfs.ext2, which claimed to succeed,
- # but then couldn't be found upon reboot. In that case we didn't
- # wait at all. So I've added a 3 second minimum wait.
- sleep 3
- secs=0
- while [[ ! -e `rootdev` ]] && (( secs < 10 )); do
- sleep 1
- secs=$((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 luksAddKey --key-file $luks_dir/host-$HOSTNAME \
- `rootdev` || [[ $? == 141 ]]
- # background: Keyfile and password are treated just
- # like 2 ways to input a passphrase, so we don't actually need to have
- # different contents of keyfile and passphrase, but it makes some
- # security sense to a really big randomly generated passphrase
- # as much as possible, so we have both.
- #
- # This would remove the keyfile.
- # yes 'test' | cryptsetup luksRemoveKey /dev/... \
- # /key/file || [[ $? == 141 ]]
-
- cryptsetup luksOpen `rootdev` `root-cryptname` \
- --key-file $luks_dir/host-$HOSTNAME
+ fi
+ ### end wipefs
+
+
+ # When we have 2 disks of at least 100g difference in size,
+ # make an extra partition on the end of the bigger one.
+ even_big_part=false
+ even_diff_min=100000
+ if $even_raid; then
+ smalli=0
+ bigi=1
+ if (( $(dev-mib ${devs[0]}) >= $(dev-mib ${devs[1]}) )); then
+ smalli=1
+ bigi=0
+ fi
+ disk_mib=$(dev-mib ${devs[smalli]})
+ even_big_dev=${devs[bigi]}
+ even_big_mib=$(dev-mib $even_big_dev)
+ if (( even_big_mib - disk_mib > even_diff_min )); then
+ even_big_part=true
+ fi
+ fi
+
+ 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.
+ #
+ # Note: parted print error output is expected. example:
+ # Error: /dev/vda: unrecognised disk label
+ if ! $even_raid; then
+ disk_mib=$(dev-mib)
+ fi
+
+ boot_part_mib=$(( boot_mib / ${#boot_devs[@]} ))
+ boot2_part_mib=$(( boot2_mib / ${#boot_devs[@]} ))
+ root2_part_mib=$(( root2_mib / ${#root_devs[@]} ))
+ build_part_mib=$(( build_mib / ${#root_devs[@]} ))
+ whonix_part_mib=$(( whonix_mib / ${#root_devs[@]} ))
+ browse_part_mib=$(( browse_mib / ${#root_devs[@]} ))
+ root_end=$(( disk_mib - boot_part_mib - boot2_part_mib ))
+ boot_end=$(( root_end + boot_part_mib ))
+
+ parted -s $dev mklabel gpt
+ # MiB because parted complains about alignment otherwise.
+ pcmd="parted -a optimal -s -- $dev"
+ # root partition, the main big one
+ $pcmd mkpart primary ext2 524MiB ${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 set $rootn lvm on
+ $pcmd name $rootn root
+ # boot partition
+ $pcmd mkpart primary "" ${root_end}MiB ${boot_end}MiB
+ $pcmd name $bootn boot
+ # boot2 partition
+ $pcmd mkpart primary "" ${boot_end}MiB ${disk_mib}MiB
+ $pcmd name $boot2n boot2
+ # uefi partition. efi sucks, half a gig, rediculous.
+ $pcmd mkpart primary "fat32" 12MiB 524MiB
+ $pcmd name $efin efi
+ $pcmd set $efin esp on
+ # note, this is shown here: https://support.system76.com/articles/bootloader/
+ # but not mentioned https://wiki.archlinux.org/index.php/EFI_system_partition
+ # probably not needed
+ $pcmd set $bootn boot on
+ $pcmd set $boot2n boot on
+ # 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.
+ # grub_ext partition
+ $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.
+ # biols grub partition
+ $pcmd mkpart primary "" 1MiB 4MiB
+ $pcmd name $bios_grubn biosgrub
+ $pcmd set $bios_grubn bios_grub on
+ if $even_big_part && [[ $dev == "$even_big_dev" ]]; then
+ $pcmd mkpart primary ext3 ${disk_mib}MiB ${even_big_mib}MiB
+ $pcmd name $even_bign even_big
+ fi
+
+ # the mkfs failed before on a vm, which prompted me to add
+ # sleep .1
+ # then it failed again on a physical machine
+ # with:
+ # Device /dev/disk/by-id/foo doesn't exist or access denied,
+ # 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.
+ secs=0
+ while [[ ! -e $(bios_grubdev) ]] && (( secs < 10 )); do
+ sleep 1
+ secs=$((secs +1))
done
- ls -la /dev/btrfs-control # this was probably for debugging...
- sleep 1
- bpart $(for dev in ${devs[@]}; do root-cryptdev; done)
- bpart ${boot_devs[@]}
+ sleep 3
+
+ mkfs.fat -F32 $(efidev)
+
+ if $even_big_part && [[ $dev == "$even_big_dev" ]]; then
+ luks-setup $(even_bigdev)
+ mkfs.btrfs -f $(crypt-dev $(even_bigdev))
+ fi
+
+ # 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)
+ luks-setup $(rootdev)
+
+ if [[ $SPECIAL_DISK ]]; then
+ exit 0
+ fi
+ done
+ ls -la /dev/btrfs-control # this was probably for debugging...
+ sleep 1
+ pvcreate $(root-cryptdev)
+ vgcreate vg$dev $(root-cryptdev)
+ lvcreate -n lvroot2 -L ${root2_part_mib}m vg$dev
+ lvcreate -n lvbuild -L ${build_part_mib}m vg$dev
+ lvcreate -n lvwhonix -L ${whonix_part_mib}m vg$dev
+ lvcreate -n lvbrowse -L ${browse_part_mib}m vg$dev
+ lvcreate -n lvroot -L 95%FREE vg$dev
+ bpart $(for dev in ${devs[@]}; do echo /dev/vg$dev/lvroot; 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
+ for dev in ${devs[@]}; do
+ if [[ -e /dev/mapper/$(root-cryptname) ]]; then
+ continue
+ fi
+ cryptsetup luksOpen $(rootdev) $(root-cryptname) \
+ --key-file $luks_file
+ done
+ sleep 1
fi
-if [[ $DISTRO != debianstable_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
+if $wipe && [[ $DISTRO != debianbullseye_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
+
+ # could set default subvol like this, but no reason to.
+ # btrfs subvolume set-default \
+ # $(btrfs subvolume list . | grep "root_$DISTRO$" | awk '{print $2}') .
+
+ # For raid systems, cow allows for error correction, for non-raid systems,
+ # protects root fs from having the plug pulled. Reprovisioning a root
+ # subvol is not my favorite thing to do.
+ # # no cow on the root filesystem. it's setup is fully scripted,
+ # # if it's messed up, we will just recreated it,
+ # # and we can get better perf with this.
+ # # I can't remember exactly why, but this is preferable to mounting with
+ # # -o nodatacow, I think because subvolumes inherit that.
+ # chattr -Rf +C root_$DISTRO
+ cd /
+ umount /mnt
fi
mount -o subvolid=0 $first_boot_dev /mnt
mkdir -p /mnt/grub2
cp $FAI/distro-install-common/libreboot_grub.cfg /mnt/grub2
-if [[ $DISTRO == debianstable_bootstrap ]]; then
- # this is just convenience for the libreboot_grub config
- # so we can glob the other ones easier.
- boot_vol=$DISTRO
+if [[ $DISTRO == debianbullseye_bootstrap ]]; then
+ # this is just convenience for the libreboot_grub config
+ # so we can glob the other ones easier.
+ boot_vol=$DISTRO
else
- boot_vol=boot_$DISTRO
+ boot_vol=boot_$DISTRO
+fi
+if $wipe && [[ -e /mnt/$boot_vol ]]; then
+ btrfs subvolume delete /mnt/$boot_vol
+fi
+if [[ ! -e /mnt/$boot_vol ]]; then
+ btrfs subvolume create $boot_vol
fi
-[[ ! -e /mnt/$boot_vol ]] || btrfs subvolume delete /mnt/$boot_vol
-btrfs subvolume create $boot_vol
cd /
umount /mnt
## end create subvols ##
grub-editenv /mnt/grubenv set last_boot=/$boot_vol
umount /mnt
-if [[ $DISTRO == debianstable_bootstrap ]]; then
- cat > /tmp/fai/fstab <<EOF
+fstabstd=x-systemd.device-timeout=30s,x-systemd.mount-timeout=30s
+if [[ $DISTRO == debianbullseye_bootstrap ]]; then
+ cat > /tmp/fai/fstab <<EOF
$first_boot_dev / btrfs noatime,subvol=$boot_vol 0 0
+$first_efi /boot/efi vfat nofail,$fstabstd 0 0
EOF
- cat >/tmp/fai/disk_var.sh <<EOF
+ cat >/tmp/fai/disk_var.sh <<EOF
BOOT_DEVICE="${short_devs[@]}"
ROOT_PARTITION=$first_boot_dev
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=$boot_vol 0 0
+ # note, fai creates the mountpoints listed here
+ cat > /tmp/fai/fstab <<EOF
+$first_root_crypt / btrfs $fstabstdopts,noatime,subvol=root_$DISTRO$mopts 0 0
+$first_root_crypt /mnt/root btrfs nofail,$fstabstd,noatime,subvolid=0$mopts 0 0
+$first_boot_dev /boot btrfs nofail,$fstabstd,noatime,subvol=$boot_vol 0 0
+$first_efi /boot/efi vfat nofail,$fstabstd 0 0
+$first_boot_dev /mnt/boot btrfs nofail,$fstabstd,noatime,subvolid=0 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
+ swaps=()
+ rm -f /tmp/fai/crypttab
+ for dev in ${devs[@]}; do
+ swaps+=($(swap-cryptname))
+ cat >>/tmp/fai/crypttab <<EOF
+$(root-cryptname) $(rootdev) none keyscript=/root/keyscript,discard,luks,initramfs
+$(swap-cryptname) $(swapdev) /dev/urandom swap,cipher=aes-xts-plain64,size=256,hash=ripemd160
EOF
- cat >> /tmp/fai/fstab <<EOF
-`swap-cryptdev` none swap sw 0 0
+ cat >> /tmp/fai/fstab <<EOF
+$(swap-cryptdev) none swap nofail,$fstabstd,sw 0 0
EOF
- done
+ done
- # fai would do this:
- #BOOT_DEVICE=\${BOOT_DEVICE:-"${devs[0]}"}
+ # fai would do this:
+ #BOOT_DEVICE=\${BOOT_DEVICE:-"${devs[0]}"}
- # note: swaplist seems to do nothing.
- cat >/tmp/fai/disk_var.sh <<EOF
+ # note: swaplist seems to do nothing.
+ cat >/tmp/fai/disk_var.sh <<EOF
BOOT_DEVICE="${short_devs[@]}"
BOOT_PARTITION=\${BOOT_PARTITION:-$first_boot_dev}
# ROOT_PARTITIONS is added by me, used in arch setup.
ROOT_PARTITION=\${ROOT_PARTITION:-$first_root_crypt}
SWAPLIST=\${SWAPLIST:-"${swaps[@]}"}
EOF
+
+ if [[ $HOSTNAME == kd ]]; then
+ # note, having these with keyscript and initramfs causes a luks error in fai.log,
+ # but it is safely ignorable and gets us the ability to just type our password
+ # in once at boot. A downside is that they are probably needed to be plugged in to boot.
+ cat >>/tmp/fai/crypttab <<EOF
+crypt_dev_ata-Samsung_SSD_870_QVO_8TB_S5VUNG0N900656V-part${even_bign} /dev/disk/by-id/ata-Samsung_SSD_870_QVO_8TB_S5VUNG0N900656V-part7 none keyscript=decrypt_keyctl,discard,luks,initramfs
+crypt_dev_ata-TOSHIBA_MD04ACA500_84R2K773FS9A-part1 /dev/disk/by-id/ata-TOSHIBA_MD04ACA500_84R2K773FS9A-part1 none keyscript=decrypt_keyctl,discard,luks,initramfs
+crypt_dev_ata-ST6000DM001-1XY17Z_Z4D29EBL-part1 /dev/disk/by-id/ata-ST6000DM001-1XY17Z_Z4D29EBL-part1 none keyscript=decrypt_keyctl,discard,luks,initramfs
+EOF
+ cat >> /tmp/fai/fstab <<EOF
+# r7 = root partition7. it isnt actually #7 anymore, not a great name, but whatever
+/dev/mapper/crypt_dev_ata-Samsung_SSD_870_QVO_8TB_S5VUNG0N900656V-part${even_bign} /mnt/r7 btrfs nofail,$fstabstd,noatime,compress=zstd,subvolid=0 0 0
+/dev/mapper/crypt_dev_ata-TOSHIBA_MD04ACA500_84R2K773FS9A-part1 /mnt/rust1 btrfs nofail,$fstabstd,noatime,compress=zstd,subvolid=0 0 0
+/dev/mapper/crypt_dev_ata-ST6000DM001-1XY17Z_Z4D29EBL-part1 /mnt/rust2 btrfs nofail,$fstabstd,noatime,compress=zstd,subvolid=0 0 0
+EOF
+ fi
+
fi
+
+# initial setup of extra data fs, mounted,
+# btrfs subvol create nocow
+# chattr +C nocow
+# chown iank.iank nocow
+