#!/bin/bash -x # Copyright (C) 2016 Ian Kelling # 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 2 # 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, 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 "$0: error: failed to find devbyid script" >&2 exit 1 fi fi #### begin configuration # this is the ordering of the /dev/sdaX, but # the ordering of the partition layout goes like this: # bios_grub # grub_ext # efi # root # swap # boot rootn=1 root2n=2 swapn=3 bootn=4 boot2n=5 efin=6 # ext partition so grub can write persistent variables, # so it can do a one time boot. grub can't write to # btrfs or any cow fs because it's more # more complicated to do and they don't want to. grub_extn=7 # bios boot partition, # https://wiki.archlinux.org/index.php/GRUB bios_grubn=8 even_bign=9 lastn=$bios_grubn ##### end configuration add-part() { # add partition suffix to $dev local d part if [[ $# == 1 ]]; then d=$dev part=$1 else d=$1 part=$2 fi echo $d-part$part } rootdev() { add-part $@ $rootn; } root2dev() { add-part $@ $root2n; } 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##*/}; } root-cryptdev() { crypt-dev $(rootdev $@); } root2-cryptdev() { crypt-dev $(root2dev $@); } swap-cryptdev() { crypt-dev $(swapdev $@); } root-cryptname() { crypt-name $(rootdev $@); } root2-cryptname() { crypt-name $(root2dev $@); } swap-cryptname() { crypt-name $(swapdev $@); } dev-mib() { local d=${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 $luks_file } ##### end function defs if ifclass REPARTITION; then partition=true # force a full wipe else partition=false # change to true to force a full wipe fi if ifclass NOWIPE; then wipe=false else wipe=true fi if (($(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=() # 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 done # 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[@]} ) else 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?*) 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 fi devs=() shopt -s extglob for short_dev in ${short_devs[@]}; do devs+=($(devbyid $short_dev)) done if [[ ! ${devs[0]} ]]; then echo "$0: error: failed to detect devs" >&2 exit 1 fi 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 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 ### Begin calculate boot partition space # due to raid duplication case $raid_level in 1*) boot_space=$(( boot_space / 2 )) ;; esac 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 elif (( boot_spa_ce > 30000 )); then boot_mib=$(( 5000 + (boot_space - 30000) / 2 )) root2_mib=100 boot2_mib=100 else # 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 )) ;; esac ### end calculate boot partition space if [[ ! $DISTRO ]]; then 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_BOOKWORM; then DISTRO=debianbookworm 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 bpart() { # btrfs a partition 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 } 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 "$0: error: no luks_dir found" >&2 exit 1 fi fi luks_file=$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 elif [[ -e $luks_dir/$HOSTNAME ]]; then lukspw=$(cat $luks_dir/$HOSTNAME) else lukspw=$(cat $luks_dir/iank) fi first_root_crypt=$(root-cryptdev ${devs[0]}) # 1.5 x based on https://access.redhat.com/documentation/en-US/Red_Hat_Enterprise_Linux/7/html/Installation_Guide/sect-disk-partitioning-setup-x86.html#sect-custom-partitioning-x86 swap_mib=$(( $(grep ^MemTotal: /proc/meminfo | \ awk '{print $2}') * 3/(${#devs[@]} * 2 ) / 1024 )) mkdir -p /tmp/fai root_devs=() for dev in ${devs[@]}; do root_devs+=($(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 <>/mnt/root/root2-fstab < 0 )) || exit 1 done done done 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 / ${#root_devs[@]} )) root_end=$(( disk_mib - root2_part_mib - swap_mib - boot_part_mib - boot2_part_mib )) root2_end=$(( root_end + root2_part_mib )) swap_end=$(( root2_end + swap_mib )) boot_end=$(( swap_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 ext3 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 name $rootn root # root2 partition $pcmd mkpart primary ext3 ${root_end}MiB ${root2_end}MiB $pcmd name $root2n root2 # normally a swap is type "linux-swap", but this is encrypted swap. using that # label will confuse systemd. # swap partition $pcmd mkpart primary "" ${root2_end}MiB ${swap_end}MiB $pcmd name $swapn swap # boot partition $pcmd mkpart primary "" ${swap_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 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 bpart $(for dev in ${devs[@]}; do root-cryptdev; done) bpart ${boot_devs[@]} else 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 $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 cd /mnt 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 [[ $DISTRO == debianbullseye_bootstrap ]]; then # this is just convenience for the libreboot_grub config # so we can glob the other ones easier. boot_vol=$DISTRO else boot_vol=boot_$DISTRO fi if $wipe && [[ -e /mnt/$boot_vol ]]; then btrfs subvolume delete /mnt/$boot_vol fi if [[ ! -e /mnt/$boot_vol ]]; then btrfs subvolume create $boot_vol fi cd / umount /mnt ## end create subvols ## dev=${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 fstabstd=x-systemd.device-timeout=30s,x-systemd.mount-timeout=30s if [[ $DISTRO == debianbullseye_bootstrap ]]; then cat > /tmp/fai/fstab </tmp/fai/disk_var.sh < /tmp/fai/fstab <>/tmp/fai/crypttab <> /tmp/fai/fstab </tmp/fai/disk_var.sh <>/tmp/fai/crypttab <> /tmp/fai/fstab <