X-Git-Url: https://iankelling.org/git/?a=blobdiff_plain;f=fai%2Fconfig%2Fhooks%2Fpartition.DEFAULT;h=11a7ee51f57839a01d04bc1436601b0c4e310d7b;hb=HEAD;hp=46887e9712d1aa558fb08bcab3be976a4fce6723;hpb=43480add5b3ab6f4e407065454bd6be1d1fe1870;p=automated-distro-installer diff --git a/fai/config/hooks/partition.DEFAULT b/fai/config/hooks/partition.DEFAULT index 46887e9..530a34d 100755 --- a/fai/config/hooks/partition.DEFAULT +++ b/fai/config/hooks/partition.DEFAULT @@ -15,438 +15,1077 @@ # along with this program; if not, write to the Free Software # Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. +# todo /boot/chboot needs update for lvm i think? + +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 without args: +# +# 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 +# +# ARGS (only 1 is valid): +# +# mkroot2: for running outside of fai and setting up the root2/boot2 luks and btrfs and tab files +# +# mkroot2tab: for running outside of fai and setting up the root2/boot2 tab files, in case luks and btrfs +# happen to already be setup. +# +# mktab: for running outside of fai and generating a crypttab for +# the main root fs in /tmp/fai. Must run with env var, eg export DISTRO=trisquelaramo. +# +# Example use in a bootstrap distro: +# scp /a/bin/fai/fai/config/{distro-install-common/devbyid,hooks/partition.DEFAULT} root@HOST: +# sl HOST +# export DISTRO=trisquelnabia; ./partition.DEFAULT mktab +# ## cryptsetup wont take within a pipeline +# mapfile -t lines < <(awk '! /swap/ {print $2,$1}' /tmp/fai/crypttab ) +# for l in "${lines[@]}"; do cryptsetup luksOpen $l; done +# +# # or alternatively, to avoid typing it many times: +# read -r lukspw; for l in "${lines[@]}"; do yes "$lukspw" | cryptsetup luksOpen $l; done + +## potentially useful later: +# sed 's#/root/keyscript,#decrypt_keyctl,#;s/$/,noauto/' /tmp/fai/crypttab >/etc/crypttab +# +# environment variables: +# +# HOSTNAME: if demohost, we set the luks password to just +# 'x'. Used in various other ways too. +# +# 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 +# +# REROOTFS: Don't reuse the root filesystem, even if we normally would +# +# ROTATIONAL: forces to install onto hdds instead of sdds. normally sdds +# are chosen if they exist. +# +# PARTITION_PROMPT: command line prompt before partitioning. This is good +# to set if we don't expect repartitioning to happen. +# +# ROTATIONAL: in a system with ssd and hdd, install to the hdd +# instead of the default ssd. +# +# RAID0: forces raid0 filesystem. Normally with 4+ devices, we use +# raid10. +# RAID1: forces raid1 filesystem. +# RAID1c3: forces raid1c3 filesystem (btrfs raid 1, 3 copies). + +mkroot2tab=false +mkroot2=false +mktab=false +if [[ $1 ]]; then + ## duplicates fai-wrapper, for convenience of not needing it + if ! type -t ifclass &>/dev/null; then + ifclass() { + local var=${1/#/CLASS_} + [[ $HOSTNAME == "$1" || ${!var} ]] + } + fi + + case $1 in + mkroot2) + mkroot2=true + ;; + mkroot2tab) + mkroot2tab=true + ;; + mktab) + mktab=true + ;; + *) + echo "$0: error: unsupported arg: $1" >&2 + exit 1 + ;; + esac +fi + +if [[ ! $SPECIAL_DISK ]] && ! $mkroot2 && ! $mkroot2tab && ! $mktab \ + && ! ifclass IANK && ! ifclass FSF; then + echo $0: error: need class IANK or FSF or SPECIAL_DISK for running in fai +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 -skiptask partition || ! type skiptask # for running not in fai +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 "$0: error: failed to find devbyid script" >&2 + exit 1 + fi +fi #### begin configuration -rootn=1 -swapn=2 -bootn=3 -# ext partition so grub can write persistent variables, -# so it can do a one time boot. -grub_extn=4 -# bios boot partition, -# https://wiki.archlinux.org/index.php/GRUB -bios_grubn=5 -lastn=$bios_grubn -boot_mib=10000 +# this is the ordering of the /dev/sdaX, but +# the ordering of the partition layout goes like this: +# bios_grub +# grub_ext +# efi +# lvm +# root +# swap +# boot ##### end configuration +##### begin function defs -add-part() { # add partition suffix to $dev - local d ret - if [[ $# == 1 ]]; then - d=$dev - part=$1 - else - d=$1 - part=$2 +bpart() { # btrfs a partition + case $raid_level in + 0) mkfs.btrfs -f $@ ;; + *) mkfs.btrfs -f -m raid$raid_level -d raid$raid_level $@ ;; + esac +} + + +zilap() { + case $HOSTNAME in + sy|bo|so) + return 0 + ;; + esac + return 1 +} + +getluks() { + 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 - if [[ $d == /dev/disk/by-id/* ]]; then - ret=$d-part$part + fi + + luks_file=$luks_dir/host-$HOSTNAME + if [[ ! -e $luks_file ]]; then + # shellcheck disable=SC2206 # globbing is intended + hostkeys=($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 - ret=$d$part + echo "$0: error: no key for hostname at $luks_file" >&2 + exit 1 fi - echo $ret + 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 + + if $mkroot2; then + luks_file=$luks_dir/host-amy + lukspw=$(cat $luks_dir/amy) + fi } -bootdev() { add-part $@ $bootn; } -rootdev() { add-part $@ $rootn; } -swapdev() { add-part $@ $swapn; } -grub_extdev() { add-part $@ $grub_extn; } -# Commented because it's not used, but left because it -# finishes the pattern and if we ever do need to use it, it's here. -#bios_grubdev() { add-part $@ $bios_grubn; } -crypt-dev() { echo /dev/mapper/crypt_dev_${1##*/}; } -crypt-name() { echo crypt_dev_${1##*/}; } -root-cryptdev() { crypt-dev $(rootdev $@); } -swap-cryptdev() { crypt-dev $(swapdev $@); } -root-cryptname() { crypt-name $(rootdev $@); } -swap-cryptname() { crypt-name $(swapdev $@); } +fsf() { + ifclass FSF +} + + +dev-mib() { + local d=${1:-$dev} + echo $(( $(parted -m $d unit MiB print | \ + sed -nr "s#^/dev/[^:]+:([0-9]+).*#\1#p") - 1)) +} + +luks-setup() { + local luksdev="$1" + local cryptname="$2" + # 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 $cryptname --key-file $luks_file +} + +mktab() { + mkdir -p /tmp/fai + dev=${boot_devs[0]} + fstabstd="x-systemd.device-timeout=30s,x-systemd.mount-timeout=30s" + + if [[ $DISTRO == *_bootstrap ]]; then + cat > /tmp/fai/fstab </tmp/fai/disk_var.sh < /tmp/fai/fstab <> /tmp/fai/fstab <>/tmp/fai/crypttab <> /tmp/fai/fstab <> /tmp/fai/fstab </tmp/fai/disk_var.sh <>/tmp/fai/crypttab <> /tmp/fai/fstab <>/tmp/fai/crypttab <> /tmp/fai/fstab <>/mnt/root/root2-crypttab <>/mnt/root/root2-fstab < 2)); then + mopts=,compress=zstd +fi + +declare -A disk_excludes +if ! $mkroot2 && ! $mkroot2tab && ! $mktab ! ifclass USE_MOUNTED; then + ## ignore disks that are mounted, eg when running from fai-cd + while read -r l; do + eval "$l" + if [[ ! $PKNAME ]]; then + # shellcheck disable=SC2153 # not a misspelling + PKNAME="$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". vda disk has empty tran (transport). This may need adjustment +# for some new type we come across. cdrom has type "rom" +for disk in $(lsblk -ndo name,type,tran | awk '$3 ~ "^(sata|nvme|)$" && $2 == "disk" { print $1 }'); do + if [[ ${disk_excludes[$disk]} ]]; then + continue + fi + case $disk in + # cdrom + sr*) continue ;; + esac + case $(cat /sys/block/$disk/queue/rotational) in + 0) + ssds+=("/dev/$disk") + echo $0: found ssd /dev/$disk + ;; + 1) + hdds+=("/dev/$disk") + echo $0: found hdd /dev/$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. if ! ifclass ROTATIONAL && (( ${#ssds[@]} > 0 )); then - short_devs=( ${ssds[@]} ) + read -ra short_devs<<<"${ssds[@]}" else - short_devs=( ${hdds[@]} ) + read -ra short_devs<<<"${hdds[@]}" fi +pvn=1 + +bootn=2 + +# rootn=1 +# root2n=2 +# swapn=3 +# bootn=4 +# boot2n=5 + +efin=3 +# 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 +# for an even raid (raid 1), when one disk is bigger, this partition goes on the big disk +even_bign=6 +# even_bign only exists in some cases +lastn=$bios_grubn # check if the partitions exist have the right filesystems #blkid="$(blkid -s TYPE)" for dev in ${short_devs[@]}; do - if $partition; then break; fi - y=$(readlink -f $dev) - x=($y[0-9]) - [[ ${#x[@]} == "${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 $dev) + # shellcheck disable=SC2206 # globbing is intended + arr=($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 "Press any key except ctrl-c to continue and partition these drives:" + if [[ $SPECIAL_DISK ]]; then + echo $SPECIAL_DISK + else echo " ${short_devs[*]}" - read -r + fi + read -r fi devs=() +vgs=() +root_devs=() +o_devs=() +swap_devs=() shopt -s extglob +partsuffix=-part for short_dev in ${short_devs[@]}; do - devs+=($(devbyid $short_dev)) + dev="$(devbyid $short_dev)" + if [[ $dev != */by-id/* ]]; then + # no by-id link, assume we are in a vm and this is true for all devs. + partsuffix= + fi + # for vms, cant name a vg the same as the short device name, they + # conflict: /dev/$vg is already taken + + dname=${dev##*/} + vg=vg$dname + vg=${vg//:/} + vgs+=("$vg") + devs+=("$dev") + if fsf; then + root_devs+=(/dev/$vg/root) + swap_devs+=(/dev/$vg/swap) + else + o_devs+=(/dev/mapper/crypt-$vg-o) + root_devs+=(/dev/mapper/crypt-$vg-root) + swap_devs+=(/dev/mapper/crypt-$vg-swap) + fi done +first_root_dev=${root_devs[0]} +if [[ ! $SPECIAL_DISK && ! ${devs[0]} ]]; then + echo "$0: error: failed to detect devs" >&2 + exit 1 +fi + + +pvsuf=$partsuffix$pvn +bootsuf=$partsuffix$bootn +efisuf=$partsuffix$efin +grub_extsuf=$partsuffix$grub_extn +# We dont do anything with this partition here, so this +# is be unused, but left as a comment for completing the pattern +# of all the suffixes. +#bios_grubsuf=$partsuffix$bios_grubn +even_bigsuf=$partsuffix$even_bign -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 && $first ]]; then - first_grub_extdev=`grub_extdev` - first=false + vg=vg${dev##*/} + vg=${vg//:/} + # I ran into a machine (frodo) 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 $bad_disk; then + continue + fi + boot_devs+=($dev$bootsuf) + boot2_devs+=(/dev/$vg/boot2) + boot_space=$(( boot_space + $(parted -m $dev unit MiB print | \ + sed -nr "s#^/dev/[^:]+:([0-9]+).*#\1#p") - 1)) + if $first; then + first_efi=$dev$efisuf + first_grub_extdev=$dev$grub_extsuf + first=false + fi done +first_boot_dev=${boot_devs[0]} -if ifclass RAID0 || (( ${#boot_devs[@]} < 4 )); then - raid_level=0 +even_raid=false +if ifclass RAID0 || (( ${#boot_devs[@]} == 1 )); then + raid_level=0 + raid_duplication=1 +elif ifclass RAID1 || (( ${#boot_devs[@]} == 2 )); then + if (( ${#boot_devs[@]} == 2 )); then + even_raid=true + fi + raid_level=1 + raid_duplication=2 +elif ifclass RAID1c3 || (( ${#boot_devs[@]} == 3 )); then + raid_level=1c3 + raid_duplication=3 else - raid_level=10 - # need double the space if we are raid 10, and then - # might as well give some extra. - boot_mib=$((boot_mib * 3)) + raid_level=10 + raid_duplication=2 fi -if [[ ! $DISTRO ]]; then - if ifclass STABLE_BOOTSTRAP; then - DISTRO=debianstable_bootstrap - elif ifclass STRETCH64; then - DISTRO=debiantesting - elif ifclass STABLE; then - DISTRO=debianstable - elif ifclass XENIAL64; then - DISTRO=ubuntuxenial - elif ifclass BELENOS64; then - DISTRO=trisquelbelenos - else - echo "PARTITIONER ERROR: no distro class/var set" >&2 - exit 1 - fi +### Begin calculate boot partition space +# due to raid duplication +case $raid_level in + 1|10) boot_space=$(( boot_space / 2 )) ;; + 1c3) boot_space=$(( boot_space / 3 )) ;; +esac +if fsf; then + boot_mib=6000 +elif (( boot_space > 900000 )); then + # this is larger than needed for several /boot subvols, + # becuase I keep a minimal debian install on it for + # recovery needs and for doing pxe-kexec. + boot_mib=10000 + root2_mib=500000 + boot2_mib=5000 +elif (( boot_space > 30000 )); then + boot_mib=$(( 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. u20.04 with 3 kernels = + # 308 mb, so things have grown significantly + boot_mib=1000 + root2_mib=100 + boot2_mib=100 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 $@ ;; - esac -} +boot_part_mib=$(( boot_mib * raid_duplication / ${#boot_devs[@]} )) +if zilap; then + boot2_part_mib=$(( boot2_mib * raid_duplication / ${#boot_devs[@]} )) + root2_part_mib=$(( root2_mib * raid_duplication / ${#root_devs[@]} )) +else + boot2_part_mib=0 + root2_part_mib=0 +fi +### end calculate boot partition space -# keyfiles generated like: -# head -c 2048 /dev/urandom | od | s dd of=/q/root/luks/host-demohost -luks_dir=${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 [[ ! $DISTRO ]]; then + if ifclass VOL_BOOKWORM_BOOTSTRAP; then + DISTRO=debianbookworm_bootstrap + elif 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_JAMMY; then + DISTRO=ubuntujammy + elif ifclass VOL_FLIDAS; then + DISTRO=trisquelflidas + elif ifclass VOL_ETIONA; then + DISTRO=trisqueletiona + elif ifclass VOL_NABIA; then + DISTRO=trisquelnabia + elif ifclass VOL_ARAMO; then + DISTRO=trisquelaramo + elif $mkroot2 || $mkroot2tab; then + : + else + echo "PARTITIONER ERROR: no distro class/var set" >&2 exit 1 + fi fi -if ifclass tp; then - lukspw=$(cat $luks_dir/traci) +if [[ $DISTRO == *_bootstrap ]]; then + # this is just convenience for the libreboot_grub config + # so we can glob the other ones easier. + boot_vol=$DISTRO else - lukspw=$(cat $luks_dir/ian) -fi -if ifclass demohost; then - lukspw=x + boot_vol=boot_$DISTRO 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 )) + 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 +##### end variable setup + + + + +if $mkroot2 || $mkroot2tab; then + getluks + doroot2 +elif $mktab; then + mktab + exit 0 +else + mktab + if ! fsf; then + getluks + fi +fi + + if $partition; 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 + ### begin wipefs + if [[ ! $SPECIAL_DISK ]]; then + + # we do lvm removals just for the disks we are using + pv_wipes=() + vg_wipes=() + declare -A vg_map + pv_devs=$(pvs --noheadings -o pvname) + for pv_dev in $pv_devs; do + for short_dev in ${short_devs[@]}; do + if [[ $pv_dev == $short_dev* ]]; then + pv_wipes+=($pv_dev) + vgs_of_pv=$(pvs --noheadings -o vgname $pv_dev) + for vg in $vgs_of_pv; do + if [[ ${vg_map[$vg]} ]]; then + continue + fi + vg_map[$vg]=t + vg_wipes+=($vg) + lvs=$(vgs --noheadings -o lv_path $vg) + for lv in $lvs; do + wipefs -a $lv done - done + done + fi + 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 + + for vg in ${vg_wipes[@]}; do + vgchange -an $vg + vgremove -ff $vg 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 pv in ${pv_wipes[@]}; do + pvremove -ff $pv + done + for dev in ${devs[@]}; do - mkfs.ext2 `grub_extdev` - cryptsetup luksOpen `rootdev` `root-cryptname` \ - --key-file $luks_dir/host-$HOSTNAME + # if we repartition to the same as an old partition, + # we don't want any old fses hanging around. + count_down=10 + # wipefs has failed, manual run works, google suggests timing issue + while ! wipefs -a $dev; do + sleep 2 + count_down=$((count_down - 1)) + if (( count_down <= 0 )); then + echo "$0: wipefs failed 10 times. exiting" >&2 + exit 1 + fi + 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 + + if [[ $SPECIAL_DISK ]]; then + devs=($(devbyid $SPECIAL_DISK)) + fi + for dev in ${devs[@]}; do + vg=vg${dev##*/} + vg=${vg//:/} + + # 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 + + + parted -s $dev mklabel gpt + # MiB because parted complains about alignment otherwise. + pcmd="parted -a optimal -s -- $dev" + # main lvm partition + + pv_end=$(( disk_mib - boot_part_mib )) + $pcmd mkpart primary ext3 524MiB ${pv_end}MiB + $pcmd name $pvn pv + + # + 794 pvcreate -y /dev/disk/by-id/ata-ST4000DM000-1F2168_Z3028BKA-part1 + # WARNING: Device /dev/sde1 not initialized in udev database even after waiting 10000000 microseconds. + # No device found for /dev/disk/by-id/ata-ST4000DM000-1F2168_Z3028BKA-part1. + # sleep 10 was not enough. + secs=0 + while [[ ! -e $dev$pvsuf ]] && (( secs < 40 )); do + sleep 1 + secs=$((secs +1)) + done + sleep 3 + pvcreate -y -ff $dev$pvsuf + vgcreate -y -ff $vg $dev$pvsuf + + if fsf; then + root_mib=40000 + elif ifclass demohost; then + # just randomish numbers that seem ok for testing. + root_mib=25000 + o_mib=1000 + else + # This would maximize it, but we are going for a separate filesystem in /o, + # so use fixed sizes to allow both to grow + # 600 = uefi 512 + grubext 8 + bios grub 3 + some extra cuz this is lvm + #root_mib=$(( disk_mib - root2_part_mib - swap_mib - boot_part_mib - boot2_part_mib - 600 )) + o_mib=$(( 180 * 1000 )) + # max minus o, minus a gig just for some extra space + max_root_mib=$(( disk_mib - root2_part_mib - swap_mib - boot_part_mib - boot2_part_mib - 600 - o_mib - 1000 )) + root_mib=$(( 1700 * 1000 )) # * 1000 to make it in gb. + if (( max_root_mib < root_mib )); then + root_mib=$max_root_mib + fi + fi + + if [[ $SPECIAL_DISK ]]; then + lvcreate -y -L $max_root_mib $vg -n data + else + # -L unit default mebibyte + lvcreate -y -L $root_mib $vg -n root + if ! fsf; then + lvcreate -y -L $o_mib $vg -n o + fi + lvcreate -y -L $swap_mib $vg -n swap + # unencrypted swap needs mkswap + if fsf; then + mkswap /dev/$vg/swap + fi + fi + + $pcmd mkpart primary "" ${pv_end}MiB ${disk_mib}MiB + $pcmd name $bootn boot + $pcmd set $bootn boot on + + # uefi partition, for normal bios systems, its just in case. + $pcmd mkpart primary "fat32" 12MiB 524MiB + $pcmd name $efin efi + # note, this is shown here: https://support.system76.com/articles/bootloader/ + # but not mentioned https://wiki.archlinux.org/index.php/EFI_system_partition + # might not be needed + $pcmd set $efin esp 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 $dev$efisuf ]] && (( secs < 40 )); do + sleep 1 + secs=$((secs +1)) done + sleep 3 + + mkfs.fat -F32 $dev$efisuf + + if ! fsf && $even_big_part && [[ $dev == "$even_big_dev" ]]; then + luks-setup $even_big_dev$even_bigsuf ${even_big_dev##*/}$even_bigsuf + mkfs.btrfs -f /dev/mapper/${even_big_dev##*/}$even_bigsuf + 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 $dev$grub_extsuf + + if [[ $SPECIAL_DISK ]]; then + luks-setup /dev/$vg/data crypt-$vg-data + exit 0 + fi + + # for fsf, no encryption of root because root will not contain any + # sensitive data. + if ! fsf; then + luks-setup /dev/$vg/root crypt-$vg-root + luks-setup /dev/$vg/o crypt-$vg-o + fi + + done + ls -la /dev/btrfs-control # this was probably for debugging... + sleep 1 + + bpart ${root_devs[@]} + if ! fsf; then + bpart ${o_devs[@]} + fi + bpart ${boot_devs[@]} + +else ## end if $partition ## + + if ! fsf; then + for vg in ${vgs[@]}; do + if [[ -e /dev/mapper/crypt-$vg-root ]]; then + continue + fi + if $rerootfs; then + luks-setup /dev/$vg/root crypt-$vg-root + else + cryptsetup luksOpen /dev/$vg/root crypt-$vg-root \ + --key-file $luks_file + cryptsetup luksOpen /dev/$vg/o crypt-$vg-o \ + --key-file $luks_file + fi + done + fi + + if $rerootfs; then sleep 1 + bpart ${root_devs[@]} + fi + 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_$DISTRO - [[ -e q ]] || btrfs subvolume create q - chown root:1000 q - - mkdir -p /mnt/root_$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 $2}') . - - # no cow on the root filesystem. it's setup is fully scripted, - # (immutable in buzzwords). if it messes up, we will just recreated it, - # and we can get better perf with this. - # I can't remember exactly why, but this is preferable to mounting with - # -o nodatacow, I think because subvolumes inherit that. - chattr -Rf +C root_$DISTRO - cd / - umount /mnt +if $wipe && [[ $DISTRO != *_bootstrap ]]; then + # bootstrap distro doesn't use separate encrypted root. + mount -o subvolid=0 ${root_devs[0]} /mnt + # systemd creates subvolumes we want to delete. + mapfile -t s < <(btrfs subvolume list --sort=-path /mnt | + sed -rn "s#^.*path\s*(root_$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. +# for libreboot systems. grub2 only reads from subvolid=0 mkdir -p /mnt/grub2 +# todo: this would need some rework if we moved boot into +# lvm. 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 -else - boot_vol=boot_$DISTRO +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 $boot_vol cd / umount /mnt ## end create subvols ## -dev=${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 -if [[ $DISTRO == debianstable_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 <