iankelling.org Git - automated-distro-installer/blob - fai/config/hooks/partition.DEFAULT

   1 #!/bin/bash -x
   2 # Copyright (C) 2016 Ian Kelling
   3
   4 # This program is free software; you can redistribute it and/or
   5 # modify it under the terms of the GNU General Public License
   6 # as published by the Free Software Foundation; either version 2
   7 # of the License, or (at your option) any later version.
   8
   9 # This program is distributed in the hope that it will be useful,
  10 # but WITHOUT ANY WARRANTY; without even the implied warranty of
  11 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12 # GNU General Public License for more details.
  13
  14 # You should have received a copy of the GNU General Public License
  15 # along with this program; if not, write to the Free Software
  16 # Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
  17
  18 # todo /boot/chboot needs update for lvm i think?
  19
  20 PS4='+ $LINENO '
  21 set -eE -o pipefail
  22 trap 'echo "$0:$LINENO:error: \"$BASH_COMMAND\" returned $?" >&2' ERR
  23
  24 if [[ $EUID != 0 ]]; then
  25   echo "$0: error: need to run as root" >&2
  26   exit 1
  27 fi
  28
  29 # for calling outside of FAI without args:
  30 #
  31 # source /b/fai/fai-wrapper
  32 # - set any appropriate classes with: fai-setclass OPT1... which sets CLASS_OPT1=true...
  33 #   or run eval-fai-classfile FILE.
  34 # - Set a VOL_DISTROVER (if not doing mkroot2) eg:
  35 #   fai-setclass VOL_NABIA
  36 #
  37 # ARGS (only 1 is valid):
  38 #
  39 # mkroot2: for running outside of fai and setting up the root2/boot2 luks and btrfs and tab files
  40 #
  41 # mkroot2tab: for running outside of fai and setting up the root2/boot2 tab files, in case luks and btrfs
  42 # happen to already be setup.
  43 #
  44 # mktab: for running outside of fai and generating a crypttab for
  45 # the main root fs in /tmp/fai. Must run with env var, eg export DISTRO=trisquelaramo.
  46 #
  47 # Example use in a bootstrap distro:
  48 # scp /a/bin/fai/fai/config/{distro-install-common/devbyid,hooks/partition.DEFAULT} root@HOST:
  49 # sl HOST
  50 # export DISTRO=trisquelnabia; ./partition.DEFAULT mktab
  51 # ## cryptsetup wont take within a pipeline
  52 # mapfile -t lines < <(awk '! /swap/ {print $2,$1}' /tmp/fai/crypttab )
  53 # for l in "${lines[@]}"; do cryptsetup luksOpen $l; done
  54 #
  55 # # or alternatively, to avoid typing it many times:
  56 # read -r lukspw; for l in "${lines[@]}"; do yes "$lukspw" | cryptsetup luksOpen $l; done
  57
  58 ## potentially useful later:
  59 # sed 's#/root/keyscript,#decrypt_keyctl,#;s/$/,noauto/' /tmp/fai/crypttab >/etc/crypttab
  60 #
  61 # environment variables:
  62 #
  63 # HOSTNAME: if demohost, we set the luks password to just
  64 # 'x'. Used in various other ways too.
  65 #
  66 # SPECIAL_DISK: For use outside of fai. A base disk name like
  67 # /dev/sdk. If set, we just cryptsetup and partition this one disk then
  68 # exit. This is useful for partitioning a disk in preparation to replace
  69 # a failed or failing disk from a raid10 array.
  70 #
  71 # classes:
  72 #
  73 # REPARTITION: forces repartitioning even if we detect the proper amount
  74 # of partitions already exist.
  75 #
  76 # NOWIPE: use existing subvolumes if they exist
  77 #
  78 # REROOTFS: Don't reuse the root filesystem, even if we normally would
  79 #
  80 # ROTATIONAL: forces to install onto hdds instead of sdds. normally sdds
  81 # are chosen if they exist.
  82 #
  83 # PARTITION_PROMPT: command line prompt before partitioning. This is good
  84 # to set if we don't expect repartitioning to happen.
  85 #
  86 # ROTATIONAL: in a system with ssd and hdd, install to the hdd
  87 # instead of the default ssd.
  88 #
  89 # RAID0: forces raid0 filesystem. Normally with 4+ devices, we use
  90 # raid10.
  91 # RAID1: forces raid1 filesystem.
  92 # RAID1c3: forces raid1c3 filesystem (btrfs raid 1, 3 copies).
  93
  94 mkroot2tab=false
  95 mkroot2=false
  96 mktab=false
  97 if [[ $1 ]]; then
  98   ## duplicates fai-wrapper, for convenience of not needing it
  99   if ! type -t ifclass &>/dev/null; then
 100     ifclass() {
 101       local var=${1/#/CLASS_}
 102       [[ $HOSTNAME == "$1" || ${!var} ]]
 103     }
 104   fi
 105
 106   case $1 in
 107     mkroot2)
 108       mkroot2=true
 109       ;;
 110     mkroot2tab)
 111       mkroot2tab=true
 112       ;;
 113     mktab)
 114       mktab=true
 115       ;;
 116     *)
 117       echo "$0: error: unsupported arg: $1" >&2
 118       exit 1
 119       ;;
 120   esac
 121 fi
 122
 123 if [[ ! $SPECIAL_DISK ]] && ! $mkroot2 && ! $mkroot2tab && ! $mktab \
 124      && ! ifclass IANK && ! ifclass FSF; then
 125   echo $0: error: need class IANK or FSF or SPECIAL_DISK for running in fai
 126 fi
 127
 128
 129 if [[ $SPECIAL_DISK ]]; then
 130   export CLASS_REPARTITION=true
 131 fi
 132
 133 # # fai's setup-storage won't do btrfs on luks,
 134 # # so we do it ourself :)
 135 # inspiration taken from files in fai-setup-storage package
 136
 137 # if we are not running in fai, skiptask won't be defined, so carry on.
 138 skiptask partition || ! type skiptask
 139
 140 if ! type -p devbyid; then
 141   for d in $FAI/distro-install-common \
 142              /a/bin/fai/fai/config/distro-install-common $FAI $PWD; do
 143     [[ -d $d ]] || continue
 144     if [[ -e $d/devbyid ]]; then
 145       devbyid=$d/devbyid
 146       devbyid() { $devbyid "$@"; }
 147       break
 148     fi
 149   done
 150   if [[ ! $devbyid ]]; then
 151     echo "$0: error: failed to find devbyid script" >&2
 152     exit 1
 153   fi
 154 fi
 155
 156 #### begin configuration
 157
 158 # this is the ordering of the /dev/sdaX, but
 159 # the ordering of the partition layout goes like this:
 160 # bios_grub
 161 # grub_ext
 162 # efi
 163 # lvm
 164 #  root
 165 #  swap
 166 #  boot
 167
 168
 169 ##### end configuration
 170
 171 ##### begin function defs
 172
 173 bpart() { # btrfs a partition
 174   case $raid_level in
 175     0) mkfs.btrfs -f $@ ;;
 176     *) mkfs.btrfs -f -m raid$raid_level -d raid$raid_level $@ ;;
 177   esac
 178 }
 179
 180
 181 zilap() {
 182   case $HOSTNAME in
 183     sy|bo|so)
 184       return 0
 185       ;;
 186   esac
 187   return 1
 188 }
 189
 190 getluks() {
 191   if [[ ! $luks_dir ]]; then
 192     # see README for docs about how to create these
 193     luks_dir=$FAI/distro-install-common/luks
 194     if [[ ! -d $luks_dir ]]; then
 195       luks_dir=/q/root/luks
 196     fi
 197     if [[ ! -d $luks_dir ]]; then
 198       echo "$0: error: no luks_dir found" >&2
 199       exit 1
 200     fi
 201   fi
 202
 203   luks_file=$luks_dir/host-$HOSTNAME
 204   if [[ ! -e $luks_file ]]; then
 205     # shellcheck disable=SC2206 # globbing is intended
 206     hostkeys=($luks_dir/host-*)
 207     # if there is only one key, we might be deploying somewhere
 208     # where dhcp doesnt give us a proper hostname, so use that.
 209     if [[ ${#hostkeys[@]} == 1 && -e ${hostkeys[0]} ]]; then
 210       luks_file=${hostkeys[0]}
 211     else
 212       echo "$0: error: no key for hostname at $luks_file" >&2
 213       exit 1
 214     fi
 215   fi
 216
 217   # # note, corresponding changes in /b/ds/keyscript-{on,off}
 218   if ifclass demohost; then
 219     lukspw=x
 220   elif [[ -e $luks_dir/$HOSTNAME ]]; then
 221     lukspw=$(cat $luks_dir/$HOSTNAME)
 222   else
 223     lukspw=$(cat $luks_dir/iank)
 224   fi
 225
 226   if $mkroot2; then
 227     luks_file=$luks_dir/host-amy
 228     lukspw=$(cat $luks_dir/amy)
 229   fi
 230 }
 231
 232
 233 fsf() {
 234   ifclass FSF
 235 }
 236
 237
 238 dev-mib() {
 239   local d=${1:-$dev}
 240   echo $(( $(parted -m $d unit MiB print | \
 241                sed -nr "s#^/dev/[^:]+:([0-9]+).*#\1#p") - 1))
 242 }
 243
 244 luks-setup() {
 245   local luksdev="$1"
 246   local cryptname="$2"
 247   # when we move to newer than trisquel 9, we can remove
 248   # --type luks1. We can also check on cryptsetup --help | less /compil
 249   # to see about the other settings. Default in debian 9 is luks2.
 250   # You can convert from luks2 to luks 1 by adding a temporary key:
 251   # cryptsetup luksAddKey --pbkdf pbkdf2
 252   # then remove the new format keys with cryptsetup luksRemoveKey
 253   # then cryptsetup convert DEV --type luks1, then readd old keys and remove temp.
 254   yes YES | cryptsetup luksFormat $luksdev $luks_file || [[ $? == 141 ]]
 255   yes "$lukspw" | \
 256     cryptsetup luksAddKey --key-file $luks_file \
 257                $luksdev || [[ $? == 141 ]]
 258   # background: Keyfile and password are treated just
 259   # like 2 ways to input a passphrase, so we don't actually need to have
 260   # different contents of keyfile and passphrase, but it makes some
 261   # security sense to a really big randomly generated passphrase
 262   # as much as possible, so we have both.
 263   #
 264   # This would remove the keyfile.
 265   #    yes 'test' | cryptsetup luksRemoveKey /dev/... \
 266     #                            /key/file || [[ $? == 141 ]]
 267   cryptsetup luksOpen $luksdev $cryptname --key-file $luks_file
 268 }
 269
 270 mktab() {
 271   mkdir -p /tmp/fai
 272   dev=${boot_devs[0]}
 273   fstabstd="x-systemd.device-timeout=30s,x-systemd.mount-timeout=30s"
 274
 275   if [[ $DISTRO == debianbullseye_bootstrap ]]; then
 276     cat > /tmp/fai/fstab <<EOF
 277 $first_boot_dev  /  btrfs  noatime,subvol=$boot_vol  0 0
 278 $first_efi  /boot/efi  vfat          nofail,$fstabstd  0 0
 279 EOF
 280     cat >/tmp/fai/disk_var.sh <<EOF
 281 BOOT_DEVICE="${short_devs[@]}"
 282 ROOT_PARTITION=$first_boot_dev
 283 EOF
 284   else
 285     # note, fai creates the mountpoints listed here
 286     cat > /tmp/fai/fstab <<EOF
 287 $first_root_dev  /  btrfs                   $fstabstd,noatime,subvol=root_$DISTRO$mopts  0 0
 288 $first_root_dev  /mnt/root  btrfs    nofail,$fstabstd,noatime,subvolid=0$mopts  0 0
 289 $first_boot_dev  /boot  btrfs        nofail,$fstabstd,noatime,subvol=$boot_vol  0 0
 290 $first_efi  /boot/efi  vfat          nofail,$fstabstd  0 0
 291 $first_boot_dev  /mnt/boot  btrfs    nofail,$fstabstd,noatime,subvolid=0  0 0
 292 EOF
 293     if ! fsf; then
 294       cat >> /tmp/fai/fstab <<EOF
 295 /dev/mapper/crypt-${vgs[0]}-o  /mnt/o  btrfs       nofail,$fstabstd,noatime,subvolid=0$mopts  0 0
 296 EOF
 297     fi
 298     rm -f /tmp/fai/crypttab
 299     for vg in ${vgs[@]}; do
 300       if ! fsf; then
 301         cat >>/tmp/fai/crypttab <<EOF
 302 crypt-$vg-root /dev/$vg/root  none  keyscript=/root/keyscript,discard,luks,initramfs
 303 crypt-$vg-o    /dev/$vg/o     none  keyscript=/root/keyscript,discard,luks,initramfs
 304 crypt-$vg-swap /dev/$vg/swap  /dev/urandom  swap,cipher=aes-xts-plain64,size=256,hash=ripemd160
 305 EOF
 306       fi
 307       if fsf; then
 308         cat >> /tmp/fai/fstab <<EOF
 309 /dev/$vg/swap  none  swap  nofail,$fstabstd,sw  0 0
 310 EOF
 311       else
 312         cat >> /tmp/fai/fstab <<EOF
 313 /dev/mapper/crypt-$vg-swap  none  swap  nofail,$fstabstd,sw  0 0
 314 EOF
 315       fi
 316     done
 317
 318     # fai would do this:
 319     #BOOT_DEVICE=\${BOOT_DEVICE:-"${devs[0]}"}
 320
 321     # note: swaplist seems to do nothing.
 322     cat >/tmp/fai/disk_var.sh <<EOF
 323 BOOT_DEVICE="${short_devs[@]}"
 324 BOOT_PARTITION=\${BOOT_PARTITION:-$first_boot_dev}
 325 # ROOT_PARTITIONS is added by me, used in arch setup.
 326 ROOT_PARTITIONS="${root_devs[@]}"
 327 ROOT_PARTITION=\${ROOT_PARTITION:-$first_root_dev}
 328 SWAPLIST=\${SWAPLIST:-"${swap_devs[@]}"}
 329 EOF
 330
 331     if [[ $HOSTNAME == frodo ]]; then
 332       big_disks=(
 333         ata-Hitachi_HDS722020ALA330_JK1121YAG7SXWS-part1
 334         ata-Hitachi_HDS722020ALA330_JK1121YAG7SY4S-part1
 335         ata-Hitachi_HDS723030ALA640_MK0311YHG2WUSA-part1
 336         ata-ST4000DM000-1F2168_Z300AZ6K-part1
 337         ata-ST6000DM001-1XY17Z_Z4D2WMZK-part1
 338         ata-TOSHIBA_MD04ACA500_8539K4TQFS9A-part1
 339         ata-TOSHIBA_MD04ACA500_85NAK4T2FS9A-part1
 340         ata-TOSHIBA_MD04ACA500_9551K615FS9A-part1
 341         ata-TOSHIBA_MD04ACA500_Y5IFK6IJFS9A-part1
 342       )
 343       for d in ${big_disks[@]}; do
 344         cat >>/tmp/fai/crypttab <<EOF
 345 crypt_dev_$d  /dev/disk/by-id/$d  /mnt/root/q/root/luks/iank  discard,luks
 346 EOF
 347       done
 348       cat >> /tmp/fai/fstab <<EOF
 349 /dev/mapper/crypt_dev_${big_disks[0]}  /mnt/i  btrfs  nofail,$fstabstd,noatime,subvolid=0  0 0
 350 EOF
 351     fi
 352     if [[ $HOSTNAME == kd ]]; then
 353       # note, having these with keyscript and initramfs causes a luks error in fai.log,
 354       # but it is safely ignorable and gets us the ability to just type our password
 355       # in once at boot. A downside is that they are probably needed to be plugged in to boot.
 356       cat >>/tmp/fai/crypttab <<EOF
 357 crypt_dev_ata-Samsung_SSD_870_QVO_8TB_S5VUNG0N900656V${even_bigsuf} /dev/disk/by-id/ata-Samsung_SSD_870_QVO_8TB_S5VUNG0N900656V${even_bigsuf}  /mnt/root/q/root/luks/iank  discard,luks
 358 crypt_dev_ata-TOSHIBA_MD04ACA500_84R2K773FS9A-part1 /dev/disk/by-id/ata-TOSHIBA_MD04ACA500_84R2K773FS9A-part1  /mnt/root/q/root/luks/iank  discard,luks
 359 crypt_dev_ata-ST6000DM001-1XY17Z_Z4D29EBL-part1 /dev/disk/by-id/ata-ST6000DM001-1XY17Z_Z4D29EBL-part1  /mnt/root/q/root/luks/iank  discard,luks
 360 EOF
 361       cat >> /tmp/fai/fstab <<EOF
 362 # r7 = root partition7. it isnt actually #7 anymore, not a great name, but whatever
 363 /dev/mapper/crypt_dev_ata-Samsung_SSD_870_QVO_8TB_S5VUNG0N900656V${even_bigsuf}  /mnt/r7  btrfs  nofail,$fstabstd,noatime,compress=zstd,subvolid=0  0 0
 364 /dev/mapper/crypt_dev_ata-TOSHIBA_MD04ACA500_84R2K773FS9A-part1  /mnt/rust1  btrfs  nofail,$fstabstd,noatime,compress=zstd,subvolid=0  0 0
 365 /dev/mapper/crypt_dev_ata-ST6000DM001-1XY17Z_Z4D29EBL-part1  /mnt/rust2  btrfs  nofail,$fstabstd,noatime,compress=zstd,subvolid=0  0 0
 366 EOF
 367     fi
 368   fi
 369 }
 370
 371
 372
 373 #### root2 non-fai run
 374
 375 # todo: update for lvm
 376 doroot2() {
 377
 378
 379   # We write to these files instead of just /etc/fstab, /etc/crypttab,
 380   # because these are filesystems created after our current root, and so
 381   # this allows us to update other root filesystems too.
 382   rm -f /mnt/root/root2-{fs,crypt}tab
 383   if $partition; then
 384     echo $0: error: found partition=true but have mkroot2 arg
 385     exit 1
 386   fi
 387
 388
 389   root2_devs=()
 390   for vg in ${vgs[@]}; do
 391
 392
 393     root2_devs+=(/dev/mapper/crypt-$vg-root2)
 394     if $mkroot2; then
 395       lvcreate -y -L $root2_part_mib $vg -n root2
 396       lvcreate -y -L $boot2_part_mib $vg -n boot2
 397       luks-setup /dev/$vg/root2 crypt-$vg-root2
 398     fi
 399     cat >>/mnt/root/root2-crypttab <<EOF
 400 crypt-$vg-root2 /dev/$vg/root2  $luks_file  discard,luks,initramfs
 401 EOF
 402   done
 403   if $mkroot2; then
 404     bpart ${root2_devs[@]}
 405     bpart ${boot2_devs[@]}
 406   fi
 407   mkdir -p /mnt/root2 /mnt/boot2
 408   cat >>/mnt/root/root2-fstab <<EOF
 409 ${root2_devs[0]} /mnt/root2  btrfs  nofail,x-systemd.device-timeout=30s,x-systemd.mount-timeout=30s,noatime,subvolid=0$mopts  0 0
 410 ${boot2_devs[0]} /mnt/boot2  btrfs  nofail,x-systemd.device-timeout=30s,x-systemd.mount-timeout=30s,noatime,subvolid=0  0 0
 411 EOF
 412   exit 0
 413 }
 414
 415
 416 ##### end function defs
 417
 418
 419 ##### begin variable setup
 420
 421
 422 partition=false
 423 if ifclass REPARTITION; then
 424   partition=true # force a full wipe
 425 fi
 426 wipe=true
 427 if ifclass NOWIPE; then
 428   wipe=false
 429 fi
 430
 431 rerootfs=false
 432 if ifclass REROOTFS; then
 433   rerootfs=true
 434 fi
 435
 436 if (($(nproc) > 2)); then
 437   mopts=,compress=zstd
 438 fi
 439
 440 declare -A disk_excludes
 441 if ! $mkroot2 && ! $mkroot2tab && ! $mktab ! ifclass USE_MOUNTED; then
 442   ## ignore disks that are mounted, eg when running from fai-cd
 443   while read -r l; do
 444     eval "$l"
 445     if [[ ! $PKNAME ]]; then
 446       # shellcheck disable=SC2153 # not a misspelling
 447       PKNAME="$KNAME"
 448     fi
 449     if [[ $MOUNTPOINT ]]; then
 450       disk_excludes[$PKNAME]=true
 451     fi
 452   done < <(lsblk -nP -o KNAME,MOUNTPOINT,PKNAME)
 453 fi
 454
 455 hdds=()
 456 ssds=()
 457 # this excludes "usb". vda disk has empty tran (transport). This may need adjustment
 458 # for some new type we come across. cdrom has type "rom"
 459 for disk in $(lsblk -ndo name,type,tran | awk '$3 ~ "^(sata|nvme|)$" && $2 == "disk" { print $1 }'); do
 460   if [[ ${disk_excludes[$disk]} ]]; then
 461     continue
 462   fi
 463   case $disk in
 464     # cdrom
 465     sr*) continue ;;
 466   esac
 467   case $(cat /sys/block/$disk/queue/rotational) in
 468     0)
 469       ssds+=("/dev/$disk")
 470       echo $0: found ssd /dev/$disk
 471       ;;
 472     1)
 473       hdds+=("/dev/$disk")
 474       echo $0: found hdd /dev/$disk
 475       ;;
 476     *) echo "$0: error: unknown /sys/block/$disk/queue/rotational: \
 477 $(cat $disk/queue/rotational)"; exit 1 ;;
 478   esac
 479 done
 480
 481 # install all ssds, or if there are none, all hdds.
 482 # Note, usb flash disks are seen as rotational, which is
 483 # very odd, but convenient for ignoring them here.
 484 if ! ifclass ROTATIONAL && (( ${#ssds[@]} > 0 )); then
 485   read -ra short_devs<<<"${ssds[@]}"
 486 else
 487   read -ra short_devs<<<"${hdds[@]}"
 488 fi
 489
 490 pvn=1
 491
 492 bootn=2
 493
 494 # rootn=1
 495 # root2n=2
 496 # swapn=3
 497 # bootn=4
 498 # boot2n=5
 499
 500 efin=3
 501 # ext partition so grub can write persistent variables,
 502 # so it can do a one time boot. grub can't write to
 503 # btrfs or any cow fs because it's more
 504 # more complicated to do and they don't want to.
 505 grub_extn=4
 506 # bios boot partition,
 507 # https://wiki.archlinux.org/index.php/GRUB
 508 bios_grubn=5
 509 # for an even raid (raid 1), when one disk is bigger, this partition goes on the big disk
 510 even_bign=6
 511 # even_bign only exists in some cases
 512 lastn=$bios_grubn
 513 # check if the partitions exist have the right filesystems
 514 #blkid="$(blkid -s TYPE)"
 515 for dev in ${short_devs[@]}; do
 516   if $partition; then break; fi
 517   y=$(readlink -f $dev)
 518   # shellcheck disable=SC2206 # globbing is intended
 519   arr=($y?*)
 520   if (( ${#arr[@]} < lastn )); then
 521     partition=true
 522   fi
 523   # On one system, blkid is missing some partitions.
 524   # maybe we need a flag, like FUZZY_BLKID or something, so we
 525   # can check that at least some exist.
 526   # for x in "`rootdev`: TYPE=\"crypto_LUKS\"" "`bootdev`: TYPE=\"btrfs\""; do
 527   #     echo "$blkid" | grep -Fx "$x" &>/dev/null || partition=true
 528   # done
 529 done
 530
 531 if $partition && ifclass PARTITION_PROMPT; then
 532   echo "Press any key except ctrl-c to continue and partition these drives:"
 533   if [[ $SPECIAL_DISK ]]; then
 534     echo $SPECIAL_DISK
 535   else
 536     echo "  ${short_devs[*]}"
 537   fi
 538   read -r
 539 fi
 540
 541 devs=()
 542 vgs=()
 543 root_devs=()
 544 o_devs=()
 545 swap_devs=()
 546 shopt -s extglob
 547 partsuffix=-part
 548 for short_dev in ${short_devs[@]}; do
 549   dev="$(devbyid $short_dev)"
 550   if [[ $dev != */by-id/* ]]; then
 551     # no by-id link, assume we are in a vm and this is true for all devs.
 552     partsuffix=
 553   fi
 554   # for vms, cant name a vg the same as the short device name, they
 555   # conflict: /dev/$vg is already taken
 556
 557   dname=${dev##*/}
 558   vg=vg$dname
 559   vg=${vg//:/}
 560   vgs+=("$vg")
 561   devs+=("$dev")
 562   if fsf; then
 563     root_devs+=(/dev/$vg/root)
 564     swap_devs+=(/dev/$vg/swap)
 565   else
 566     o_devs+=(/dev/mapper/crypt-$vg-o)
 567     root_devs+=(/dev/mapper/crypt-$vg-root)
 568     swap_devs+=(/dev/mapper/crypt-$vg-swap)
 569   fi
 570 done
 571 first_root_dev=${root_devs[0]}
 572 if [[ ! $SPECIAL_DISK && ! ${devs[0]} ]]; then
 573   echo "$0: error: failed to detect devs" >&2
 574   exit 1
 575 fi
 576
 577
 578 pvsuf=$partsuffix$pvn
 579 bootsuf=$partsuffix$bootn
 580 efisuf=$partsuffix$efin
 581 grub_extsuf=$partsuffix$grub_extn
 582 # We dont do anything with this partition here, so this
 583 # is be unused, but left as a comment for completing the pattern
 584 # of all the suffixes.
 585 #bios_grubsuf=$partsuffix$bios_grubn
 586 even_bigsuf=$partsuffix$even_bign
 587
 588
 589 boot_space=0
 590 first=true
 591 boot_devs=()
 592 boot2_devs=()
 593 for dev in ${devs[@]}; do
 594   vg=vg${dev##*/}
 595   vg=${vg//:/}
 596   # I ran into a machine (frodo) where the bios doesn't know about some disks,
 597   # so 1st stage of grub also doesn't know about them.
 598   # Also, grub does not support mounting degraded btrfs as far as
 599   # I can tell with some googling.
 600   # From within an arch install env, I could detect them by noting
 601   # their partitions were mixed with the next disk in /dev/disk/by-path,
 602   # and I have mixed model disks, and I could see the 8 models which showed
 603   # up in the bios, and thus see which 2 models were missing.
 604   # hdparm -I /dev/sdh will give model info in linux.
 605   # However, in fai on jessie, /dev/disk/by-path dir doesn't exist,
 606   # and I don't see another way, so I'm hardcoding them.
 607   # We still put grub on them and partition them the same, for uniformity
 608   # and in case they get moved to a system that can recognize them,
 609   # we just exclude them from the boot filesystem.
 610   cd /dev/disk/by-id/
 611   bad_disk=false
 612   for id in ata-TOSHIBA_MD04ACA500_8539K4TQFS9A \
 613               ata-TOSHIBA_MD04ACA500_Y5IFK6IJFS9A; do
 614     if [[ $(readlink -f $id) == "$(readlink -f $dev)" ]]; then
 615       bad_disk=true
 616       break
 617     fi
 618   done
 619   if $bad_disk; then
 620     continue
 621   fi
 622   boot_devs+=($dev$bootsuf)
 623   boot2_devs+=(/dev/$vg/boot2)
 624   boot_space=$(( boot_space + $(parted -m $dev unit MiB print | \
 625                                   sed -nr "s#^/dev/[^:]+:([0-9]+).*#\1#p") - 1))
 626   if $first; then
 627     first_efi=$dev$efisuf
 628     first_grub_extdev=$dev$grub_extsuf
 629     first=false
 630   fi
 631 done
 632 first_boot_dev=${boot_devs[0]}
 633
 634 even_raid=false
 635 if ifclass RAID0 || (( ${#boot_devs[@]} == 1 )); then
 636   raid_level=0
 637   raid_duplication=1
 638 elif ifclass RAID1 || (( ${#boot_devs[@]} == 2 )); then
 639   if (( ${#boot_devs[@]} == 2 )); then
 640     even_raid=true
 641   fi
 642   raid_level=1
 643   raid_duplication=2
 644 elif ifclass RAID1c3 || (( ${#boot_devs[@]} == 3 )); then
 645   raid_level=1c3
 646   raid_duplication=3
 647 else
 648   raid_level=10
 649   raid_duplication=2
 650 fi
 651
 652
 653
 654 ### Begin calculate boot partition space
 655 # due to raid duplication
 656 case $raid_level in
 657   1|10) boot_space=$(( boot_space / 2 )) ;;
 658   1c3) boot_space=$(( boot_space / 3 )) ;;
 659 esac
 660 if fsf; then
 661   boot_mib=6000
 662 elif (( boot_space > 900000 )); then
 663   # this is larger than needed for several /boot subvols,
 664   # becuase I keep a minimal debian install on it for
 665   # recovery needs and for doing pxe-kexec.
 666   boot_mib=10000
 667   root2_mib=500000
 668   boot2_mib=5000
 669 elif (( boot_space > 30000 )); then
 670   boot_mib=$(( 5000 + (boot_space - 30000) / 2 ))
 671   root2_mib=100
 672   boot2_mib=100
 673 else
 674   # Small vms don't have room for /boot recovery.  With 3 kernels
 675   # installed, i'm using 132M on t8, so this seems like plenty of
 676   # room. note: rhel 8 recomments 1g for /boot.  u20.04 with 3 kernels =
 677   # 308 mb, so things have grown significantly
 678   boot_mib=1000
 679   root2_mib=100
 680   boot2_mib=100
 681 fi
 682 boot_part_mib=$(( boot_mib * raid_duplication / ${#boot_devs[@]} ))
 683
 684 if zilap; then
 685   boot2_part_mib=$(( boot2_mib * raid_duplication / ${#boot_devs[@]}  ))
 686   root2_part_mib=$(( root2_mib * raid_duplication / ${#root_devs[@]}  ))
 687 else
 688   boot2_part_mib=0
 689   root2_part_mib=0
 690 fi
 691 ### end calculate boot partition space
 692
 693
 694 if [[ ! $DISTRO ]]; then
 695   if ifclass VOL_BULLSEYE_BOOTSTRAP; then
 696     DISTRO=debianbullseye_bootstrap
 697   elif ifclass VOL_STRETCH; then
 698     DISTRO=debianstretch
 699   elif ifclass VOL_BUSTER; then
 700     DISTRO=debianbuster
 701   elif ifclass VOL_BULLSEYE; then
 702     DISTRO=debianbullseye
 703   elif ifclass VOL_BOOKWORM; then
 704     DISTRO=debianbookworm
 705   elif ifclass VOL_TESTING; then
 706     DISTRO=debiantesting
 707   elif ifclass VOL_XENIAL; then
 708     DISTRO=ubuntuxenial
 709   elif ifclass VOL_BIONIC; then
 710     DISTRO=ubuntubionic
 711   elif ifclass VOL_FOCAL; then
 712     DISTRO=ubuntufocal
 713   elif ifclass VOL_JAMMY; then
 714     DISTRO=ubuntujammy
 715   elif ifclass VOL_FLIDAS; then
 716     DISTRO=trisquelflidas
 717   elif ifclass VOL_ETIONA; then
 718     DISTRO=trisqueletiona
 719   elif ifclass VOL_NABIA; then
 720     DISTRO=trisquelnabia
 721   elif ifclass VOL_ARAMO; then
 722     DISTRO=trisquelaramo
 723   elif $mkroot2 || $mkroot2tab; then
 724     :
 725   else
 726     echo "PARTITIONER ERROR: no distro class/var set" >&2
 727     exit 1
 728   fi
 729 fi
 730
 731 if [[ $DISTRO == debianbullseye_bootstrap ]]; then
 732   # this is just convenience for the libreboot_grub config
 733   # so we can glob the other ones easier.
 734   boot_vol=$DISTRO
 735 else
 736   boot_vol=boot_$DISTRO
 737 fi
 738
 739
 740 # 1.5 x based on https://access.redhat.com/documentation/en-US/Red_Hat_Enterprise_Linux/7/html/Installation_Guide/sect-disk-partitioning-setup-x86.html#sect-custom-partitioning-x86
 741 swap_mib=$(( $(grep ^MemTotal: /proc/meminfo | \
 742                  awk '{print $2}') * 3/(${#devs[@]} * 2 ) / 1024 ))
 743
 744 shopt -s nullglob
 745 ##### end variable setup
 746
 747
 748
 749
 750 if $mkroot2 || $mkroot2tab; then
 751   getluks
 752   doroot2
 753 elif $mktab; then
 754   mktab
 755   exit 0
 756 else
 757   mktab
 758   if ! fsf; then
 759     getluks
 760   fi
 761 fi
 762
 763
 764 if $partition; then
 765   ### begin wipefs
 766   if [[ ! $SPECIAL_DISK ]]; then
 767
 768     # we do lvm removals just for the disks we are using
 769     pv_wipes=()
 770     vg_wipes=()
 771     declare -A vg_map
 772     pv_devs=$(pvs --noheadings -o pvname)
 773     for pv_dev in $pv_devs; do
 774       for short_dev in ${short_devs[@]}; do
 775         if [[ $pv_dev == $short_dev* ]]; then
 776           pv_wipes+=($pv_dev)
 777           vgs_of_pv=$(pvs --noheadings -o vgname $pv_dev)
 778           for vg in $vgs_of_pv; do
 779             if [[ ${vg_map[$vg]} ]]; then
 780               continue
 781             fi
 782             vg_map[$vg]=t
 783             vg_wipes+=($vg)
 784             lvs=$(vgs --noheadings -o lv_path $vg)
 785             for lv in $lvs; do
 786               wipefs -a $lv
 787             done
 788           done
 789         fi
 790       done
 791     done
 792
 793     for vg in ${vg_wipes[@]}; do
 794       vgchange -an $vg
 795       vgremove -ff $vg
 796     done
 797
 798     for pv in ${pv_wipes[@]}; do
 799       pvremove -ff $pv
 800     done
 801
 802     for dev in ${devs[@]}; do
 803       # if we repartition to the same as an old partition,
 804       # we don't want any old fses hanging around.
 805       count_down=10
 806       # wipefs has failed, manual run works, google suggests timing issue
 807       while ! wipefs -a $dev; do
 808         sleep 2
 809         count_down=$((count_down - 1))
 810         if (( count_down <= 0 )); then
 811           echo "$0: wipefs failed 10 times. exiting" >&2
 812           exit 1
 813         fi
 814       done
 815     done
 816   fi
 817   ### end wipefs
 818
 819
 820   # When we have 2 disks of at least 100g difference in size,
 821   # make an extra partition on the end of the bigger one.
 822   even_big_part=false
 823   even_diff_min=100000
 824   if $even_raid; then
 825     smalli=0
 826     bigi=1
 827     if (( $(dev-mib ${devs[0]}) >= $(dev-mib ${devs[1]}) )); then
 828       smalli=1
 829       bigi=0
 830     fi
 831     disk_mib=$(dev-mib ${devs[smalli]})
 832     even_big_dev=${devs[bigi]}
 833     even_big_mib=$(dev-mib $even_big_dev)
 834     if (( even_big_mib - disk_mib > even_diff_min )); then
 835       even_big_part=true
 836     fi
 837   fi
 838
 839   if [[ $SPECIAL_DISK ]]; then
 840     devs=($(devbyid $SPECIAL_DISK))
 841   fi
 842   for dev in ${devs[@]}; do
 843     vg=vg${dev##*/}
 844     vg=${vg//:/}
 845
 846     # parted will round up the disk size. Do -1 so we can have
 847     # fully 1MiB unit partitions for easy resizing of the last partition.
 848     # Otherwise we would pass in -0 for the end argument for the last partition.
 849     #
 850     # Note: parted print error output is expected. example:
 851     # Error: /dev/vda: unrecognised disk label
 852     if ! $even_raid; then
 853       disk_mib=$(dev-mib)
 854     fi
 855
 856
 857     parted -s $dev mklabel gpt
 858     # MiB because parted complains about alignment otherwise.
 859     pcmd="parted -a optimal -s -- $dev"
 860     # main lvm partition
 861
 862     pv_end=$(( disk_mib - boot_part_mib ))
 863     $pcmd mkpart primary ext3 524MiB ${pv_end}MiB
 864     $pcmd name $pvn pv
 865
 866     #   + 794 pvcreate -y /dev/disk/by-id/ata-ST4000DM000-1F2168_Z3028BKA-part1
 867     # WARNING: Device /dev/sde1 not initialized in udev database even after waiting 10000000 microseconds.
 868     # No device found for /dev/disk/by-id/ata-ST4000DM000-1F2168_Z3028BKA-part1.
 869     # sleep 10 was not enough.
 870     secs=0
 871     while [[ ! -e $dev$pvsuf ]] && (( secs < 40 )); do
 872       sleep 1
 873       secs=$((secs +1))
 874     done
 875     sleep 3
 876     pvcreate -y -ff $dev$pvsuf
 877     vgcreate -y -ff $vg $dev$pvsuf
 878
 879     if fsf; then
 880       root_mib=40000
 881     else
 882       # This would maximize it, but we are going for a separate filesystem in /o,
 883       # so use fixed sizes to allow both to grow
 884       # 600 = uefi 512 + grubext 8 + bios grub 3 + some extra cuz this is lvm
 885       #root_mib=$(( disk_mib - root2_part_mib - swap_mib - boot_part_mib - boot2_part_mib - 600 ))
 886       o_mib=$(( 120 * 1000 ))
 887       # max minus o, minus a gig just for some extra space
 888       max_root_mib=$(( disk_mib - root2_part_mib - swap_mib - boot_part_mib - boot2_part_mib - 600 - o_mib - 1000 ))
 889       root_mib=$(( 1000 * 1000 )) # * 1000 to make it in gb.
 890       if (( max_root_mib < root_mib )); then
 891         root_mib=$max_root_mib
 892       fi
 893     fi
 894
 895     if [[ $SPECIAL_DISK ]]; then
 896       lvcreate -y -L $max_root_mib $vg -n data
 897     else
 898       # -L unit default mebibyte
 899       lvcreate -y -L $root_mib $vg -n root
 900       if ! fsf; then
 901         lvcreate -y -L $o_mib $vg -n o
 902       fi
 903       lvcreate -y -L $swap_mib $vg -n swap
 904       # unencrypted swap needs mkswap
 905       if fsf; then
 906         mkswap /dev/$vg/swap
 907       fi
 908     fi
 909
 910     $pcmd mkpart primary "" ${pv_end}MiB ${disk_mib}MiB
 911     $pcmd name $bootn boot
 912     $pcmd set $bootn boot on
 913
 914     # uefi partition, for normal bios systems, its just in case.
 915     $pcmd mkpart primary "fat32" 12MiB 524MiB
 916     $pcmd name $efin efi
 917     # note, this is shown here: https://support.system76.com/articles/bootloader/
 918     # but not mentioned https://wiki.archlinux.org/index.php/EFI_system_partition
 919     # might not be needed
 920     $pcmd set $efin esp on
 921
 922     # i only need a few k, but googling min size,
 923     # I found someone saying that gparted required
 924     # required at least 8 because of their hard drive cylinder size.
 925     # And 8 is still very tiny.
 926     # grub_ext partition
 927     $pcmd mkpart primary "ext2" 4MiB 12MiB
 928     $pcmd name $grub_extn grubext
 929     # gpt ubuntu cloud image uses ~4 mb for this partition. fai uses 1 MiB.
 930     # so, I use 3, whatever.
 931     # note: parted manual saying cheap flash media
 932     # should to start at 4.
 933     # biols grub partition
 934     $pcmd mkpart primary "" 1MiB 4MiB
 935     $pcmd name $bios_grubn biosgrub
 936     $pcmd set $bios_grubn bios_grub on
 937     if $even_big_part  && [[ $dev == "$even_big_dev" ]]; then
 938       $pcmd mkpart primary ext3 ${disk_mib}MiB ${even_big_mib}MiB
 939       $pcmd name $even_bign even_big
 940     fi
 941
 942     # the mkfs failed before on a vm, which prompted me to add
 943     # sleep .1
 944     # then it failed again on a physical machine
 945     # with:
 946     # Device /dev/disk/by-id/foo doesn't exist or access denied,
 947     # so I added a wait until it existed.
 948     # Then I added the mkfs.ext2, which claimed to succeed,
 949     # but then couldn't be found upon reboot. In that case we didn't
 950     # wait at all. So I've added a 3 second minimum wait.
 951     secs=0
 952     while [[ ! -e $dev$efisuf ]] && (( secs < 40 )); do
 953       sleep 1
 954       secs=$((secs +1))
 955     done
 956     sleep 3
 957
 958     mkfs.fat -F32 $dev$efisuf
 959
 960     if ! fsf && $even_big_part  && [[ $dev == "$even_big_dev" ]]; then
 961       luks-setup $even_big_dev$even_bigsuf ${even_big_dev##*/}$even_bigsuf
 962       mkfs.btrfs -f /dev/mapper/${even_big_dev##*/}$even_bigsuf
 963     fi
 964
 965     # Holds just a single file, rarely written, so
 966     # use ext2, like was often used for the /boot partition.
 967     # This exists because grub can only persist data to a non-cow fs.
 968     # And we use persisting a var in grub to do a one time boot.
 969     # We could pass the data on the kernel command line and persist it
 970     # to grubenv after booting, but that relies on the boot always succeeding.
 971     # This is just a bit more robust, and it could work for booting
 972     # into ipxe which can't persist data, if we ever got that working.
 973     mkfs.ext2 $dev$grub_extsuf
 974
 975     if [[ $SPECIAL_DISK ]]; then
 976       luks-setup /dev/$vg/data crypt-$vg-data
 977       exit 0
 978     fi
 979
 980     # for fsf, no encryption of root because root will not contain any
 981     # sensitive data.
 982     if ! fsf; then
 983       luks-setup /dev/$vg/root crypt-$vg-root
 984       luks-setup /dev/$vg/o crypt-$vg-o
 985     fi
 986
 987   done
 988   ls -la /dev/btrfs-control # this was probably for debugging...
 989   sleep 1
 990
 991   bpart ${root_devs[@]}
 992   if ! fsf; then
 993     bpart ${o_devs[@]}
 994   fi
 995   bpart ${boot_devs[@]}
 996
 997 else ## end if $partition ##
 998
 999   if ! fsf; then
1000     for vg in ${vgs[@]}; do
1001       if [[ -e /dev/mapper/crypt-$vg-root ]]; then
1002         continue
1003       fi
1004       if $rerootfs; then
1005         luks-setup /dev/$vg/root crypt-$vg-root
1006       else
1007         cryptsetup luksOpen /dev/$vg/root crypt-$vg-root \
1008                    --key-file $luks_file
1009         cryptsetup luksOpen /dev/$vg/o crypt-$vg-o \
1010                    --key-file $luks_file
1011       fi
1012     done
1013   fi
1014
1015   if $rerootfs; then
1016     sleep 1
1017     bpart ${root_devs[@]}
1018   fi
1019   sleep 1
1020 fi
1021
1022
1023 if $wipe && [[ $DISTRO != debianbullseye_bootstrap ]]; then
1024   # bootstrap distro doesn't use separate encrypted root.
1025   mount -o subvolid=0 ${root_devs[0]} /mnt
1026   # systemd creates subvolumes we want to delete.
1027   mapfile -t s < <(btrfs subvolume list --sort=-path /mnt |
1028                      sed -rn "s#^.*path\s*(root_$DISTRO/\S+)\s*\$#\1#p")
1029   for subvol in ${s[@]}; do btrfs subvolume delete /mnt/$subvol; done
1030   btrfs subvolume set-default 0 /mnt
1031   [[ ! -e /mnt/root_$DISTRO ]] || btrfs subvolume delete /mnt/root_$DISTRO
1032
1033   ## create subvols ##
1034   cd /mnt
1035
1036   btrfs subvolume create root_$DISTRO
1037
1038   # could set default subvol like this, but no reason to.
1039   # btrfs subvolume set-default \
1040     #       $(btrfs subvolume list . | grep "root_$DISTRO$" | awk '{print $2}') .
1041
1042   # For raid systems, cow allows for error correction, for non-raid systems,
1043   # protects root fs from having the plug pulled. Reprovisioning a root
1044   # subvol is not my favorite thing to do.
1045   # # no cow on the root filesystem. it's setup is fully scripted,
1046   # # if it's messed up, we will just recreated it,
1047   # # and we can get better perf with this.
1048   # # I can't remember exactly why, but this is preferable to mounting with
1049   # # -o nodatacow, I think because subvolumes inherit that.
1050   # chattr -Rf +C root_$DISTRO
1051   cd /
1052   umount /mnt
1053 fi
1054
1055 mount -o subvolid=0 $first_boot_dev /mnt
1056 cd /mnt
1057 btrfs subvolume set-default 0 /mnt # already default, just ensuring it.
1058
1059 # for libreboot systems. grub2 only reads from subvolid=0
1060 mkdir -p /mnt/grub2
1061 # todo: this would need some rework if we moved boot into
1062 # lvm.
1063 cp $FAI/distro-install-common/libreboot_grub.cfg /mnt/grub2
1064
1065 if $wipe && [[ -e /mnt/$boot_vol ]]; then
1066   btrfs subvolume delete /mnt/$boot_vol
1067 fi
1068 if [[ ! -e /mnt/$boot_vol ]]; then
1069   btrfs subvolume create $boot_vol
1070 fi
1071 cd /
1072 umount /mnt
1073 ## end create subvols ##
1074
1075 mount $first_grub_extdev /mnt
1076 grub-editenv /mnt/grubenv set did_fai_check=true
1077 grub-editenv /mnt/grubenv set last_boot=/$boot_vol
1078 umount /mnt
1079
1080
1081 # initial setup of extra data fs, mounted,
1082 # btrfs subvol create nocow
1083 # chattr +C nocow
1084 # chown iank.iank nocow
1085