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