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 PS4='+ $LINENO '
  19 set -eE -o pipefail
  20 trap 'echo "$0:$LINENO:error: \"$BASH_COMMAND\" returned $?" >&2' ERR
  21
  22 if [[ $EUID != 0 ]]; then
  23   echo "$0: error: need to run as root" >&2
  24   exit 1
  25 fi
  26
  27 # for calling outside of FAI:
  28 # fai-redep
  29 #
  30 # source /b/fai/fai-wrapper
  31 # - set any appropriate classes with: fai-setclass OPT1... which sets CLASS_OPT1=true...
  32 #   or run eval-fai-classfile FILE.
  33 # - Set a VOL_DISTROVER (if not doing mkroot2) eg:
  34 #   fai-setclass VOL_NABIA
  35 #
  36 # OPTIONS:
  37 #
  38 # mkroot2: for running outside of fai and setting up the root2/boot2 luks and btrfs and tab files
  39 #
  40 # mkroot2tab: for running outside of fai and setting up the root2/boot2 tab files, in case luks and btrfs
  41 # happen to already be setup.
  42 #
  43 #
  44 # environment variables:
  45 #
  46 # HOSTNAME: if frodo, we exclude 2 devices from the /boot array, which
  47 # the bios does not see. if demohost, we set the luks password to just
  48 # 'x'.
  49 #
  50 # SPECIAL_DISK: For use outside of fai. A base disk name like
  51 # /dev/sdk. If set, we just cryptsetup and partition this one disk then
  52 # exit. This is useful for partitioning a disk in preparation to replace
  53 # a failed or failing disk from a raid10 array.
  54 #
  55 # classes:
  56 #
  57 # REPARTITION: forces repartitioning even if we detect the proper amount
  58 # of partitions already exist.
  59 #
  60 # NOWIPE: use existing subvolumes if they exist
  61 #
  62 # ROTATIONAL: forces to install onto hdds instead of sdds. normally sdds
  63 # are chosen if they exist.
  64 #
  65 # PARTITION_PROMPT: command line prompt before partitioning
  66 #
  67 # RAID0: forces raid0 filesystem. Normally with 4+ devices, we use
  68 # raid10.
  69 # RAID1: forces raid1 filesystem.
  70
  71 mkroot2tab=false
  72 mkroot2=false
  73 if [[ $1 ]]; then
  74   case $1 in
  75     mkroot2)
  76       mkroot2=true
  77       ;;
  78     mkroot2tab)
  79       mkroot2tab=true
  80       ;;
  81     *)
  82       echo "$0: error: unsupported arg: $1" >&2
  83       exit 1
  84       ;;
  85   esac
  86 fi
  87
  88
  89 if [[ $SPECIAL_DISK ]]; then
  90   export CLASS_REPARTITION=true
  91 fi
  92
  93 # # fai's setup-storage won't do btrfs on luks,
  94 # # so we do it ourself :)
  95 # inspiration taken from files in fai-setup-storage package
  96
  97 # if we are not running in fai, skiptask won't be defined, so carry on.
  98 skiptask partition || ! type skiptask
  99
 100 if ! type -p devbyid; then
 101   for d in $FAI/distro-install-common \
 102              /a/bin/fai/fai/config/distro-install-common $FAI $PWD; do
 103     [[ -d $d ]] || continue
 104     if [[ -e $d/devbyid ]]; then
 105       devbyid=$d/devbyid
 106       devbyid() { $devbyid "$@"; }
 107       break
 108     fi
 109   done
 110   if [[ ! $devbyid ]]; then
 111     echo "$0: error: failed to find devbyid script" >&2
 112     exit 1
 113   fi
 114 fi
 115
 116
 117
 118 #### begin configuration
 119
 120 # this is the ordering of the /dev/sdaX, but
 121 # the ordering of the partition layout goes like this:
 122 # bios_grub
 123 # grub_ext
 124 # efi
 125 # root
 126 # swap
 127 # boot
 128
 129 rootn=1
 130 root2n=2
 131 swapn=3
 132 bootn=4
 133 boot2n=5
 134 efin=6
 135 # ext partition so grub can write persistent variables,
 136 # so it can do a one time boot. grub can't write to
 137 # btrfs or any cow fs because it's more
 138 # more complicated to do and they don't want to.
 139 grub_extn=7
 140 # bios boot partition,
 141 # https://wiki.archlinux.org/index.php/GRUB
 142 bios_grubn=8
 143 even_bign=9
 144 lastn=$bios_grubn
 145
 146
 147
 148 ##### end configuration
 149
 150
 151 add-part() { # add partition suffix to $dev
 152   local d part
 153   if [[ $# == 1 ]]; then
 154     d=$dev
 155     part=$1
 156   else
 157     d=$1
 158     part=$2
 159   fi
 160   echo $d-part$part
 161 }
 162
 163 rootdev() { add-part $@ $rootn; }
 164 root2dev() { add-part $@ $root2n; }
 165 swapdev() { add-part $@ $swapn; }
 166 bootdev() { add-part $@ $bootn; }
 167 boot2dev() { add-part $@ $boot2n; }
 168 efidev() { add-part $@ $efin; }
 169 grub_extdev() { add-part $@ $grub_extn; }
 170 bios_grubdev() { add-part $@ $bios_grubn; }
 171 even_bigdev() { add-part $@ $even_bign; }
 172
 173 crypt-dev() { echo /dev/mapper/crypt_dev_${1##*/}; }
 174 crypt-name() { echo crypt_dev_${1##*/}; }
 175 root-cryptdev() { crypt-dev $(rootdev $@); }
 176 root2-cryptdev() { crypt-dev $(root2dev $@); }
 177 swap-cryptdev() { crypt-dev $(swapdev $@); }
 178 root-cryptname() { crypt-name $(rootdev $@); }
 179 root2-cryptname() { crypt-name $(root2dev $@); }
 180 swap-cryptname() { crypt-name $(swapdev $@); }
 181
 182 dev-mib() {
 183   local d=${1:-$dev}
 184   echo $(( $(parted -m $d unit MiB print | \
 185                sed -nr "s#^/dev/[^:]+:([0-9]+).*#\1#p") - 1))
 186 }
 187
 188 luks-setup() {
 189   local luksdev="$1"
 190   # when we move to newer than trisquel 9, we can remove
 191   # --type luks1. We can also check on cryptsetup --help | less /compil
 192   # to see about the other settings. Default in debian 9 is luks2.
 193   # You can convert from luks2 to luks 1 by adding a temporary key:
 194   # cryptsetup luksAddKey --pbkdf pbkdf2
 195   # then remove the new format keys with cryptsetup luksRemoveKey
 196   # then cryptsetup convert DEV --type luks1, then readd old keys and remove temp.
 197   yes YES | cryptsetup luksFormat $luksdev $luks_file || [[ $? == 141 ]]
 198   yes "$lukspw" | \
 199     cryptsetup luksAddKey --key-file $luks_file \
 200                $luksdev || [[ $? == 141 ]]
 201   # background: Keyfile and password are treated just
 202   # like 2 ways to input a passphrase, so we don't actually need to have
 203   # different contents of keyfile and passphrase, but it makes some
 204   # security sense to a really big randomly generated passphrase
 205   # as much as possible, so we have both.
 206   #
 207   # This would remove the keyfile.
 208   #    yes 'test' | cryptsetup luksRemoveKey /dev/... \
 209     #                            /key/file || [[ $? == 141 ]]
 210   cryptsetup luksOpen $luksdev $(crypt-name $luksdev) --key-file $luks_file
 211 }
 212
 213 ##### end function defs
 214
 215 if ifclass REPARTITION; then
 216   partition=true # force a full wipe
 217 else
 218   partition=false # change to true to force a full wipe
 219 fi
 220 if ifclass NOWIPE; then
 221   wipe=false
 222 else
 223   wipe=true
 224 fi
 225
 226 if (($(nproc) > 2)); then
 227   mopts=,compress=zstd
 228 fi
 229
 230 declare -A disk_excludes
 231 if ! $mkroot2 && ! $mkroot2tab ! ifclass USE_MOUNTED; then
 232   ## ignore disks that are mounted, eg when running from fai-cd
 233   while read -r l; do
 234     eval "$l"
 235     if [[ ! $PKNAME ]]; then
 236       PKNAME="$KNAME"
 237     fi
 238     if [[ $MOUNTPOINT ]]; then
 239       disk_excludes[$PKNAME]=true
 240     fi
 241   done < <(lsblk -nP -o KNAME,MOUNTPOINT,PKNAME)
 242 fi
 243
 244 hdds=()
 245 ssds=()
 246 # this excludes usb. note: i may encounter some other type in the future.
 247 for disk in $(lsblk -do name,tran -n | awk '$2 ~ "^(sata|nvme)$" { print $1 }'); do
 248   if [[ ${disk_excludes[$disk]} ]]; then
 249     continue
 250   fi
 251   case $(cat /sys/block/$disk/queue/rotational) in
 252     0) ssds+=(/dev/$disk) ;;
 253     1) hdds+=(/dev/$disk) ;;
 254     *) echo "$0: error: unknown /sys/block/$disk/queue/rotational: \
 255 $(cat $disk/queue/rotational)"; exit 1 ;;
 256   esac
 257 done
 258
 259 # install all ssds, or if there are none, all hdds.
 260 # Note, usb flash disks are seen as rotational, which is
 261 # very odd, but convenient for ignoring them here.
 262 # TODO: find a reliable way to ignore them.
 263 if ! ifclass ROTATIONAL && (( ${#ssds[@]} > 0 )); then
 264   short_devs=( ${ssds[@]} )
 265 else
 266   short_devs=( ${hdds[@]} )
 267 fi
 268
 269 # check if the partitions exist have the right filesystems
 270 #blkid="$(blkid -s TYPE)"
 271 for dev in ${short_devs[@]}; do
 272   if $partition; then break; fi
 273   y=$(readlink -f $dev)
 274   arr=($y?*)
 275   if (( ${#arr[@]} < lastn )); then
 276     partition=true
 277   fi
 278   # On one system, blkid is missing some partitions.
 279   # maybe we need a flag, like FUZZY_BLKID or something, so we
 280   # can check that at least some exist.
 281   # for x in "`rootdev`: TYPE=\"crypto_LUKS\"" "`bootdev`: TYPE=\"btrfs\""; do
 282   #     echo "$blkid" | grep -Fx "$x" &>/dev/null || partition=true
 283   # done
 284 done
 285
 286 if $partition && ifclass PARTITION_PROMPT; then
 287   echo "Press any key except ctrl-c to continue and partition these drives:"
 288   echo "  ${short_devs[*]}"
 289   read -r
 290 fi
 291
 292 devs=()
 293 shopt -s extglob
 294 for short_dev in ${short_devs[@]}; do
 295   devs+=($(devbyid $short_dev))
 296 done
 297 if [[ ! ${devs[0]} ]]; then
 298   echo "$0: error: failed to detect devs" >&2
 299   exit 1
 300 fi
 301
 302 boot_space=0
 303 first=true
 304 boot_devs=()
 305 boot2_devs=()
 306 for dev in ${devs[@]}; do
 307   if ifclass frodo; then
 308     # I ran into a machine where the bios doesn't know about some disks,
 309     # so 1st stage of grub also doesn't know about them.
 310     # Also, grub does not support mounting degraded btrfs as far as
 311     # I can tell with some googling.
 312     # From within an arch install env, I could detect them by noting
 313     # their partitions were mixed with the next disk in /dev/disk/by-path,
 314     # and I have mixed model disks, and I could see the 8 models which showed
 315     # up in the bios, and thus see which 2 models were missing.
 316     # hdparm -I /dev/sdh will give model info in linux.
 317     # However, in fai on jessie, /dev/disk/by-path dir doesn't exist,
 318     # and I don't see another way, so I'm hardcoding them.
 319     # We still put grub on them and partition them the same, for uniformity
 320     # and in case they get moved to a system that can recognize them,
 321     # we just exclude them from the boot filesystem.
 322     cd /dev/disk/by-id/
 323     bad_disk=false
 324     for id in ata-TOSHIBA_MD04ACA500_8539K4TQFS9A \
 325                 ata-TOSHIBA_MD04ACA500_Y5IFK6IJFS9A; do
 326       if [[ $(readlink -f $id) == "$(readlink -f $dev)" ]]; then
 327         bad_disk=true
 328         break
 329       fi
 330     done
 331     if ! $bad_disk; then
 332       boot_devs+=($(bootdev))
 333       boot2_devs+=($(boot2dev))
 334     fi
 335   else
 336     boot_space=$(( boot_space + $(parted -m $dev unit MiB print | \
 337                                     sed -nr "s#^/dev/[^:]+:([0-9]+).*#\1#p") - 1))
 338     boot_devs+=($(bootdev))
 339     boot2_devs+=($(boot2dev))
 340   fi
 341   if $first && [[ $boot_devs ]]; then
 342     first_efi=$(efidev)
 343     first_grub_extdev=$(grub_extdev)
 344     first=false
 345   fi
 346 done
 347 first_boot_dev=${boot_devs[0]}
 348
 349 even_raid=false
 350 if ifclass RAID0 || (( ${#boot_devs[@]} == 1 )); then
 351   raid_level=0
 352 elif ifclass RAID1 || (( ${#boot_devs[@]} <= 3 )); then
 353   if (( ${#boot_devs[@]} == 2 )); then
 354     even_raid=true
 355   fi
 356   raid_level=1
 357 else
 358   raid_level=10
 359 fi
 360
 361 ### Begin calculate boot partition space
 362 # due to raid duplication
 363 case $raid_level in
 364   1*) boot_space=$(( boot_space / 2 )) ;;
 365 esac
 366 if (( boot_space > 60000 )); then
 367   # this is larger than needed for several /boot subvols,
 368   # becuase I keep a minimal debian install on it for
 369   # recovery needs and for doing pxe-kexec.
 370   boot_mib=10000
 371   root2_mib=1000000
 372   boot2_mib=5000
 373 elif (( boot_space > 30000 )); then
 374   boot_mib=$(( 5000 + (boot_space - 30000) / 2 ))
 375   root2_mib=100
 376   boot2_mib=100
 377 else
 378   # Small vms don't have room for /boot recovery.  With 3 kernels
 379   # installed, i'm using 132M on t8, so this seems like plenty of
 380   # room. note: rhel 8 recomments 1g for /boot.  u20.04 with 3 kernels =
 381   # 308 mb, so things have grown significantly
 382   boot_mib=1000
 383   root2_mib=100
 384   boot2_mib=100
 385 fi
 386 case $raid_level in
 387   1*)
 388     boot_mib=$(( boot_mib * 2 ))
 389     boot2_mib=$(( boot2_mib * 2 ))
 390     root2_mib=$(( root2_mib * 2 ))
 391     ;;
 392 esac
 393 ### end calculate boot partition space
 394
 395
 396
 397 if [[ ! $DISTRO ]]; then
 398   if ifclass VOL_BULLSEYE_BOOTSTRAP; then
 399     DISTRO=debianbullseye_bootstrap
 400   elif ifclass VOL_STRETCH; then
 401     DISTRO=debianstretch
 402   elif ifclass VOL_BUSTER; then
 403     DISTRO=debianbuster
 404   elif ifclass VOL_BULLSEYE; then
 405     DISTRO=debianbullseye
 406   elif ifclass VOL_BOOKWORM; then
 407     DISTRO=debianbookworm
 408   elif ifclass VOL_TESTING; then
 409     DISTRO=debiantesting
 410   elif ifclass VOL_XENIAL; then
 411     DISTRO=ubuntuxenial
 412   elif ifclass VOL_BIONIC; then
 413     DISTRO=ubuntubionic
 414   elif ifclass VOL_FOCAL; then
 415     DISTRO=ubuntufocal
 416   elif ifclass VOL_FLIDAS; then
 417     DISTRO=trisquelflidas
 418   elif ifclass VOL_ETIONA; then
 419     DISTRO=trisqueletiona
 420   elif ifclass VOL_NABIA; then
 421     DISTRO=trisquelnabia
 422   elif $mkroot2 || $mkroot2tab; then
 423     :
 424   else
 425     echo "PARTITIONER ERROR: no distro class/var set" >&2
 426     exit 1
 427   fi
 428 fi
 429
 430
 431 bpart() { # btrfs a partition
 432   case $raid_level in
 433     0) mkfs.btrfs -f $@ ;;
 434     1) mkfs.btrfs -f -m raid1 -d raid1 $@ ;;
 435     10) mkfs.btrfs -f -m raid10 -d raid10 $@ ;;
 436   esac
 437 }
 438
 439
 440 if [[ ! $luks_dir ]]; then
 441   # see README for docs about how to create these
 442   luks_dir=$FAI/distro-install-common/luks
 443   if [[ ! -d $luks_dir ]]; then
 444     luks_dir=/q/root/luks
 445   fi
 446   if [[ ! -d $luks_dir ]]; then
 447     echo "$0: error: no luks_dir found" >&2
 448     exit 1
 449   fi
 450 fi
 451
 452 luks_file=$luks_dir/host-$HOSTNAME
 453 if [[ ! -e $luks_file ]]; then
 454   hostkeys=($luks_dir/host-*)
 455   # if there is only one key, we might be deploying somewhere
 456   # where dhcp doesnt give us a proper hostname, so use that.
 457   if [[ ${#hostkeys[@]} == 1 && -e ${hostkeys[0]} ]]; then
 458     luks_file=${hostkeys[0]}
 459   else
 460     echo "$0: error: no key for hostname at $luks_file" >&2
 461     exit 1
 462   fi
 463 fi
 464
 465 # # note, corresponding changes in /b/ds/keyscript-{on,off}
 466 if ifclass demohost; then
 467   lukspw=x
 468 elif [[ -e $luks_dir/$HOSTNAME ]]; then
 469   lukspw=$(cat $luks_dir/$HOSTNAME)
 470 else
 471   lukspw=$(cat $luks_dir/iank)
 472 fi
 473
 474
 475 first_root_crypt=$(root-cryptdev ${devs[0]})
 476
 477 # 1.5 x based on https://access.redhat.com/documentation/en-US/Red_Hat_Enterprise_Linux/7/html/Installation_Guide/sect-disk-partitioning-setup-x86.html#sect-custom-partitioning-x86
 478 swap_mib=$(( $(grep ^MemTotal: /proc/meminfo | \
 479                  awk '{print $2}') * 3/(${#devs[@]} * 2 ) / 1024 ))
 480
 481 mkdir -p /tmp/fai
 482 root_devs=()
 483 for dev in ${devs[@]}; do
 484   root_devs+=($(rootdev))
 485 done
 486 shopt -s nullglob
 487
 488 # We write to these files instead of just /etc/fstab, /etc/crypttab,
 489 # because these are filesystems created after our current root, and so
 490 # this allows us to update other root filesystems too.
 491 rm -f /mnt/root/root2-{fs,crypt}tab
 492 if $mkroot2 || $mkroot2tab; then
 493   if $partition; then
 494     echo $0: error: found partition=true but have mkroot2 arg
 495     exit 1
 496   fi
 497   for dev in ${devs[@]}; do
 498     if $mkroot2; then
 499       luks_file=$luks_dir/host-amy
 500       lukspw=$(cat $luks_dir/amy)
 501       luks-setup $(root2dev)
 502     fi
 503     cat >>/mnt/root/root2-crypttab <<EOF
 504 $(root2-cryptname) $(root2dev)  $luks_file  discard,luks,initramfs
 505 EOF
 506   done
 507   if $mkroot2; then
 508     bpart $(for dev in ${devs[@]}; do root2-cryptdev; done)
 509     bpart ${boot2_devs[@]}
 510   fi
 511   mkdir -p /mnt/root2 /mnt/boot2
 512   cat >>/mnt/root/root2-fstab <<EOF
 513 $(root2-cryptdev ${devs[0]}) /mnt/root2  btrfs  nofail,x-systemd.device-timeout=30s,x-systemd.mount-timeout=30s,noatime,subvolid=0$mopts  0 0
 514 ${boot2_devs[0]} /mnt/boot2  btrfs    nofail,x-systemd.device-timeout=30s,x-systemd.mount-timeout=30s,noatime,subvolid=0  0 0
 515 EOF
 516   exit 0
 517 fi
 518
 519
 520 if $partition; then
 521   ### begin wipefs
 522   if [[ ! $SPECIAL_DISK ]]; then
 523     for dev in ${devs[@]}; do
 524       # if we repartition to the same as an old partition,
 525       # we don't want any old fses hanging around.
 526       for (( i=1; i <= lastn; i++ )); do
 527         x=$(add-part $i)
 528         [[ -e $x ]] || continue
 529         count_down=10
 530         # wipefs has failed, manual run works, google suggests timing issue
 531         while ! wipefs -a $x; do
 532           sleep 2
 533           count_down=$((count_down - 1))
 534           (( count_down > 0 )) || exit 1
 535         done
 536       done
 537     done
 538   fi
 539   ### end wipefs
 540
 541
 542   # When we have 2 disks of at least 100g difference in size,
 543   # make an extra partition on the end of the bigger one.
 544   even_big_part=false
 545   even_diff_min=100000
 546   if $even_raid; then
 547     smalli=0
 548     bigi=1
 549     if (( $(dev-mib ${devs[0]}) >= $(dev-mib ${devs[1]}) )); then
 550       smalli=1
 551       bigi=0
 552     fi
 553     disk_mib=$(dev-mib ${devs[smalli]})
 554     even_big_dev=${devs[bigi]}
 555     even_big_mib=$(dev-mib $even_big_dev)
 556     if (( even_big_mib - disk_mib > even_diff_min )); then
 557       even_big_part=true
 558     fi
 559   fi
 560
 561   for dev in ${devs[@]}; do
 562     if [[ $SPECIAL_DISK ]]; then
 563       dev=$(devbyid $SPECIAL_DISK)
 564     fi
 565
 566     # parted will round up the disk size. Do -1 so we can have
 567     # fully 1MiB unit partitions for easy resizing of the last partition.
 568     # Otherwise we would pass in -0 for the end argument for the last partition.
 569     #
 570     # Note: parted print error output is expected. example:
 571     # Error: /dev/vda: unrecognised disk label
 572     if ! $even_raid; then
 573       disk_mib=$(dev-mib)
 574     fi
 575
 576     boot_part_mib=$(( boot_mib / ${#boot_devs[@]} ))
 577     boot2_part_mib=$(( boot2_mib / ${#boot_devs[@]} ))
 578     root2_part_mib=$(( root2_mib / ${#root_devs[@]} ))
 579     root_end=$(( disk_mib - root2_part_mib - swap_mib - boot_part_mib - boot2_part_mib ))
 580     root2_end=$(( root_end + root2_part_mib ))
 581     swap_end=$(( root2_end + swap_mib ))
 582     boot_end=$(( swap_end + boot_part_mib ))
 583
 584     parted -s $dev mklabel gpt
 585     # MiB because parted complains about alignment otherwise.
 586     pcmd="parted -a optimal -s -- $dev"
 587     # root partition, the main big one
 588     $pcmd mkpart primary ext3 524MiB ${root_end}MiB
 589     # without naming, systemd gives us misc errors like:
 590     # dev-disk-by\x2dpartlabel-primary.device: Dev dev-disk-by\x2dpartlabel-primary.device appeared twice
 591     $pcmd name $rootn root
 592     # root2 partition
 593     $pcmd mkpart primary ext3 ${root_end}MiB ${root2_end}MiB
 594     $pcmd name $root2n root2
 595     # normally a swap is type "linux-swap", but this is encrypted swap. using that
 596     # label will confuse systemd.
 597     # swap partition
 598     $pcmd mkpart primary "" ${root2_end}MiB ${swap_end}MiB
 599     $pcmd name $swapn swap
 600     # boot partition
 601     $pcmd mkpart primary "" ${swap_end}MiB ${boot_end}MiB
 602     $pcmd name $bootn boot
 603     # boot2 partition
 604     $pcmd mkpart primary "" ${boot_end}MiB ${disk_mib}MiB
 605     $pcmd name $boot2n boot2
 606     # uefi partition. efi sucks, half a gig, rediculous.
 607     $pcmd mkpart primary "fat32" 12MiB 524MiB
 608     $pcmd name $efin efi
 609     $pcmd set $efin esp on
 610     # note, this is shown here: https://support.system76.com/articles/bootloader/
 611     # but not mentioned https://wiki.archlinux.org/index.php/EFI_system_partition
 612     # probably not needed
 613     $pcmd set $bootn boot on
 614     $pcmd set $boot2n boot on
 615     # i only need a few k, but googling min size,
 616     # I found someone saying that gparted required
 617     # required at least 8 because of their hard drive cylinder size.
 618     # And 8 is still very tiny.
 619     # grub_ext partition
 620     $pcmd mkpart primary "ext2" 4MiB 12MiB
 621     $pcmd name $grub_extn grubext
 622     # gpt ubuntu cloud image uses ~4 mb for this partition. fai uses 1 MiB.
 623     # so, I use 3, whatever.
 624     # note: parted manual saying cheap flash media
 625     # should to start at 4.
 626     # biols grub partition
 627     $pcmd mkpart primary "" 1MiB 4MiB
 628     $pcmd name $bios_grubn biosgrub
 629     $pcmd set $bios_grubn bios_grub on
 630     if $even_big_part  && [[ $dev == "$even_big_dev" ]]; then
 631       $pcmd mkpart primary ext3 ${disk_mib}MiB ${even_big_mib}MiB
 632       $pcmd name $even_bign even_big
 633     fi
 634
 635     # the mkfs failed before on a vm, which prompted me to add
 636     # sleep .1
 637     # then it failed again on a physical machine
 638     # with:
 639     # Device /dev/disk/by-id/foo doesn't exist or access denied,
 640     # so I added a wait until it existed.
 641     # Then I added the mkfs.ext2, which claimed to succeed,
 642     # but then couldn't be found upon reboot. In that case we didn't
 643     # wait at all. So I've added a 3 second minimum wait.
 644     secs=0
 645     while [[ ! -e $(bios_grubdev) ]] && (( secs < 10 )); do
 646       sleep 1
 647       secs=$((secs +1))
 648     done
 649     sleep 3
 650
 651     mkfs.fat -F32 $(efidev)
 652
 653     if $even_big_part  && [[ $dev == "$even_big_dev" ]]; then
 654       luks-setup $(even_bigdev)
 655       mkfs.btrfs -f $(crypt-dev $(even_bigdev))
 656     fi
 657
 658     # Holds just a single file, rarely written, so
 659     # use ext2, like was often used for the /boot partition.
 660     # This exists because grub can only persist data to a non-cow fs.
 661     # And we use persisting a var in grub to do a one time boot.
 662     # We could pass the data on the kernel command line and persist it
 663     # to grubenv after booting, but that relies on the boot always succeeding.
 664     # This is just a bit more robust, and it could work for booting
 665     # into ipxe which can't persist data, if we ever got that working.
 666     mkfs.ext2 $(grub_extdev)
 667     luks-setup $(rootdev)
 668
 669     if [[ $SPECIAL_DISK ]]; then
 670       exit 0
 671     fi
 672   done
 673   ls -la /dev/btrfs-control # this was probably for debugging...
 674   sleep 1
 675   bpart $(for dev in ${devs[@]}; do root-cryptdev; done)
 676   bpart ${boot_devs[@]}
 677 else
 678   for dev in ${devs[@]}; do
 679     if [[ -e /dev/mapper/$(root-cryptname) ]]; then
 680       continue
 681     fi
 682     cryptsetup luksOpen $(rootdev) $(root-cryptname) \
 683                --key-file $luks_file
 684   done
 685   sleep 1
 686 fi
 687
 688
 689 if $wipe && [[ $DISTRO != debianbullseye_bootstrap ]]; then
 690   # bootstrap distro doesn't use separate encrypted root.
 691   mount -o subvolid=0 $first_root_crypt /mnt
 692   # systemd creates subvolumes we want to delete.
 693   s=($(btrfs subvolume list --sort=-path /mnt |
 694          sed -rn "s#^.*path\s*(root_$DISTRO/\S+)\s*\$#\1#p"))
 695   for subvol in ${s[@]}; do btrfs subvolume delete /mnt/$subvol; done
 696   btrfs subvolume set-default 0 /mnt
 697   [[ ! -e /mnt/root_$DISTRO ]] || btrfs subvolume delete /mnt/root_$DISTRO
 698
 699   ## create subvols ##
 700   cd /mnt
 701
 702   btrfs subvolume create root_$DISTRO
 703
 704   # could set default subvol like this, but no reason to.
 705   # btrfs subvolume set-default \
 706     #       $(btrfs subvolume list . | grep "root_$DISTRO$" | awk '{print $2}') .
 707
 708   # For raid systems, cow allows for error correction, for non-raid systems,
 709   # protects root fs from having the plug pulled. Reprovisioning a root
 710   # subvol is not my favorite thing to do.
 711   # # no cow on the root filesystem. it's setup is fully scripted,
 712   # # if it's messed up, we will just recreated it,
 713   # # and we can get better perf with this.
 714   # # I can't remember exactly why, but this is preferable to mounting with
 715   # # -o nodatacow, I think because subvolumes inherit that.
 716   # chattr -Rf +C root_$DISTRO
 717   cd /
 718   umount /mnt
 719 fi
 720
 721 mount -o subvolid=0 $first_boot_dev /mnt
 722 cd /mnt
 723 btrfs subvolume set-default 0 /mnt # already default, just ensuring it.
 724
 725 # for libreboot systems. grub2 only reads from subvolid=0
 726 mkdir -p /mnt/grub2
 727 cp $FAI/distro-install-common/libreboot_grub.cfg /mnt/grub2
 728
 729 if [[ $DISTRO == debianbullseye_bootstrap ]]; then
 730   # this is just convenience for the libreboot_grub config
 731   # so we can glob the other ones easier.
 732   boot_vol=$DISTRO
 733 else
 734   boot_vol=boot_$DISTRO
 735 fi
 736 if $wipe && [[ -e /mnt/$boot_vol ]]; then
 737   btrfs subvolume delete /mnt/$boot_vol
 738 fi
 739 if [[ ! -e /mnt/$boot_vol ]]; then
 740   btrfs subvolume create $boot_vol
 741 fi
 742 cd /
 743 umount /mnt
 744 ## end create subvols ##
 745
 746 dev=${boot_devs[0]}
 747 mount $first_grub_extdev /mnt
 748 grub-editenv /mnt/grubenv set did_fai_check=true
 749 grub-editenv /mnt/grubenv set last_boot=/$boot_vol
 750 umount /mnt
 751
 752 fstabstd=x-systemd.device-timeout=30s,x-systemd.mount-timeout=30s
 753 if [[ $DISTRO == debianbullseye_bootstrap ]]; then
 754   cat > /tmp/fai/fstab <<EOF
 755 $first_boot_dev  /  btrfs  noatime,subvol=$boot_vol  0 0
 756 $first_efi  /boot/efi  vfat          nofail,$fstabstd  0 0
 757 EOF
 758   cat >/tmp/fai/disk_var.sh <<EOF
 759 BOOT_DEVICE="${short_devs[@]}"
 760 ROOT_PARTITION=$first_boot_dev
 761 EOF
 762 else
 763   # note, fai creates the mountpoints listed here
 764   cat > /tmp/fai/fstab <<EOF
 765 $first_root_crypt  /  btrfs $fstabstdopts,noatime,subvol=root_$DISTRO$mopts  0 0
 766 $first_root_crypt  /mnt/root  btrfs  nofail,$fstabstd,noatime,subvolid=0$mopts  0 0
 767 $first_boot_dev  /boot  btrfs        nofail,$fstabstd,noatime,subvol=$boot_vol  0 0
 768 $first_efi  /boot/efi  vfat          nofail,$fstabstd  0 0
 769 $first_boot_dev  /mnt/boot  btrfs    nofail,$fstabstd,noatime,subvolid=0  0 0
 770 EOF
 771   swaps=()
 772   rm -f /tmp/fai/crypttab
 773   for dev in ${devs[@]}; do
 774     swaps+=($(swap-cryptname))
 775     cat >>/tmp/fai/crypttab <<EOF
 776 $(root-cryptname) $(rootdev)  none  keyscript=/root/keyscript,discard,luks,initramfs
 777 $(swap-cryptname) $(swapdev)  /dev/urandom  swap,cipher=aes-xts-plain64,size=256,hash=ripemd160
 778 EOF
 779     cat >> /tmp/fai/fstab <<EOF
 780 $(swap-cryptdev)  none  swap  nofail,$fstabstd,sw  0 0
 781 EOF
 782   done
 783
 784   # fai would do this:
 785   #BOOT_DEVICE=\${BOOT_DEVICE:-"${devs[0]}"}
 786
 787   # note: swaplist seems to do nothing.
 788   cat >/tmp/fai/disk_var.sh <<EOF
 789 BOOT_DEVICE="${short_devs[@]}"
 790 BOOT_PARTITION=\${BOOT_PARTITION:-$first_boot_dev}
 791 # ROOT_PARTITIONS is added by me, used in arch setup.
 792 ROOT_PARTITIONS="${root_devs[@]}"
 793 ROOT_PARTITION=\${ROOT_PARTITION:-$first_root_crypt}
 794 SWAPLIST=\${SWAPLIST:-"${swaps[@]}"}
 795 EOF
 796
 797   if [[ $HOSTNAME == kd ]]; then
 798     # note, having these with keyscript and initramfs causes a luks error in fai.log,
 799     # but it is safely ignorable and gets us the ability to just type our password
 800     # in once at boot. A downside is that they are probably needed to be plugged in to boot.
 801     cat >>/tmp/fai/crypttab <<EOF
 802 crypt_dev_ata-Samsung_SSD_870_QVO_8TB_S5VUNG0N900656V-part${even_bign} /dev/disk/by-id/ata-Samsung_SSD_870_QVO_8TB_S5VUNG0N900656V-part7  none  keyscript=decrypt_keyctl,discard,luks,initramfs
 803 crypt_dev_ata-TOSHIBA_MD04ACA500_84R2K773FS9A-part1 /dev/disk/by-id/ata-TOSHIBA_MD04ACA500_84R2K773FS9A-part1  none  keyscript=decrypt_keyctl,discard,luks,initramfs
 804 crypt_dev_ata-ST6000DM001-1XY17Z_Z4D29EBL-part1 /dev/disk/by-id/ata-ST6000DM001-1XY17Z_Z4D29EBL-part1  none  keyscript=decrypt_keyctl,discard,luks,initramfs
 805 EOF
 806     cat >> /tmp/fai/fstab <<EOF
 807 # r7 = root partition7. it isnt actually #7 anymore, not a great name, but whatever
 808 /dev/mapper/crypt_dev_ata-Samsung_SSD_870_QVO_8TB_S5VUNG0N900656V-part${even_bign}  /mnt/r7  btrfs  nofail,$fstabstd,noatime,compress=zstd,subvolid=0  0 0
 809 /dev/mapper/crypt_dev_ata-TOSHIBA_MD04ACA500_84R2K773FS9A-part1  /mnt/rust1  btrfs  nofail,$fstabstd,noatime,compress=zstd,subvolid=0  0 0
 810 /dev/mapper/crypt_dev_ata-ST6000DM001-1XY17Z_Z4D29EBL-part1  /mnt/rust2  btrfs  nofail,$fstabstd,noatime,compress=zstd,subvolid=0  0 0
 811 EOF
 812   fi
 813
 814 fi
 815
 816 # initial setup of extra data fs, mounted,
 817 # btrfs subvol create nocow
 818 # chattr +C nocow
 819 # chown iank.iank nocow
 820