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
  39 #
  40 # environment variables:
  41 #
  42 # HOSTNAME: if frodo, we exclude 2 devices from the /boot array, which
  43 # the bios does not see. if demohost, we set the luks password to just
  44 # 'x'.
  45 #
  46 # SPECIAL_DISK: For use outside of fai. A base disk name like
  47 # /dev/sdk. If set, we just cryptsetup and partition this one disk then
  48 # exit. This is useful for partitioning a disk in preparation to replace
  49 # a failed or failing disk from a raid10 array.
  50 #
  51 # classes:
  52 #
  53 # REPARTITION: forces repartitioning even if we detect the proper amount
  54 # of partitions already exist.
  55 #
  56 # NOWIPE: use existing subvolumes if they exist
  57 #
  58 # ROTATIONAL: forces to install onto hdds instead of sdds. normally sdds
  59 # are chosen if they exist.
  60 #
  61 # PARTITION_PROMPT: command line prompt before partitioning
  62 #
  63 # RAID0: forces raid0 filesystem. Normally with 4+ devices, we use
  64 # raid10.
  65 # RAID1: forces raid1 filesystem.
  66
  67 mkroot2=false
  68 case $1 in
  69   mkroot2)
  70     mkroot2=true
  71     ;;
  72   *)
  73     echo "$0: error: unsupported arg: $1" >&2
  74     exit 1
  75     ;;
  76 esac
  77
  78
  79
  80 if [[ $SPECIAL_DISK ]]; then
  81   export CLASS_REPARTITION=true
  82 fi
  83
  84 # # fai's setup-storage won't do btrfs on luks,
  85 # # so we do it ourself :)
  86 # inspiration taken from files in fai-setup-storage package
  87
  88 # if we are not running in fai, skiptask won't be defined, so carry on.
  89 skiptask partition || ! type skiptask
  90
  91 if ! type -p devbyid; then
  92   for d in $FAI/distro-install-common \
  93                /a/bin/fai/fai/config/distro-install-common $FAI $PWD; do
  94     [[ -d $d ]] || continue
  95     if [[ -e $d/devbyid ]]; then
  96       devbyid=$d/devbyid
  97       devbyid() { $devbyid "$@"; }
  98       break
  99     fi
 100   done
 101   if [[ ! $devbyid ]]; then
 102     echo "$0: error: failed to find devbyid script" >&2
 103     exit 1
 104   fi
 105 fi
 106
 107
 108
 109 #### begin configuration
 110
 111 # this is the ordering of the /dev/sdaX, but
 112 # the ordering of the partition layout goes like this:
 113 # bios_grub
 114 # grub_ext
 115 # efi
 116 # root
 117 # swap
 118 # boot
 119
 120 rootn=1
 121 root2n=2
 122 swapn=3
 123 bootn=4
 124 boot2n=5
 125 efin=6
 126 # ext partition so grub can write persistent variables,
 127 # so it can do a one time boot. grub can't write to
 128 # btrfs or any cow fs because it's more
 129 # more complicated to do and they don't want to.
 130 grub_extn=7
 131 # bios boot partition,
 132 # https://wiki.archlinux.org/index.php/GRUB
 133 bios_grubn=8
 134 even_bign=9
 135 lastn=$bios_grubn
 136
 137
 138
 139 ##### end configuration
 140
 141
 142 add-part() { # add partition suffix to $dev
 143   local d part
 144   if [[ $# == 1 ]]; then
 145     d=$dev
 146     part=$1
 147   else
 148     d=$1
 149     part=$2
 150   fi
 151   echo $d-part$part
 152 }
 153
 154 rootdev() { add-part $@ $rootn; }
 155 root2dev() { add-part $@ $root2n; }
 156 swapdev() { add-part $@ $swapn; }
 157 bootdev() { add-part $@ $bootn; }
 158 boot2dev() { add-part $@ $boot2n; }
 159 efidev() { add-part $@ $efin; }
 160 grub_extdev() { add-part $@ $grub_extn; }
 161 bios_grubdev() { add-part $@ $bios_grubn; }
 162 even_bigdev() { add-part $@ $even_bign; }
 163
 164 crypt-dev() { echo /dev/mapper/crypt_dev_${1##*/}; }
 165 crypt-name() { echo crypt_dev_${1##*/}; }
 166 root-cryptdev() { crypt-dev $(rootdev $@); }
 167 root2-cryptdev() { crypt-dev $(root2dev $@); }
 168 swap-cryptdev() { crypt-dev $(swapdev $@); }
 169 root-cryptname() { crypt-name $(rootdev $@); }
 170 root2-cryptname() { crypt-name $(root2dev $@); }
 171 swap-cryptname() { crypt-name $(swapdev $@); }
 172
 173 dev-mib() {
 174   local d=${1:-$dev}
 175   echo $(( $(parted -m $d unit MiB print | \
 176                sed -nr "s#^/dev/[^:]+:([0-9]+).*#\1#p") - 1))
 177 }
 178
 179 luks-setup() {
 180   local luksdev="$1"
 181   # when we move to newer than trisquel 9, we can remove
 182   # --type luks1. We can also check on cryptsetup --help | less /compil
 183   # to see about the other settings. Default in debian 9 is luks2.
 184   # You can convert from luks2 to luks 1 by adding a temporary key:
 185   # cryptsetup luksAddKey --pbkdf pbkdf2
 186   # then remove the new format keys with cryptsetup luksRemoveKey
 187   # then cryptsetup convert DEV --type luks1, then readd old keys and remove temp.
 188   yes YES | cryptsetup luksFormat $luksdev $luks_file \
 189                        --type luks1 -c aes-cbc-essiv:sha256 -s 256 || [[ $? == 141 ]]
 190   yes "$lukspw" | \
 191     cryptsetup luksAddKey --key-file $luks_file \
 192                $luksdev || [[ $? == 141 ]]
 193   # background: Keyfile and password are treated just
 194   # like 2 ways to input a passphrase, so we don't actually need to have
 195   # different contents of keyfile and passphrase, but it makes some
 196   # security sense to a really big randomly generated passphrase
 197   # as much as possible, so we have both.
 198   #
 199   # This would remove the keyfile.
 200   #    yes 'test' | cryptsetup luksRemoveKey /dev/... \
 201     #                            /key/file || [[ $? == 141 ]]
 202   cryptsetup luksOpen $luksdev $(crypt-name $luksdev) --key-file $luks_file
 203 }
 204
 205 ##### end function defs
 206
 207 if ifclass REPARTITION; then
 208   partition=true # force a full wipe
 209 else
 210   partition=false # change to true to force a full wipe
 211 fi
 212 if ifclass NOWIPE; then
 213   wipe=false
 214 else
 215   wipe=true
 216 fi
 217
 218 if (($(nproc) > 2)); then
 219   mopts=,compress=zstd
 220 fi
 221
 222 declare -A disk_excludes
 223 if ! $mkroot2 && ! ifclass USE_MOUNTED; then
 224   ## ignore disks that are mounted, eg when running from fai-cd
 225   while read -r l; do
 226     eval "$l"
 227     if [[ ! $PKNAME ]]; then
 228       PKNAME="$KNAME"
 229     fi
 230     if [[ $MOUNTPOINT ]]; then
 231       disk_excludes[$PKNAME]=true
 232     fi
 233   done < <(lsblk -nP -o KNAME,MOUNTPOINT,PKNAME)
 234 fi
 235
 236 hdds=()
 237 ssds=()
 238 # this excludes usb. note: i may encounter some other type in the future.
 239 for disk in $(lsblk -do name,tran -n | awk '$2 ~ "^(sata|nvme)$" { print $1 }'); do
 240   if [[ ${disk_excludes[$disk]} ]]; then
 241     continue
 242   fi
 243   case $(cat /sys/block/$disk/queue/rotational) in
 244     0) ssds+=(/dev/$disk) ;;
 245     1) hdds+=(/dev/$disk) ;;
 246     *) echo "$0: error: unknown /sys/block/$disk/queue/rotational: \
 247 $(cat $disk/queue/rotational)"; exit 1 ;;
 248   esac
 249 done
 250
 251 # install all ssds, or if there are none, all hdds.
 252 # Note, usb flash disks are seen as rotational, which is
 253 # very odd, but convenient for ignoring them here.
 254 # TODO: find a reliable way to ignore them.
 255 if ! ifclass ROTATIONAL && (( ${#ssds[@]} > 0 )); then
 256   short_devs=( ${ssds[@]} )
 257 else
 258   short_devs=( ${hdds[@]} )
 259 fi
 260
 261 # check if the partitions exist have the right filesystems
 262 #blkid="$(blkid -s TYPE)"
 263 for dev in ${short_devs[@]}; do
 264   if $partition; then break; fi
 265   y=$(readlink -f $dev)
 266   arr=($y?*)
 267   if (( ${#arr[@]} < lastn )); then
 268     partition=true
 269   fi
 270   # On one system, blkid is missing some partitions.
 271   # maybe we need a flag, like FUZZY_BLKID or something, so we
 272   # can check that at least some exist.
 273   # for x in "`rootdev`: TYPE=\"crypto_LUKS\"" "`bootdev`: TYPE=\"btrfs\""; do
 274   #     echo "$blkid" | grep -Fx "$x" &>/dev/null || partition=true
 275   # done
 276 done
 277
 278 if $partition && ifclass PARTITION_PROMPT; then
 279   echo "Press any key except ctrl-c to continue and partition these drives:"
 280   echo "  ${short_devs[*]}"
 281   read -r
 282 fi
 283
 284 devs=()
 285 shopt -s extglob
 286 for short_dev in ${short_devs[@]}; do
 287   devs+=($(devbyid $short_dev))
 288 done
 289 if [[ ! ${devs[0]} ]]; then
 290   echo "$0: error: failed to detect devs" >&2
 291   exit 1
 292 fi
 293
 294 boot_space=0
 295 first=true
 296 boot_devs=()
 297 boot2_devs=()
 298 for dev in ${devs[@]}; do
 299   if ifclass frodo; then
 300     # I ran into a machine where the bios doesn't know about some disks,
 301     # so 1st stage of grub also doesn't know about them.
 302     # Also, grub does not support mounting degraded btrfs as far as
 303     # I can tell with some googling.
 304     # From within an arch install env, I could detect them by noting
 305     # their partitions were mixed with the next disk in /dev/disk/by-path,
 306     # and I have mixed model disks, and I could see the 8 models which showed
 307     # up in the bios, and thus see which 2 models were missing.
 308     # hdparm -I /dev/sdh will give model info in linux.
 309     # However, in fai on jessie, /dev/disk/by-path dir doesn't exist,
 310     # and I don't see another way, so I'm hardcoding them.
 311     # We still put grub on them and partition them the same, for uniformity
 312     # and in case they get moved to a system that can recognize them,
 313     # we just exclude them from the boot filesystem.
 314     cd /dev/disk/by-id/
 315     bad_disk=false
 316     for id in ata-TOSHIBA_MD04ACA500_8539K4TQFS9A \
 317                 ata-TOSHIBA_MD04ACA500_Y5IFK6IJFS9A; do
 318       if [[ $(readlink -f $id) == "$(readlink -f $dev)" ]]; then
 319         bad_disk=true
 320         break
 321       fi
 322     done
 323     if ! $bad_disk; then
 324       boot_devs+=($(bootdev))
 325       boot2_devs+=($(boot2dev))
 326     fi
 327   else
 328     boot_space=$(( boot_space + $(parted -m $dev unit MiB print | \
 329                                     sed -nr "s#^/dev/[^:]+:([0-9]+).*#\1#p") - 1))
 330     boot_devs+=($(bootdev))
 331     boot2_devs+=($(boot2dev))
 332   fi
 333   if $first && [[ $boot_devs ]]; then
 334     first_efi=$(efidev)
 335     first_grub_extdev=$(grub_extdev)
 336     first=false
 337   fi
 338 done
 339 first_boot_dev=${boot_devs[0]}
 340
 341 even_raid=false
 342 if ifclass RAID0 || (( ${#boot_devs[@]} == 1 )); then
 343   raid_level=0
 344 elif ifclass RAID1 || (( ${#boot_devs[@]} <= 3 )); then
 345   if (( ${#boot_devs[@]} == 2 )); then
 346     even_raid=true
 347   fi
 348   raid_level=1
 349 else
 350   raid_level=10
 351 fi
 352
 353 ### Begin calculate boot partition space
 354 # due to raid duplication
 355 case $raid_level in
 356   1*) boot_space=$(( boot_space / 2 )) ;;
 357 esac
 358 if (( boot_space > 60000 )); then
 359   # this is larger than needed for several /boot subvols,
 360   # becuase I keep a minimal debian install on it for
 361   # recovery needs and for doing pxe-kexec.
 362   boot_mib=10000
 363   root2_mib=200000
 364   boot2_mib=500
 365 elif (( boot_space > 30000 )); then
 366   boot_mib=$(( 5000 + (boot_space - 30000) / 2 ))
 367   root2_mib=100
 368   boot2_mib=100
 369 else
 370   # Small vms don't have room for /boot recovery.  With 3 kernels
 371   # installed, i'm using 132M on t8, so this seems like plenty of
 372   # room. note: rhel 8 recomments 1g for /boot.
 373   boot_mib=500
 374   root2_mib=100
 375   boot2_mib=100
 376 fi
 377 case $raid_level in
 378   1*)
 379     boot_mib=$(( boot_mib * 2 ))
 380     boot2_mib=$(( boot2_mib * 2 ))
 381     root2_mib=$(( root2_mib * 2 ))
 382     ;;
 383 esac
 384 ### end calculate boot partition space
 385
 386
 387
 388 if [[ ! $DISTRO ]]; then
 389   if ifclass VOL_BUSTER_BOOTSTRAP; then
 390     DISTRO=debianbuster_bootstrap
 391   elif ifclass VOL_STRETCH; then
 392     DISTRO=debianstretch
 393   elif ifclass VOL_BUSTER; then
 394     DISTRO=debianbuster
 395   elif ifclass VOL_BULLSEYE; then
 396     DISTRO=debianbullseye
 397   elif ifclass VOL_TESTING; then
 398     DISTRO=debiantesting
 399   elif ifclass VOL_XENIAL; then
 400     DISTRO=ubuntuxenial
 401   elif ifclass VOL_BIONIC; then
 402     DISTRO=ubuntubionic
 403   elif ifclass VOL_FOCAL; then
 404     DISTRO=ubuntufocal
 405   elif ifclass VOL_FLIDAS; then
 406     DISTRO=trisquelflidas
 407   elif ifclass VOL_ETIONA; then
 408     DISTRO=trisqueletiona
 409   elif ifclass VOL_NABIA; then
 410     DISTRO=trisquelnabia
 411   elif $mkroot2; then
 412     :
 413   else
 414     echo "PARTITIONER ERROR: no distro class/var set" >&2
 415     exit 1
 416   fi
 417 fi
 418
 419
 420 bpart() { # btrfs a partition
 421   case $raid_level in
 422     0) mkfs.btrfs -f $@ ;;
 423     1) mkfs.btrfs -f -m raid1 -d raid1 $@ ;;
 424     10) mkfs.btrfs -f -m raid10 -d raid10 $@ ;;
 425   esac
 426 }
 427
 428
 429 if [[ ! $luks_dir ]]; then
 430   # see README for docs about how to create these
 431   luks_dir=$FAI/distro-install-common/luks
 432   if [[ ! -d $luks_dir ]]; then
 433     luks_dir=/q/root/luks
 434   fi
 435   if [[ ! -d $luks_dir ]]; then
 436     echo "$0: error: no luks_dir found" >&2
 437     exit 1
 438   fi
 439 fi
 440
 441 luks_file=$luks_dir/host-$HOSTNAME
 442 if [[ ! -e $luks_file ]]; then
 443   hostkeys=($luks_dir/host-*)
 444   # if there is only one key, we might be deploying somewhere
 445   # where dhcp doesnt give us a proper hostname, so use that.
 446   if [[ ${#hostkeys[@]} == 1 && -e ${hostkeys[0]} ]]; then
 447     luks_file=${hostkeys[0]}
 448   else
 449     echo "$0: error: no key for hostname at $luks_file" >&2
 450     exit 1
 451   fi
 452 fi
 453
 454 # # note, corresponding changes in /b/ds/keyscript-{on,off}
 455 if ifclass demohost; then
 456   lukspw=x
 457 elif [[ -e $luks_dir/$HOSTNAME ]]; then
 458   lukspw=$(cat $luks_dir/$HOSTNAME)
 459 else
 460   lukspw=$(cat $luks_dir/iank)
 461 fi
 462
 463
 464 first_root_crypt=$(root-cryptdev ${devs[0]})
 465
 466 # 1.5 x based on https://access.redhat.com/documentation/en-US/Red_Hat_Enterprise_Linux/7/html/Installation_Guide/sect-disk-partitioning-setup-x86.html#sect-custom-partitioning-x86
 467 swap_mib=$(( $(grep ^MemTotal: /proc/meminfo | \
 468                  awk '{print $2}') * 3/(${#devs[@]} * 2 ) / 1024 ))
 469
 470 mkdir -p /tmp/fai
 471 root_devs=()
 472 for dev in ${devs[@]}; do
 473   root_devs+=($(rootdev))
 474 done
 475 shopt -s nullglob
 476
 477 rm -f /mnt/root/root2-{fs,crypt}tab
 478 if $mkroot2; then
 479   if $partition; then
 480     echo $0: error: found partition=true but have mkroot2 arg
 481     exit 1
 482   fi
 483   for dev in ${devs[@]}; do
 484     luks_file=$luks_dir/host-amy
 485     lukspw=$(cat $luks_dir/amy)
 486     luks-setup $(root2dev)
 487     cat >>/mnt/root/root2-crypttab <<EOF
 488 $(root2-cryptname) $(root2dev)  $luks_file  discard,luks,initramfs
 489 EOF
 490   done
 491   bpart $(for dev in ${devs[@]}; do root2-cryptdev; done)
 492   bpart ${boot2_devs[@]}
 493   mkdir -p /mnt/root2 /mnt/boot2
 494   cat >>/mnt/root/root2-fstab <<EOF
 495 $(root2-cryptdev ${devs[0]}) /mnt/root2  btrfs  nofail,x-systemd.device-timeout=30s,x-systemd.mount-timeout=30s,noatime,subvolid=0$mopts  0 0
 496 ${boot2_devs[0]} /mnt/boot2  btrfs    nofail,x-systemd.device-timeout=30s,x-systemd.mount-timeout=30s,noatime,subvolid=0  0 0
 497 EOF
 498   exit 0
 499 fi
 500
 501
 502 if $partition; then
 503   ### begin wipefs
 504   if [[ ! $SPECIAL_DISK ]]; then
 505     for dev in ${devs[@]}; do
 506       # if we repartition to the same as an old partition,
 507       # we don't want any old fses hanging around.
 508       for (( i=1; i <= lastn; i++ )); do
 509         x=$(add-part $i)
 510         [[ -e $x ]] || continue
 511         count_down=10
 512         # wipefs has failed, manual run works, google suggests timing issue
 513         while ! wipefs -a $x; do
 514           sleep 2
 515           count_down=$((count_down - 1))
 516           (( count_down > 0 )) || exit 1
 517         done
 518       done
 519     done
 520   fi
 521   ### end wipefs
 522
 523
 524   # When we have 2 disks of at least 100g difference in size,
 525   # make an extra partition on the end of the bigger one.
 526   even_big_part=false
 527   even_diff_min=100000
 528   if $even_raid; then
 529     smalli=0
 530     bigi=1
 531     if (( $(dev-mib ${devs[0]}) >= $(dev-mib ${devs[1]}) )); then
 532       smalli=1
 533       bigi=0
 534     fi
 535     disk_mib=$(dev-mib ${devs[smalli]})
 536     even_big_dev=${devs[bigi]}
 537     even_big_mib=$(dev-mib $even_big_dev)
 538     if (( even_big_mib - disk_mib > even_diff_min )); then
 539       even_big_part=true
 540     fi
 541   fi
 542
 543   for dev in ${devs[@]}; do
 544     if [[ $SPECIAL_DISK ]]; then
 545       dev=$(devbyid $SPECIAL_DISK)
 546     fi
 547
 548     # parted will round up the disk size. Do -1 so we can have
 549     # fully 1MiB unit partitions for easy resizing of the last partition.
 550     # Otherwise we would pass in -0 for the end argument for the last partition.
 551     #
 552     # Note: parted print error output is expected. example:
 553     # Error: /dev/vda: unrecognised disk label
 554     if ! $even_raid; then
 555       disk_mib=$(dev-mib)
 556     fi
 557
 558     boot_part_mib=$(( boot_mib / ${#boot_devs[@]} ))
 559     boot2_part_mib=$(( boot2_mib / ${#boot_devs[@]} ))
 560     root2_part_mib=$(( root2_mib / ${#root_devs[@]} ))
 561     root_end=$(( disk_mib - root2_part_mib - swap_mib - boot_part_mib - boot2_part_mib ))
 562     root2_end=$(( root_end + root2_part_mib ))
 563     swap_end=$(( root2_end + swap_mib ))
 564     boot_end=$(( swap_end + boot_part_mib ))
 565
 566     parted -s $dev mklabel gpt
 567     # MiB because parted complains about alignment otherwise.
 568     pcmd="parted -a optimal -s -- $dev"
 569     # root partition, the main big one
 570     $pcmd mkpart primary ext3 524MiB ${root_end}MiB
 571     # without naming, systemd gives us misc errors like:
 572     # dev-disk-by\x2dpartlabel-primary.device: Dev dev-disk-by\x2dpartlabel-primary.device appeared twice
 573     $pcmd name $rootn root
 574     # root2 partition
 575     $pcmd mkpart primary ext3 ${root_end}MiB ${root2_end}MiB
 576     $pcmd name $root2n root2
 577     # normally a swap is type "linux-swap", but this is encrypted swap. using that
 578     # label will confuse systemd.
 579     # swap partition
 580     $pcmd mkpart primary "" ${root2_end}MiB ${swap_end}MiB
 581     $pcmd name $swapn swap
 582     # boot partition
 583     $pcmd mkpart primary "" ${swap_end}MiB ${boot_end}MiB
 584     $pcmd name $bootn boot
 585     # boot2 partition
 586     $pcmd mkpart primary "" ${boot_end}MiB ${disk_mib}MiB
 587     $pcmd name $boot2n boot2
 588     # uefi partition. efi sucks, half a gig, rediculous.
 589     $pcmd mkpart primary "fat32" 12MiB 524MiB
 590     $pcmd name $efin efi
 591     $pcmd set $efin esp on
 592     # note, this is shown here: https://support.system76.com/articles/bootloader/
 593     # but not mentioned https://wiki.archlinux.org/index.php/EFI_system_partition
 594     # probably not needed
 595     $pcmd set $bootn boot on
 596     $pcmd set $boot2n boot on
 597     # i only need a few k, but googling min size,
 598     # I found someone saying that gparted required
 599     # required at least 8 because of their hard drive cylinder size.
 600     # And 8 is still very tiny.
 601     # grub_ext partition
 602     $pcmd mkpart primary "ext2" 4MiB 12MiB
 603     $pcmd name $grub_extn grubext
 604     # gpt ubuntu cloud image uses ~4 mb for this partition. fai uses 1 MiB.
 605     # so, I use 3, whatever.
 606     # note: parted manual saying cheap flash media
 607     # should to start at 4.
 608     # biols grub partition
 609     $pcmd mkpart primary "" 1MiB 4MiB
 610     $pcmd name $bios_grubn biosgrub
 611     $pcmd set $bios_grubn bios_grub on
 612     if $even_big_part  && [[ $dev == "$even_big_dev" ]]; then
 613       $pcmd mkpart primary ext3 ${disk_mib}MiB ${even_big_mib}MiB
 614       $pcmd name $even_bign even_big
 615     fi
 616
 617     # the mkfs failed before on a vm, which prompted me to add
 618     # sleep .1
 619     # then it failed again on a physical machine
 620     # with:
 621     # Device /dev/disk/by-id/foo doesn't exist or access denied,
 622     # so I added a wait until it existed.
 623     # Then I added the mkfs.ext2, which claimed to succeed,
 624     # but then couldn't be found upon reboot. In that case we didn't
 625     # wait at all. So I've added a 3 second minimum wait.
 626     secs=0
 627     while [[ ! -e $(bios_grubdev) ]] && (( secs < 10 )); do
 628       sleep 1
 629       secs=$((secs +1))
 630     done
 631     sleep 3
 632
 633     mkfs.fat -F32 $(efidev)
 634
 635     if $even_big_part  && [[ $dev == "$even_big_dev" ]]; then
 636       luks-setup $(even_bigdev)
 637       mkfs.btrfs -f $(crypt-dev $(even_bigdev))
 638     fi
 639
 640     # Holds just a single file, rarely written, so
 641     # use ext2, like was often used for the /boot partition.
 642     # This exists because grub can only persist data to a non-cow fs.
 643     # And we use persisting a var in grub to do a one time boot.
 644     # We could pass the data on the kernel command line and persist it
 645     # to grubenv after booting, but that relies on the boot always succeeding.
 646     # This is just a bit more robust, and it could work for booting
 647     # into ipxe which can't persist data, if we ever got that working.
 648     mkfs.ext2 $(grub_extdev)
 649     luks-setup $(rootdev)
 650
 651     if [[ $SPECIAL_DISK ]]; then
 652       exit 0
 653     fi
 654   done
 655   ls -la /dev/btrfs-control # this was probably for debugging...
 656   sleep 1
 657   bpart $(for dev in ${devs[@]}; do root-cryptdev; done)
 658   bpart ${boot_devs[@]}
 659 else
 660   for dev in ${devs[@]}; do
 661     if [[ -e /dev/mapper/$(root-cryptname) ]]; then
 662       continue
 663     fi
 664     cryptsetup luksOpen $(rootdev) $(root-cryptname) \
 665                --key-file $luks_file
 666   done
 667   sleep 1
 668 fi
 669
 670
 671 if $wipe && [[ $DISTRO != debianbuster_bootstrap ]]; then
 672   # bootstrap distro doesn't use separate encrypted root.
 673   mount -o subvolid=0 $first_root_crypt /mnt
 674   # systemd creates subvolumes we want to delete.
 675   s=($(btrfs subvolume list --sort=-path /mnt |
 676          sed -rn "s#^.*path\s*(root_$DISTRO/\S+)\s*\$#\1#p"))
 677   for subvol in ${s[@]}; do btrfs subvolume delete /mnt/$subvol; done
 678   btrfs subvolume set-default 0 /mnt
 679   [[ ! -e /mnt/root_$DISTRO ]] || btrfs subvolume delete /mnt/root_$DISTRO
 680
 681   ## create subvols ##
 682   cd /mnt
 683
 684   btrfs subvolume create root_$DISTRO
 685
 686   # could set default subvol like this, but no reason to.
 687   # btrfs subvolume set-default \
 688     #       $(btrfs subvolume list . | grep "root_$DISTRO$" | awk '{print $2}') .
 689
 690   # For raid systems, cow allows for error correction, for non-raid systems,
 691   # protects root fs from having the plug pulled. Reprovisioning a root
 692   # subvol is not my favorite thing to do.
 693   # # no cow on the root filesystem. it's setup is fully scripted,
 694   # # if it's messed up, we will just recreated it,
 695   # # and we can get better perf with this.
 696   # # I can't remember exactly why, but this is preferable to mounting with
 697   # # -o nodatacow, I think because subvolumes inherit that.
 698   # chattr -Rf +C root_$DISTRO
 699   cd /
 700   umount /mnt
 701 fi
 702
 703 mount -o subvolid=0 $first_boot_dev /mnt
 704 cd /mnt
 705 btrfs subvolume set-default 0 /mnt # already default, just ensuring it.
 706
 707 # for libreboot systems. grub2 only reads from subvolid=0
 708 mkdir -p /mnt/grub2
 709 cp $FAI/distro-install-common/libreboot_grub.cfg /mnt/grub2
 710
 711 if [[ $DISTRO == debianbuster_bootstrap ]]; then
 712   # this is just convenience for the libreboot_grub config
 713   # so we can glob the other ones easier.
 714   boot_vol=$DISTRO
 715 else
 716   boot_vol=boot_$DISTRO
 717 fi
 718 if $wipe && [[ -e /mnt/$boot_vol ]]; then
 719   btrfs subvolume delete /mnt/$boot_vol
 720 fi
 721 if [[ ! -e /mnt/$boot_vol ]]; then
 722   btrfs subvolume create $boot_vol
 723 fi
 724 cd /
 725 umount /mnt
 726 ## end create subvols ##
 727
 728 dev=${boot_devs[0]}
 729 mount $first_grub_extdev /mnt
 730 grub-editenv /mnt/grubenv set did_fai_check=true
 731 grub-editenv /mnt/grubenv set last_boot=/$boot_vol
 732 umount /mnt
 733
 734 fstabstd=x-systemd.device-timeout=30s,x-systemd.mount-timeout=30s
 735 if [[ $DISTRO == debianbuster_bootstrap ]]; then
 736   cat > /tmp/fai/fstab <<EOF
 737 $first_boot_dev  /  btrfs  noatime,subvol=$boot_vol  0 0
 738 $first_efi  /boot/efi  vfat          nofail,$fstabstd  0 0
 739 EOF
 740   cat >/tmp/fai/disk_var.sh <<EOF
 741 BOOT_DEVICE="${short_devs[@]}"
 742 ROOT_PARTITION=$first_boot_dev
 743 EOF
 744 else
 745   # note, fai creates the mountpoints listed here
 746   cat > /tmp/fai/fstab <<EOF
 747 $first_root_crypt  /  btrfs $fstabstdopts,noatime,subvol=root_$DISTRO$mopts  0 0
 748 $first_root_crypt  /mnt/root  btrfs  nofail,$fstabstd,noatime,subvolid=0$mopts  0 0
 749 $first_boot_dev  /boot  btrfs        nofail,$fstabstd,noatime,subvol=$boot_vol  0 0
 750 $first_efi  /boot/efi  vfat          nofail,$fstabstd  0 0
 751 $first_boot_dev  /mnt/boot  btrfs    nofail,$fstabstd,noatime,subvolid=0  0 0
 752 EOF
 753   swaps=()
 754   rm -f /tmp/fai/crypttab
 755   for dev in ${devs[@]}; do
 756     swaps+=($(swap-cryptname))
 757     cat >>/tmp/fai/crypttab <<EOF
 758 $(root-cryptname) $(rootdev)  none  keyscript=/root/keyscript,discard,luks,initramfs
 759 $(swap-cryptname) $(swapdev)  /dev/urandom  swap,cipher=aes-xts-plain64,size=256,hash=ripemd160
 760 EOF
 761     cat >> /tmp/fai/fstab <<EOF
 762 $(swap-cryptdev)  none  swap  nofail,$fstabstd,sw  0 0
 763 EOF
 764   done
 765
 766   # fai would do this:
 767   #BOOT_DEVICE=\${BOOT_DEVICE:-"${devs[0]}"}
 768
 769   # note: swaplist seems to do nothing.
 770   cat >/tmp/fai/disk_var.sh <<EOF
 771 BOOT_DEVICE="${short_devs[@]}"
 772 BOOT_PARTITION=\${BOOT_PARTITION:-$first_boot_dev}
 773 # ROOT_PARTITIONS is added by me, used in arch setup.
 774 ROOT_PARTITIONS="${root_devs[@]}"
 775 ROOT_PARTITION=\${ROOT_PARTITION:-$first_root_crypt}
 776 SWAPLIST=\${SWAPLIST:-"${swaps[@]}"}
 777 EOF
 778
 779   if [[ $HOSTNAME == kd ]]; then
 780     # note, having these with keyscript and initramfs causes a luks error in fai.log,
 781     # but it is safely ignorable and gets us the ability to just type our password
 782     # in once at boot. A downside is that they are probably needed to be plugged in to boot.
 783     cat >>/tmp/fai/crypttab <<EOF
 784 crypt_dev_ata-Samsung_SSD_870_QVO_8TB_S5VUNG0N900656V-part${even_bign} /dev/disk/by-id/ata-Samsung_SSD_870_QVO_8TB_S5VUNG0N900656V-part7  none  keyscript=decrypt_keyctl,discard,luks,initramfs
 785 crypt_dev_ata-TOSHIBA_MD04ACA500_84R2K773FS9A-part1 /dev/disk/by-id/ata-TOSHIBA_MD04ACA500_84R2K773FS9A-part1  none  keyscript=decrypt_keyctl,discard,luks,initramfs
 786 crypt_dev_ata-ST6000DM001-1XY17Z_Z4D29EBL-part1 /dev/disk/by-id/ata-ST6000DM001-1XY17Z_Z4D29EBL-part1  none  keyscript=decrypt_keyctl,discard,luks,initramfs
 787 EOF
 788     cat >> /tmp/fai/fstab <<EOF
 789 # r7 = root partition7. it isnt actually #7 anymore, not a great name, but whatever
 790 /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
 791 /dev/mapper/crypt_dev_ata-TOSHIBA_MD04ACA500_84R2K773FS9A-part1  /mnt/rust1  btrfs  nofail,$fstabstd,noatime,compress=zstd,subvolid=0  0 0
 792 /dev/mapper/crypt_dev_ata-ST6000DM001-1XY17Z_Z4D29EBL-part1  /mnt/rust2  btrfs  nofail,$fstabstd,noatime,compress=zstd,subvolid=0  0 0
 793 EOF
 794   fi
 795
 796 fi
 797
 798 # initial setup of extra data fs, mounted,
 799 # btrfs subvol create nocow
 800 # chattr +C nocow
 801 # chown iank.iank nocow
 802