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 set -eE -o pipefail
  19 trap 'echo "$0:$LINENO:error: \"$BASH_COMMAND\" returned $?" >&2' ERR
  20
  21 # # fai's setup-storage won't do btrfs on luks,
  22 # # so we do it ourself :)
  23 # inspiration taken from files in fai-setup-storage package
  24
  25
  26 skiptask partition || ! type skiptask # for running not in fai
  27
  28 #### begin configuration
  29
  30 rootn=1
  31 swapn=2
  32 bootn=3
  33 # ext partition so grub can write persistent variables,
  34 # so it can do a one time boot.
  35 grub_extn=4
  36 # bios boot partition,
  37 # https://wiki.archlinux.org/index.php/GRUB
  38 bios_grubn=5
  39 lastn=$bios_grubn
  40 boot_mib=10000
  41
  42
  43 ##### end configuration
  44
  45
  46 add-part() { # add partition suffix to $dev
  47     local d ret
  48     if [[ $# == 1 ]]; then
  49         d=$dev
  50         part=$1
  51     else
  52         d=$1
  53         part=$2
  54     fi
  55     if [[ $d == /dev/disk/by-id/* ]]; then
  56         ret=$d-part$part
  57     else
  58         ret=$d$part
  59     fi
  60     echo $ret
  61 }
  62
  63 bootdev() { add-part $@ $bootn; }
  64 rootdev() { add-part $@ $rootn; }
  65 swapdev() { add-part $@ $swapn; }
  66 grub_extdev() { add-part $@ $grub_extn; }
  67 # Commented because it's not used, but left because it
  68 # finishes the pattern and if we ever do need to use it, it's here.
  69 #bios_grubdev() { add-part $@ $bios_grubn; }
  70
  71 crypt-dev() { echo /dev/mapper/crypt_dev_${1##*/}; }
  72 crypt-name() { echo crypt_dev_${1##*/}; }
  73 root-cryptdev() { crypt-dev $(rootdev $@); }
  74 swap-cryptdev() { crypt-dev $(swapdev $@); }
  75 root-cryptname() { crypt-name $(rootdev $@); }
  76 swap-cryptname() { crypt-name $(swapdev $@); }
  77
  78
  79 ##### end function defs
  80
  81 if ifclass REPARTITION;then
  82     partition=true # force a full wipe
  83 else
  84     partition=false # change to true to force a full wipe
  85 fi
  86
  87
  88
  89 hdds=()
  90 ssds=()
  91 cd /sys/block
  92 for disk in [sv]d[a-z]; do
  93     case $(cat $disk/queue/rotational) in
  94         0) ssds+=(/dev/$disk) ;;
  95         1) hdds+=(/dev/$disk) ;;
  96         *) echo "$0: error: unknown /sys/block/$disk/queue/rotational: \
  97 $(cat $disk/queue/rotational)"; exit 1 ;;
  98     esac
  99 done
 100
 101 # install all ssds, or if there are none, all hdds
 102 if ! ifclass ROTATIONAL && (( ${#ssds[@]} > 0 )); then
 103     short_devs=( ${ssds[@]} )
 104 else
 105     short_devs=( ${hdds[@]} )
 106 fi
 107
 108 # check if the partitions exist have the right filesystems
 109 #blkid="$(blkid -s TYPE)"
 110 for dev in ${short_devs[@]}; do
 111     if $partition; then break; fi
 112     y=$(readlink -f $dev)
 113     x=($y[0-9])
 114     [[ ${#x[@]}  == "${lastn}" ]] || partition=true
 115     for (( i=1; i <= lastn; i++ )); do
 116         [[ -e ${dev}$i ]] || partition=true
 117     done
 118     # On one system, blkid is missing some partitions.
 119     # maybe we need a flag, like FUZZY_BLKID or something, so we
 120     # can check that at least some exist.
 121     # for x in "`rootdev`: TYPE=\"crypto_LUKS\"" "`bootdev`: TYPE=\"btrfs\""; do
 122     #     echo "$blkid" | grep -Fx "$x" &>/dev/null || partition=true
 123     # done
 124 done
 125
 126 if $partition && ifclass PARTITION_PROMPT; then
 127     echo "Press any key except ctrl-c to continue and partition these drives:"
 128     echo "  ${short_devs[*]}"
 129     read -r
 130 fi
 131
 132 devs=()
 133 shopt -s extglob
 134 for short_dev in ${short_devs[@]}; do
 135     devs+=($(devbyid $short_dev))
 136 done
 137
 138
 139 first=false
 140 boot_devs=()
 141 for dev in ${devs[@]}; do
 142     if ifclass frodo; then
 143         # I ran into a machine where the bios doesn't know about some disks,
 144         # so 1st stage of grub also doesn't know about them.
 145         # Also, grub does not support mounting degraded btrfs as far as
 146         # I can tell with some googling.
 147         # From within an arch install env, I could detect them by noting
 148         # their partitions were mixed with the next disk in /dev/disk/by-path,
 149         # and I have mixed model disks, and I could see the 8 models which showed
 150         # up in the bios, and thus see which 2 models were missing.
 151         # hdparm -I /dev/sdh will give model info in linux.
 152         # However, in fai on jessie, /dev/disk/by-path dir doesn't exist,
 153         # and I don't see another way, so I'm hardcoding them.
 154         # We still put grub on them and partition them the same, for uniformity
 155         # and in case they get moved to a system that can recognize them,
 156         # we just exclude them from the boot filesystem.
 157         cd /dev/disk/by-id/
 158         bad_disk=false
 159         for id in ata-TOSHIBA_MD04ACA500_8539K4TQFS9A \
 160                       ata-TOSHIBA_MD04ACA500_Y5IFK6IJFS9A; do
 161             if [[ $(readlink -f $id) == "$(readlink -f $dev)" ]]; then
 162                 bad_disk=true
 163                 break
 164             fi
 165         done
 166         $bad_disk || boot_devs+=(`bootdev`)
 167     else
 168         boot_devs+=(`bootdev`)
 169     fi
 170     if [[ $boot_devs && $first ]]; then
 171         first_grub_extdev=`grub_extdev`
 172         first=false
 173     fi
 174 done
 175
 176 if ifclass RAID0 || (( ${#boot_devs[@]} < 4 )); then
 177     raid_level=0
 178 else
 179     raid_level=10
 180     # need double the space if we are raid 10, and then
 181     # might as well give some extra.
 182     boot_mib=$((boot_mib * 3))
 183 fi
 184
 185
 186
 187 if [[ ! $DISTRO ]]; then
 188     if ifclass STABLE_BOOTSTRAP; then
 189         DISTRO=debianstable_bootstrap
 190     elif ifclass STRETCH64; then
 191         DISTRO=debiantesting
 192     elif ifclass STABLE; then
 193         DISTRO=debianstable
 194     elif ifclass XENIAL64; then
 195         DISTRO=ubuntuxenial
 196     elif ifclass BELENOS64; then
 197         DISTRO=trisquelbelenos
 198     else
 199         echo "PARTITIONER ERROR: no distro class/var set" >&2
 200         exit 1
 201     fi
 202 fi
 203 first_boot_dev=${boot_devs[0]}
 204
 205
 206 bpart() { # btrfs a partition
 207     case $raid_level in
 208         0) mkfs.btrfs -f $@ ;;
 209         10) mkfs.btrfs -f -m raid10 -d raid10 $@ ;;
 210     esac
 211 }
 212
 213
 214 # keyfiles generated like:
 215 # head -c 2048 /dev/urandom | od | s dd of=/q/root/luks/host-demohost
 216 luks_dir=${LUKS_DIR:-/var/lib/fai/config/distro-install-common/luks}
 217
 218 if [[ ! -e $luks_dir/host-$HOSTNAME ]]; then
 219     echo "$0: error: no key for hostname at $luks_dir/host-$HOSTNAME" >&2
 220     exit 1
 221 fi
 222
 223 if ifclass tp; then
 224     lukspw=$(cat $luks_dir/traci)
 225 else
 226     lukspw=$(cat $luks_dir/ian)
 227 fi
 228 if ifclass demohost; then
 229     lukspw=x
 230 fi
 231
 232
 233 first_root_crypt=$(root-cryptdev ${devs[0]})
 234
 235 # 1.5 x based on https://access.redhat.com/documentation/en-US/Red_Hat_Enterprise_Linux/7/html/Installation_Guide/sect-disk-partitioning-setup-x86.html#sect-custom-partitioning-x86
 236 swap_mib=$(( $(grep ^MemTotal: /proc/meminfo | \
 237                    awk '{print $2}') * 3/(${#devs[@]} * 2 ) / 1024 ))
 238
 239 mkdir -p /tmp/fai
 240 root_devs=()
 241 for dev in ${devs[@]}; do
 242     root_devs+=(`rootdev`)
 243 done
 244 shopt -s nullglob
 245 if $partition; then
 246     for dev in ${devs[@]}; do
 247         # if we repartition to the same as an old partition,
 248         # we don't want any old fses hanging around.
 249         for (( i=1; i <= lastn; i++ )); do
 250             x=$(add-part $dev $i)
 251             [[ -e $x ]] || continue
 252             count_down=10
 253             # wipefs has failed, manual run works, google suggests timing issue
 254             while ! wipefs -a $x; do
 255                 sleep 2
 256                 count_down=$((count_down - 1))
 257                 (( count_down > 0 )) || exit 1
 258             done
 259         done
 260     done
 261     for dev in ${devs[@]}; do
 262         # parted will round up the disk size. Do -1 so we can have
 263         # fully 1MiB unit partitions for easy resizing of the last partition.
 264         # Otherwise we would pass in -0 for the end argument for the last partition.
 265         #
 266         # parted print error output is expected. example:
 267         # Error: /dev/vda: unrecognised disk label
 268         disk_mib=$(( $(parted -m $dev unit MiB print | \
 269                            sed -nr "s#^/dev/[^:]+:([0-9]+).*#\1#p") - 1))
 270         root_end=$(( disk_mib - swap_mib - boot_mib /  ${#boot_devs[@]} ))
 271         swap_end=$(( root_end + swap_mib))
 272
 273         parted -s $dev mklabel gpt
 274         # MiB because parted complains about alignment otherwise.
 275         pcmd="parted -a optimal -s -- $dev"
 276         $pcmd mkpart primary "ext3" 12MiB ${root_end}MiB
 277         $pcmd mkpart primary "linux-swap" ${root_end}MiB ${swap_end}MiB
 278         $pcmd mkpart primary "" ${swap_end}MiB ${disk_mib}MiB
 279         # i only need a few k, but googling min size,
 280         # I found someone saying that gparted required
 281         # required at least 8 because of their hard drive cylinder size.
 282         # And 8 is still very tiny.
 283         $pcmd mkpart primary "ext2" 4MiB 12MiB
 284         # gpt ubuntu cloud image uses ~4 mb for this partition. fai uses 1 MiB.
 285         # so, I use 3, whatever.
 286         # note: parted manual saying cheap flash media
 287         # should to start at 4.
 288         $pcmd mkpart primary "" 1MiB 4MiB
 289         $pcmd set $bios_grubn bios_grub on
 290         $pcmd set $bootn boot on # generally not needed on modern systems
 291         # the mkfs failed before on a vm, which prompted me to add
 292         # sleep .1
 293         # then it failed again on a physical machine
 294         # with:
 295         # Device /dev/disk/by-id/foo doesn't exist or access denied,
 296         # so I added a wait until it existed.
 297         # Then I added the mkfs.ext2, which claimed to succeed,
 298         # but then couldn't be found upon reboot. In that case we didn't
 299         # wait at all. So I've added a 3 second minimum wait.
 300         sleep 3
 301         secs=0
 302         while [[ ! -e `rootdev` ]] && (( secs < 10 )); do
 303             sleep 1
 304             secs=$((secs +1))
 305         done
 306         # Holds just a single file, rarely written, so
 307         # use ext2, like was often used for the /boot partition.
 308         # This exists because grub can only persist data to a non-cow fs.
 309         # And we use persisting a var in grub to do a one time boot.
 310         # We could pass the data on the kernel command line and persist it
 311         # to grubenv after booting, but that relies on the boot always succeeding.
 312         # This is just a bit more robust, and it could work for booting
 313         # into ipxe which can't persist data, if we ever got that working.
 314         mkfs.ext2 `grub_extdev`
 315         yes YES | cryptsetup luksFormat `rootdev` $luks_dir/host-$HOSTNAME \
 316                              -c aes-cbc-essiv:sha256 -s 256 || [[ $? == 141 ]]
 317         yes "$lukspw" | \
 318             cryptsetup luksAddKey --key-file $luks_dir/host-$HOSTNAME \
 319                        `rootdev` || [[ $? == 141 ]]
 320         # background: Keyfile and password are treated just
 321         # like 2 ways to input a passphrase, so we don't actually need to have
 322         # different contents of keyfile and passphrase, but it makes some
 323         # security sense to a really big randomly generated passphrase
 324         # as much as possible, so we have both.
 325         #
 326         # This would remove the keyfile.
 327         #    yes 'test' | cryptsetup luksRemoveKey /dev/... \
 328             #                            /key/file || [[ $? == 141 ]]
 329
 330         cryptsetup luksOpen `rootdev` `root-cryptname` \
 331                    --key-file $luks_dir/host-$HOSTNAME
 332     done
 333     ls -la /dev/btrfs-control # this was probably for debugging...
 334     sleep 1
 335     bpart $(for dev in ${devs[@]}; do root-cryptdev; done)
 336     bpart ${boot_devs[@]}
 337 else
 338     for dev in ${devs[@]}; do
 339         mkfs.ext2 `grub_extdev`
 340         cryptsetup luksOpen `rootdev` `root-cryptname` \
 341                    --key-file $luks_dir/host-$HOSTNAME
 342     done
 343     sleep 1
 344 fi
 345
 346
 347 if [[ $DISTRO != debianstable_bootstrap ]]; then
 348     # bootstrap distro doesn't use separate encrypted root.
 349     mount -o subvolid=0 $first_root_crypt /mnt
 350     # systemd creates subvolumes we want to delete.
 351     s=($(btrfs subvolume list --sort=-path /mnt |
 352              sed -rn "s#^.*path\s*(root_$DISTRO/\S+)\s*\$#\1#p"))
 353     for subvol in ${s[@]}; do btrfs subvolume delete /mnt/$subvol; done
 354     btrfs subvolume set-default 0 /mnt
 355     [[ ! -e /mnt/root_$DISTRO ]] || btrfs subvolume delete /mnt/root_$DISTRO
 356
 357     ## create subvols ##
 358     cd /mnt
 359
 360     btrfs subvolume create root_$DISTRO
 361     [[ -e q ]] || btrfs subvolume create q
 362     chown root:1000 q
 363
 364     mkdir -p /mnt/root_$DISTRO/boot
 365     for x in q/a q/i; do
 366         mkdir -p $x
 367         chown 1000:1000 $x
 368         chmod 755 $x
 369     done
 370     # could set default like this, but no reason to.
 371     # btrfs subvolume set-default \
 372         #       $(btrfs subvolume list . | grep "root_$DISTRO$" | awk '{print $2}') .
 373
 374     # no cow on the root filesystem. it's setup is fully scripted,
 375     # (immutable in buzzwords). if it messes up, we will just recreated it,
 376     # and we can get better perf with this.
 377     # I can't remember exactly why, but this is preferable to mounting with
 378     # -o nodatacow, I think because subvolumes inherit that.
 379     chattr -Rf +C root_$DISTRO
 380     cd /
 381     umount /mnt
 382 fi
 383
 384 mount -o subvolid=0 $first_boot_dev /mnt
 385 cd /mnt
 386 btrfs subvolume set-default 0 /mnt # already default, just ensuring it.
 387
 388 # for libreboot systems.
 389 mkdir -p /mnt/grub2
 390 cp $FAI/distro-install-common/libreboot_grub.cfg /mnt/grub2
 391
 392 if [[ $DISTRO == debianstable_bootstrap ]]; then
 393     # this is just convenience for the libreboot_grub config
 394     # so we can glob the other ones easier.
 395     boot_vol=$DISTRO
 396 else
 397     boot_vol=boot_$DISTRO
 398 fi
 399 [[ ! -e /mnt/$boot_vol ]] || btrfs subvolume delete /mnt/$boot_vol
 400 btrfs subvolume create $boot_vol
 401 cd /
 402 umount /mnt
 403 ## end create subvols ##
 404
 405 dev=${boot_devs[0]}
 406 mount $first_grub_extdev /mnt
 407 grub-editenv /mnt/grubenv set did_fai_check=true
 408 grub-editenv /mnt/grubenv set last_boot=/$boot_vol
 409 umount /mnt
 410
 411 if [[ $DISTRO == debianstable_bootstrap ]]; then
 412     cat > /tmp/fai/fstab <<EOF
 413 $first_boot_dev  /  btrfs  noatime,subvol=$boot_vol  0 0
 414 EOF
 415     cat >/tmp/fai/disk_var.sh <<EOF
 416 BOOT_DEVICE="${short_devs[@]}"
 417 ROOT_PARTITION=$first_boot_dev
 418 EOF
 419 else
 420     # note, the mount point /a seems to get automatically created somewhere
 421     cat > /tmp/fai/fstab <<EOF
 422 $first_root_crypt  /  btrfs  noatime,subvol=root_$DISTRO  0 0
 423 $first_root_crypt  /q  btrfs  noatime,subvol=q  0 0
 424 /q/a  /a  none  bind  0 0
 425 $first_boot_dev  /boot  btrfs  noatime,subvol=$boot_vol  0 0
 426 EOF
 427
 428     swaps=()
 429     for dev in ${devs[@]}; do
 430         swaps+=(`swap-cryptname`)
 431         cat >>/tmp/fai/crypttab <<EOF
 432 `root-cryptname`  `rootdev`  none  keyscript=/root/keyscript,discard,luks
 433 `swap-cryptname` `swapdev`  /dev/urandom  swap,cipher=aes-xts-plain64,size=256,hash=ripemd160
 434 EOF
 435         cat >> /tmp/fai/fstab <<EOF
 436 `swap-cryptdev`  none  swap  sw  0 0
 437 EOF
 438     done
 439
 440     # fai would do this:
 441     #BOOT_DEVICE=\${BOOT_DEVICE:-"${devs[0]}"}
 442
 443     # note: swaplist seems to do nothing.
 444     cat >/tmp/fai/disk_var.sh <<EOF
 445 BOOT_DEVICE="${short_devs[@]}"
 446 BOOT_PARTITION=\${BOOT_PARTITION:-$first_boot_dev}
 447 # ROOT_PARTITIONS is added by me, used in arch setup.
 448 ROOT_PARTITIONS="${root_devs[@]}"
 449 ROOT_PARTITION=\${ROOT_PARTITION:-$first_root_crypt}
 450 SWAPLIST=\${SWAPLIST:-"${swaps[@]}"}
 451 EOF
 452 fi