X-Git-Url: https://iankelling.org/git/?a=blobdiff_plain;f=fai%2Fconfig%2Fhooks%2Fpartition.DEFAULT;h=aa7440968b8eaca3cde4c43672a69ad5825441c0;hb=1d331faf4f315ae3356ceecb648170d3d74473df;hp=c45c4dc67a0917002d3393d6a84b3dd65d354e66;hpb=dbca9252a2039a851af5358ff866478135ae3ccc;p=automated-distro-installer diff --git a/fai/config/hooks/partition.DEFAULT b/fai/config/hooks/partition.DEFAULT index c45c4dc..aa74409 100755 --- a/fai/config/hooks/partition.DEFAULT +++ b/fai/config/hooks/partition.DEFAULT @@ -1,72 +1,184 @@ #!/bin/bash -x set -eE -o pipefail -trap 'echo "$0:$LINENO:error: \"$BASH_COMMAND\" returned $?"' ERR +trap 'echo "$0:$LINENO:error: \"$BASH_COMMAND\" returned $?" >&2' ERR # # fai's setup-storage won't do btrfs on luks, # # so we do it ourself :) -skiptask partition ||: # for running out of fai +skiptask partition || ! type skiptask # for running not in fai #### begin configuration -bootn=1 +rootn=1 swapn=2 -rootn=3 +bootn=3 bios_grubn=4 -boot_end=804 -lastn=$bios_grubn +boot_mib=1500 + + +##### end configuration + + +add-part() { + local d ret + if [[ $# == 1 ]]; then + d=$dev + part=$1 + else + d=$1 + part=$2 + fi + if [[ $d == /dev/disk/by-id/* ]]; then + ret=$d-part$part + else + ret=$d$part + fi + echo $ret +} + +bootdev() { add-part $@ $bootn; } +rootdev() { add-part $@ $rootn; } +swapdev() { add-part $@ $swapn; } +bios_grubdev() { add-part $@ $bios_grubn; } + +crypt-dev() { echo /dev/mapper/crypt_dev_${1##*/}; } +crypt-name() { echo crypt_dev_${1##*/}; } +root-cryptdev() { crypt-dev $(rootdev $@); } +swap-cryptdev() { crypt-dev $(swapdev $@); } +root-cryptname() { crypt-name $(rootdev $@); } +swap-cryptname() { crypt-name $(swapdev $@); } + -if ifclass VM; then - d=vd +##### end function defs + +if ifclass REPARTITION;then + partition=true # force a full wipe else - d=sd + partition=false # change to true to force a full wipe fi -letters=() -if ifclass TWO_DISK; then - letters=(a b) -elif ifclass ONE_DISK; then - letters=(a) -elif ifclass MANY_DISK; then - for dev in /dev/${d}?; do letters+=(${dev#/dev/${d}}); done +lastn=$bios_grubn + + +hdds=() +ssds=() +cd /sys/block +for disk in [sv]d[a-z]; do + case $(cat $disk/queue/rotational) in + 0) ssds+=(/dev/$disk) ;; + 1) hdds+=(/dev/$disk) ;; + *) echo "$0: error: unknown /sys/block/$disk/queue/rotational: \ +$(cat $disk/queue/rotational)"; exit 1 ;; + esac +done + +# install all ssds, or if there are none, all hdds +if (( ${#ssds[@]} > 0 )); then + short_devs=( ${ssds[@]} ) else - exit 1 + short_devs=( ${hdds[@]} ) fi -##### end configuration +# check if the partitions exist have the right filesystems +#blkid="$(blkid -s TYPE)" +for dev in ${short_devs[@]}; do + ! $partition || break + y=$(readlink -f $dev) + x=($y[0-9]) + [[ ${#x[@]} == "${lastn}" ]] || partition=true + for (( i=1; i <= lastn; i++ )); do + [[ -e ${dev}$i ]] || partition=true + done + # On one system, blkid is missing some partitions. + # maybe we need a flag, like FUZZY_BLKID or something, so we + # can check that at least some exist. + # for x in "`rootdev`: TYPE=\"crypto_LUKS\"" "`bootdev`: TYPE=\"btrfs\""; do + # echo "$blkid" | grep -Fx "$x" &>/dev/null || partition=true + # done +done + +if $partition && ifclass PARTITION_PROMPT; then + echo "Press any key except ctrl-c to continue and partition these drives:" + echo " ${short_devs[@]}" + read +fi -bpart() { - dev_n=$1 - case ${#@} in - [1-3]) mkfs.btrfs -f $@ ;; - [4-9]*|[1-3]?*) mkfs.btrfs -f -m raid10 -d raid10 $@ ;; - esac -} +devs=() +shopt -s extglob +for short_dev in ${short_devs[@]}; do + devs+=($(devbyid $short_dev)) +done -devs=(${letters[@]/#//dev/${d}}) -crypt_devs=(${letters[@]/#//dev/mapper/crypt_dev_${d}}) -partition=true -# somewhat crude detection of whether to partition +boot_devs=() for dev in ${devs[@]}; do - x=($dev[0-9]) - [[ ${#x[@]} == ${lastn} ]] || partition=true - for (( i=1; i <= $lastn; i++ )); do - [[ -e ${dev}$i ]] || partition=true - done - for part in $dev$rootn $dev$bootn; do - # type tells us it's not totally blank - blkid | grep "^${part}:.*TYPE=" &>/dev/null || partition=true - done + if ifclass frodo; then + # I ran into a machine where the bios doesn't know about some disks, + # so 1st stage of grub also doesn't know about them. + # Also, grub does not support mounting degraded btrfs as far as + # I can tell with some googling. + # From within an arch install env, I could detect them by noting + # their partitions were mixed with the next disk in /dev/disk/by-path, + # and I have mixed model disks, and I could see the 8 models which showed + # up in the bios, and thus see which 2 models were missing. + # hdparm -I /dev/sdh will give model info in linux. + # However, in fai on jessie, /dev/disk/by-path dir doesn't exist, + # and I don't see another way, so I'm hardcoding them. + # We still put grub on them and partition them the same, for uniformity + # and in case they get moved to a system that can recognize them, + # we just exclude them from the boot filesystem. + cd /dev/disk/by-id/ + bad_disk=false + for id in ata-TOSHIBA_MD04ACA500_8539K4TQFS9A \ + ata-TOSHIBA_MD04ACA500_Y5IFK6IJFS9A; do + if [[ $(readlink -f $id) == "$(readlink -f $dev)" ]]; then + bad_disk=true + break + fi + done + $bad_disk || boot_devs+=(`bootdev`) + else + boot_devs+=(`bootdev`) + fi done -#partition=true # for temporarily override +if [[ ! $DISTRO ]]; then + if ifclass STABLE; then + DISTRO=debianstable + else + DISTRO=debiantesting + fi +fi + + + +case ${#boot_devs[@]} in + # need double the space if we are raid 10, and then + # might as well give some extra overhead. + [4-9]*|[1-3]?*) boot_mib=$((boot_mib * 3)) ;; +esac + + +bpart() { # btrfs a partition + case ${#@} in + [1-3]) mkfs.btrfs -f $@ ;; + [4-9]*|[1-3]?*) mkfs.btrfs -f -m raid10 -d raid10 $@ ;; + esac +} + +first_boot_dev=${boot_devs[0]} # keyfiles generated like: # head -c 2048 /dev/urandom | od | s dd of=/q/root/luks/host-demohost luks_dir=${LUKS_DIR:-/var/lib/fai/config/distro-install-common/luks} + +if [[ ! -e $luks_dir/host-$HOSTNAME ]]; then + echo "$0: error: no key for hostname at $luks_dir/host-$HOSTNAME" >&2 + exit 1 +fi + if ifclass tp; then lukspw=$(cat $luks_dir/traci) else @@ -77,130 +189,175 @@ if ifclass demohost; then fi -crypt=/dev/mapper/crypt_dev_${d##/dev/}a$rootn +first_root_crypt=$(root-cryptdev ${devs[0]}) - -if ifclass frodo; then - # next upgrade is prolly 16 gigs of memory, across ~8 devices - swap_end=$((3500 + boot_end)) -else - # 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 - swap_end=$(( $(grep ^MemTotal: /proc/meminfo| awk '{print $2}') * 3/(${#devs[@]} * 2 ) / 1000 + boot_end )) -fi +bios_grubn=4 +# 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 +swap_mib=$(( $(grep ^MemTotal: /proc/meminfo | \ + awk '{print $2}') * 3/(${#devs[@]} * 2 ) / 1024 )) mkdir -p /tmp/fai +root_devs=() +for dev in ${devs[@]}; do + root_devs+=(`rootdev`) +done shopt -s nullglob if $partition; then for dev in ${devs[@]}; do - for x in $dev[0-9]; do wipefs -a $x; done + for x in $dev[0-9]; do + count_down=10 + # wipefs has failed, manual run works, google suggests timing issue + while ! wipefs -a $x; do + sleep 2 + count_down=$((count_down - 1)) + (( count_down > 0 )) || exit 1 + done + done done for dev in ${devs[@]}; do - parted -s $dev mklabel gpt - # gpt ubuntu cloud image uses ~4. fai uses 1 MiB. ehh, i'll do 4. - # also, using MB instead of MiB causes complains about alignment. - parted -s $dev mkpart primary "ext3" 4MB ${boot_end}MiB - parted -s $dev mkpart primary "linux-swap" ${boot_end}MiB ${swap_end}MiB - parted -s -- $dev mkpart primary "" ${swap_end}MiB -0 - parted -s $dev mkpart primary "" 1MiB 4MiB - parted -s $dev set $bios_grubn bios_grub on - parted -s $dev set $bootn boot on # generally not needed on modern systems - # the mkfs failed randomly on a vm, so I threw a sleep in here. - sleep .1 - mkfs.ext4 -F ${dev}1 - # 3 is device which simply holds a key for the 4's, - # so we can unlock multi-device btrfs fs with 1 manually entered passphrase. - # - # Background: It's of course possible modify the initramfs to - # put the input from a passphrase prompt into a variable and use - # it to unlock multiple devices, but that would require figuring - # more things out. + # parted will round up the disk size. Do -1 so we can have + # fully 1MiB unit partitions for easy resizing of the last partition. + # Otherwise we would pass in -0 for the end argument for the last partition. # - for luks_dev in ${dev}3; do - yes YES | cryptsetup luksFormat $luks_dev $luks_dir/host-$HOSTNAME \ - -c aes-cbc-essiv:sha256 -s 256 || [[ $? == 141 ]] - yes "$lukspw" | \ - cryptsetup luksAddKey --key-file $luks_dir/host-$HOSTNAME \ - $luks_dev || [[ $? == 141 ]] - # background: Keyfile and password are treated just - # like 2 ways to input a passphrase, so we don't actually need to have - # different contents of keyfile and passphrase, but it makes some - # security sense to a really big randomly generated passphrase - # as much as possible, so we have both. - # - # This would remove the keyfile. - # yes 'test' | cryptsetup luksRemoveKey /dev/... \ - # /key/file || [[ $? == 141 ]] - - cryptsetup luksOpen $luks_dev crypt_dev_${luks_dev##/dev/} \ - --key-file $luks_dir/host-$HOSTNAME + # parted print error output is expected. example: + # Error: /dev/vda: unrecognised disk label + disk_mib=$(( $(parted -m $dev unit MiB print | \ + sed -nr "s#^/dev/[^:]+:([0-9]+).*#\1#p") - 1)) + root_end=$(( disk_mib - swap_mib - boot_mib / ${#boot_devs[@]} )) + swap_end=$(( root_end + swap_mib)) + + parted -s $dev mklabel gpt + # gpt ubuntu cloud image uses ~4. fai uses 1 MiB. + # I read something in the parted manual saying cheap flash media + # likes to start at 4. + # MiB because parted complains about alignment otherwise. + pcmd="parted -a optimal -s -- $dev" + $pcmd mkpart primary "ext3" 4MiB ${root_end}MiB + $pcmd mkpart primary "linux-swap" ${root_end}MiB ${swap_end}MiB + $pcmd mkpart primary "" ${swap_end}MiB ${disk_mib}MiB + $pcmd mkpart primary "" 1MiB 4MiB + $pcmd set $bios_grubn bios_grub on + $pcmd set $bootn boot on # generally not needed on modern systems + # the mkfs failed before on a vm, which prompted me to add + # sleep .1 + # then failed on a physical machine + # with: + # Device /dev/disk/by-id/foo doesn't exist or access denied, + # so here we wait. + secs=0 + while [[ ! -e `rootdev` ]] && (( secs < 10 )); do + sleep 1 + secs=$((secs +1)) done + yes YES | cryptsetup luksFormat `rootdev` $luks_dir/host-$HOSTNAME \ + -c aes-cbc-essiv:sha256 -s 256 || [[ $? == 141 ]] + yes "$lukspw" | \ + cryptsetup luksAddKey --key-file $luks_dir/host-$HOSTNAME \ + `rootdev` || [[ $? == 141 ]] + # background: Keyfile and password are treated just + # like 2 ways to input a passphrase, so we don't actually need to have + # different contents of keyfile and passphrase, but it makes some + # security sense to a really big randomly generated passphrase + # as much as possible, so we have both. + # + # This would remove the keyfile. + # yes 'test' | cryptsetup luksRemoveKey /dev/... \ + # /key/file || [[ $? == 141 ]] + + cryptsetup luksOpen `rootdev` `root-cryptname` \ + --key-file $luks_dir/host-$HOSTNAME done - #bpart ${devs[@]/%/$rootn} - bpart ${crypt_devs[@]/%/3} - parted ${devs[0]} set 1 boot on - mount $crypt /mnt + ls -la /dev/btrfs-control + sleep 1 + bpart $(for dev in ${devs[@]}; do root-cryptdev; done) + bpart ${boot_devs[@]} else for dev in ${devs[@]}; do - mkfs.ext4 -F ${dev}1 - cryptsetup luksOpen ${dev}3 crypt_dev_${dev##/dev/}3 \ - --key-file $luks_dir/host-$HOSTNAME || [[ $? == 141 ]] + cryptsetup luksOpen `rootdev` `root-cryptname` \ + --key-file $luks_dir/host-$HOSTNAME done sleep 1 - mount -o subvolid=0 $crypt /mnt - # systemd creates subvolumes we want to delete. - s=($(btrfs subvolume list --sort=-path /mnt | - sed -rn 's#^.*path\s*(root/\S+)\s*$#\1#p')) - for subvol in ${s[@]}; do btrfs subvolume delete /mnt/$subvol; done - btrfs subvolume set-default 0 /mnt - btrfs subvolume delete /mnt/root fi +mount -o subvolid=0 $first_root_crypt /mnt +# systemd creates subvolumes we want to delete. +s=($(btrfs subvolume list --sort=-path /mnt | + sed -rn "s#^.*path\s*(root_$DISTRO/\S+)\s*\$#\1#p")) +for subvol in ${s[@]}; do btrfs subvolume delete /mnt/$subvol; done +btrfs subvolume set-default 0 /mnt +[[ ! -e /mnt/root_$DISTRO ]] || btrfs subvolume delete /mnt/root_$DISTRO + ## create subvols ## cd /mnt -for x in q home root; do - btrfs subvolume list . | grep "$x$" >/dev/null || btrfs subvolume create $x -done -for x in root/a q/a; do +btrfs subvolume create root_$DISTRO +[[ -e q ]] || btrfs subvolume create q +chown root:1000 q + +mkdir -p /mnt/root_$DISTRO/boot +for x in q/a q/i; do mkdir -p $x chown 1000:1000 $x chmod 755 $x done -btrfs subvolume set-default \ - $(btrfs subvolume list . | grep 'root$' | awk '{print $2}') . -chattr -Rf +C root +# could set default like this, but no reason to. +# btrfs subvolume set-default \ + # $(btrfs subvolume list . | grep "root_$DISTRO$" | awk '{print $2}') . +chattr -Rf +C root_$DISTRO +cd / +umount /mnt +mount -o subvolid=0 $first_boot_dev /mnt +cd /mnt +btrfs subvolume set-default 0 /mnt +[[ ! -e /mnt/boot_$DISTRO ]] || btrfs subvolume delete /mnt/boot_$DISTRO +btrfs subvolume create boot_$DISTRO cd / umount /mnt ## end create subvols ## - +# note, the mount point /a seems to get automatically created somewhere cat > /tmp/fai/fstab <> /tmp/fai/fstab <<'EOF' +UUID=3f7b31cd-f299-40b4-a86b-7604282e2715 /i btrfs noatime 0 2 EOF +fi +if ifclass frodo; then + cat >> /tmp/fai/fstab <<'EOF' +/q/i /i none bind 0 0 +EOF +fi swaps=() for dev in ${devs[@]}; do - s=crypt_swap_${dev##/dev/}$swapn - swaps+=(/dev/mapper/$s) + swaps+=(`swap-cryptname`) cat >>/tmp/fai/crypttab <> /tmp/fai/fstab </tmp/fai/disk_var.sh <