Root on ZFS with native encryption: Difference between revisions
(order) |
(→Enable persistent device names: find and replace) |
||
Line 316: | Line 316: | ||
echo /etc/zfs/zpool.cache >> /etc/mkinitfs/features.d/zfs.files | echo /etc/zfs/zpool.cache >> /etc/mkinitfs/features.d/zfs.files | ||
replace {{ic|zpool import}} with {{ic|zpool import -c /etc/zfs/zpool.cache}} | replace {{ic|zpool import}} with {{ic|zpool import -c /etc/zfs/zpool.cache}} | ||
find and replace | |||
-d /dev $_root_pool | |||
with | |||
-c /etc/zfs/zpool.cache | |||
and | |||
-d /dev -f $_root_pool | |||
with | |||
-f -c /etc/zfs/zpool.cache | |||
Rebuild initramfs with | Rebuild initramfs with | ||
mkinitfs | mkinitfs |
Revision as of 04:00, 31 December 2020
Objectives
This guide aims to setup encrypted Alpine Linux on ZFS with a layout compatible with boot environments. Mirror and RAID-Z supported.
Except EFI system partition and boot pool /boot
, everything is encrypted. Root pool is encrypted with ZFS native encryption and swap partition is encrypted with dm-crypt.
To do an unencrypted setup, simply omit -O keylocation -O keyformat
when creating root pool.
Notes
Swap on ZFS will cause dead lock
You shouldn't use a ZVol as a swap device, as it can deadlock under memory pressure. See [1] This guide will set up swap on a separate partition with plain dm-crypt.
Resume from swap is not possible, because the key of swap partition can not be stored in the unencrypted boot pool. Busybox initramfs only supports unlocking exactly one LUKS container at boot, therefore boot pool and swap partition can not be both LUKS encrypted. A possible workaround is to import and mount boot pool after booting the system via systemd service.
Resume from ZFS will corrupt the pool
ZFS does not support freeze/thaw operations, which is required for resuming from hibernation, or suspend to disk. Attempt to resume from a swap on ZFS WILL corrupt the pool. See [2]
Encrypted boot pool
GRUB supports booting from LUKS-1 encrypted containers. Therefore, it is possible to encrypt both boot pool and root pool to achieve full disk encryption.
To do this, format boot pool partition as LUKS-1 container and supply the encryption password here. Use keyfile for root pool and embed the keyfile in initramfs.
Since there isn't any sensitive information in /boot
anyway (unless you want to use a persistent LUKS encrypted swap partition for resume from hibernation), encrypting boot pool provides no meaningful benefit and complicates the installation and recovery process.
Pre-installation
UEFI is required. Supports single disk & multi-disk (stripe, mirror, RAID-Z) installation.
Existing data on target disk(s) will be destroyed.
Download the extended release from https://www.alpinelinux.org/downloads/, it's shipped with ZFS kernel module.
Write it to a USB and boot from it.
Setup live environment
Run the following command to setup the live environment, select disk=none at the last step when asked for disk mode. See Installation#Questions_asked_by_setup-alpine.
setup-alpine
The settings given here will be copied to the target system later by setup-disk
.
Install system utilities
apk update apk add eudev sgdisk grub-efi zfs modprobe zfs
Here we must install eudev to have persistent block device names. Do not use /dev/sda for ZFS pools.
rc-update add udev-trigger sysinit /etc/init.d/udev-trigger start
Variables
In this step, we will set some variables to make our installation process easier.
DISK=/dev/disk/by-id/ata-HXY_120G_YS
Use unique disk path instead of /dev/sda
to ensure the correct partition can be found by ZFS.
Other variables
TARGET_USERNAME='your username' ENCRYPTION_PWD='your root pool encryption password, 8 characters min' TARGET_USERPWD='user account password'
Create a mountpoint
MOUNTPOINT=`mktemp -d`
Create a unique suffix for the ZFS pools: this will prevent name conflict when importing pools on another Root on ZFS system.
poolUUID=$(dd if=/dev/urandom of=/dev/stdout bs=1 count=100 2>/dev/null |tr -dc 'a-z0-9' | cut -c-6)
Partitioning
For a single disk, UEFI installation, we need to create at lease 3 partitions:
- EFI system partition
- Boot pool partition
- Root pool partition
Since GRUB only partially support ZFS, many features needs to be disabled on the boot pool. By creating a separate root pool, we can then utilize the full potential of ZFS.
Clear the partition table on the target disk and create EFI, boot and root pool parititions:
sgdisk --zap-all $DISK sgdisk -n1:0:+512M -t1:EF00 $DISK sgdisk -n2:0:+2G $DISK # boot pool sgdisk -n3:0:0 $DISK # root pool
If you want to use a multi-disk setup, such as mirror or RAID-Z, partition every target disk with the same commands above.
Optional: Swap partition
Swap support on ZFS is also problematic, therefore it is recommended to create a separate Swap partition if needed. This guide will cover the creation of a separate swap partition.(can not be used for hibernation since the encryption key is discarded when power off.)
If you want to use swap, reserve some space at the end of disk when creating root pool:
sgdisk -n3:0:-8G $DISK # root pool, reserve 8GB for swap at the end of the disk sgdisk -n4:0:0 $DISK # swap partition
Boot and root pool creation
As mentioned above, ZFS features need to be selectively enabled for GRUB. All available features are enabled when no feature@
is supplied.
Here we explicitly enable those GRUB can support.
zpool create \ -o ashift=12 -d \ -o feature@async_destroy=enabled \ -o feature@bookmarks=enabled \ -o feature@embedded_data=enabled \ -o feature@empty_bpobj=enabled \ -o feature@enabled_txg=enabled \ -o feature@extensible_dataset=enabled \ -o feature@filesystem_limits=enabled \ -o feature@hole_birth=enabled \ -o feature@large_blocks=enabled \ -o feature@lz4_compress=enabled \ -o feature@spacemap_histogram=enabled \ -O acltype=posixacl -O canmount=off -O compression=lz4 \ -O devices=off -O normalization=formD -O relatime=on -O xattr=sa \ -O mountpoint=/boot -R $MOUNTPOINT \ bpool_$poolUUID $DISK-part2
Nothing is stored directly under bpool and rpool, hence canmount=off
. The respective mountpoint
properties are more symbolic than practical.
For root pool all available features are enabled by default
echo $ENCRYPTION_PWD | zpool create \ -o ashift=12 \ -O encryption=aes-256-gcm \ -O keylocation=prompt -O keyformat=passphrase \ -O acltype=posixacl -O canmount=off -O compression=lz4 \ -O dnodesize=auto -O normalization=formD -O relatime=on \ -O xattr=sa -O mountpoint=/ -R $MOUNTPOINT \ rpool_$poolUUID $DISK-part3
Notes for multi-disk
For mirror:
zpool create \ ... \ bpool_$poolUUID mirror \ /dev/disk/by-id/target_disk1-part2 \ /dev/disk/by-id/target_disk2-part2 zpool create \ ... \ rpool_$poolUUID mirror \ /dev/disk/by-id/target_disk1-part3 \ /dev/disk/by-id/target_disk2-part3
For RAID-Z, replace mirror with raidz, raidz2 or raidz3.
Dataset creation
This layout is intended to separate root file system from persistent files. See [3] for a description.
zfs create -o canmount=off -o mountpoint=none rpool_$poolUUID/HOME zfs create -o canmount=off -o mountpoint=none rpool_$poolUUID/ROOT zfs create -o canmount=off -o mountpoint=none bpool_$poolUUID/BOOT zfs create -o mountpoint=/ -o canmount=noauto rpool_$poolUUID/ROOT/default zfs create -o mountpoint=/boot -o canmount=noauto bpool_$poolUUID/BOOT/default zfs mount rpool_$poolUUID/ROOT/default zfs mount bpool_$poolUUID/BOOT/default d='usr var var/lib' for i in $d; do zfs create -o canmount=off rpool_$poolUUID/ROOT/default/$i; done d='srv usr/local' for i in $d; do zfs create rpool_$poolUUID/ROOT/default/$i; done d='log spool tmp' for i in $d; do zfs create rpool_$poolUUID/ROOT/default/var/$i; done zfs create -o mountpoint=/home rpool_$poolUUID/HOME/default zfs create -o mountpoint=/root rpool_$poolUUID/HOME/default/root zfs create rpool_$poolUUID/HOME/default/$TARGET_USERNAME
Format and mount EFI partition
mkfs.vfat -n EFI $DISK-part1 mkdir $MOUNTPOINT/boot/efi mount -t vfat $DISK-part1 $MOUNTPOINT/boot/efi
Install Alpine Linux to target disk
Preparations
GRUB will not find the correct path of root device without ZPOOL_VDEV_NAME_PATH=YES.
export ZPOOL_VDEV_NAME_PATH=YES
setup-disk refuse to run on ZFS by default, we need to add ZFS to the supported filesystem array.
sed -i 's|supported="ext|supported="zfs ext|g' /sbin/setup-disk
Run setup-disk
BOOTLOADER=grub USE_EFI=y setup-disk -v $MOUNTPOINT
Note that grub-probe will still fail despite ZPOOL_VDEV_NAME_PATH=YES variable set above. We will deal with this later inside chroot.
Chroot into new system
mount --rbind /dev $MOUNTPOINT/dev mount --rbind /proc $MOUNTPOINT/proc mount --rbind /sys $MOUNTPOINT/sys chroot $MOUNTPOINT /usr/bin/env TARGET_USERPWD=$TARGET_USERPWD TARGET_USERNAME=$TARGET_USERNAME poolUUID=$poolUUID /bin/sh
Finish GRUB installation
As GRUB installation failed half-way in #Run setup-disk, we will finish it here.
Apply GRUB ZFS fix:
export ZPOOL_VDEV_NAME_PATH=YES
Apply fixes in WARNING.
WARNING
1. GRUB will fail to detect the ZFS filesystem of /boot
with stat
from Busybox.
See source file of grub-mkconfig, the problem is:
GRUB_DEVICE="`${grub_probe} --target=device /`" # will fail with `grub-probe: error: unknown filesystem.` GRUB_FS="`${grub_probe} --device ${GRUB_DEVICE} --target=fs 2> /dev/null || echo unknown`" # will also fail. The final fall back is if [ x"$GRUB_FS" = xunknown ]; then GRUB_FS="$(stat -f -c %T / || echo unknown)" fi # `stat` from coreutils will return `zfs`, the correct answer # `stat` from busybox will return `UNKNOWN`, cause `10_linux` script to fail
Therefore we need to install coreutils
.
apk add coreutils
2. GRUB will stuff an empty result if it does not support root pool.
GRUB is lagging behind development of OpenZFS, see [4]. A temporary fix is to replace detection of rpool with the method given in patch.
sed -i "s/rpool=.*/rpool=\`zdb -l \${GRUB_DEVICE} \| awk -F \\\' '\/ name\/ { print \$2 }'\`/" /etc/grub.d/10_linux
Need to be applied upon every GRUB update until the patch is merged.
Generate grub.cfg
After applying fixes, finally run
grub-mkconfig -o /boot/grub/grub.cfg
Initramfs fixes
Fix zfs decrypt
As of this writing, the initramfs has a bug for entering ZFS password at boot. When booting the system, root dataset will fail to mount sh: `active`, unknown operand
and drop into emergency shell.
In /usr/share/mkinitfs/initramfs-init
:
# Ask for encryption password if [ $(zpool list -H -o feature@encryption $_root_pool) = "active" ]; then # replace underline with true will fix it # if true; then local _encryption_root=$(zfs get -H -o value encryptionroot $_root_vol) if [ "$_encryption_root" != "-" ]; then eval zfs load-key $_encryption_root fi fi
Enable persistent device names
Special modifications need to be made to populate /dev/disk/by-*
in initramfs.
Ensure eudev
is installed
apk add eudev
Create /etc/mkinitfs/features.d/eudev.files
to add eudev
to initramfs.
tee /etc/mkinitfs/features.d/eudev.files << EOF /bin/udevadm /sbin/udevadm /sbin/udevd /etc/udev/* /lib/udev/* /usr/lib/libudev* EOF
Edit /usr/share/mkinitfs/initramfs-init
.
Add functions from /etc/init.d/udev*
at the beginning of the file.
# persistent device names from eudev, see /etc/init.d/udev* eudev_start_pre() { # load unix domain sockets if built as module, Bug #221253 # and not yet loaded, Bug #363549 if [ ! -e /proc/net/unix ]; then if ! modprobe unix; then eerror "Cannot load the unix domain socket module" return 1 fi fi if [ -e /proc/sys/kernel/hotplug ]; then echo "" >/proc/sys/kernel/hotplug fi return 0 } eudev_dir_writeable() { touch "$1"/.test.$$ 2>/dev/null && rm "$1"/.test.$$ } # store persistent-rules that got created while booting # when / was still read-only eudev_store_persistent_rules() { # create /etc/udev/rules.d if it does not exist and /etc/udev is writable [ -d /etc/udev/rules.d ] || \ eudev_dir_writeable /etc/udev && \ mkdir -p /etc/udev/rules.d # only continue if rules-directory is writable eudev_dir_writeable /etc/udev/rules.d || return 0 local file dest for file in /run/udev/tmp-rules--*; do dest=${file##*tmp-rules--} [ "$dest" = '*' ] && break type=${dest##70-persistent-} type=${type%%.rules} cat "$file" >> /etc/udev/rules.d/"$dest" && rm -f "$file" done } eudev_start() { eudev_start_pre udevd -d # store persistent-rules that got created while booting # when / was still read-only eudev_store_persistent_rules # Populating /dev with existing devices through uevents" udevadm trigger --type=subsystems --action=add udevadm trigger --type=devices --action=add }
After the following section, which populates /dev
mount -t devtmpfs -o exec,nosuid,mode=0755,size=2M devtmpfs /dev 2>/dev/null \ || mount -t tmpfs -o exec,nosuid,mode=0755,size=2M tmpfs /dev # pty device nodes (later system will need it) [ -c /dev/ptmx ] || mknod -m 666 /dev/ptmx c 5 2 [ -d /dev/pts ] || mkdir -m 755 /dev/pts mount -t devpts -o gid=5,mode=0620,noexec,nosuid devpts /dev/pts # shared memory area (later system will need it) [ -d /dev/shm ] || mkdir /dev/shm mount -t tmpfs -o nodev,nosuid,noexec shm /dev/shm
add this paragraph.
# persistent device names from eudev if [ -f /sbin/udevadm ]; then eudev_start fi
To let zpool import
to use persistent names, use zpool.cache to import pool
zpool set cachefile=/etc/zfs/zpool.cache rpool_$poolUUID zpool set cachefile=/etc/zfs/zpool.cache bpool_$poolUUID echo /etc/zfs/zpool.cache >> /etc/mkinitfs/features.d/zfs.files
replace zpool import
with zpool import -c /etc/zfs/zpool.cache
find and replace -d /dev $_root_pool with -c /etc/zfs/zpool.cache and -d /dev -f $_root_pool with -f -c /etc/zfs/zpool.cache
Rebuild initramfs with
mkinitfs
Install packages
These packages are used for creating a common user account. Root account is accessed with sudo. Also package for persisitent block device name must be installed.
shadow
is available in community repo. Enable it first:
vi /etc/apk/repositories # uncomment community line
Install
apk add shadow sudo eudev
Enable sudo access for wheel group
mv /etc/sudoers /etc/sudoers.original tee /etc/sudoers << EOF root ALL=(ALL) ALL %wheel ALL=(ALL) ALL EOF
Add normal user account
useradd -s /bin/sh -U -G wheel,video -d /home/$TARGET_USERNAME $TARGET_USERNAME chown -R $TARGET_USERNAME:$TARGET_USERNAME /home/$TARGET_USERNAME echo "$TARGET_USERNAME:$TARGET_USERPWD" | chpasswd
Optional: Enable encrypted swap partition
Install cryptsetup
apk add cryptsetup
Edit the /mnt/etc/mkinitfs/mkinitfs.conf
file and append the cryptsetup
module to the features
parameter:
features="ata base ide scsi usb virtio ext4 lvm cryptsetup zfs"
Add relevant lines in fstab
and crypttab
. Replace $DISK
with actual disk.
echo swap $DISK-part4 /dev/urandom swap,cipher=aes-cbc-essiv:sha256,size=256 >> /etc/crypttab echo /dev/mapper/swap none swap defaults 0 0 >> /etc/fstab
Rebuild initramfs with mkinitfs
.
Finish installation
Take a snapshot for the clean installation for future use and export all pools.
exit zfs snapshot -r rpool_$poolUUID/ROOT/default@install zfs snapshot -r bpool_$poolUUID/BOOT/default@install
Pools must be exported before reboot, or they will fail to be imported on boot.
mount | grep -v zfs | tac | grep $MOUNTPOINT | awk '{print $3}' | \ xargs -i{} umount -lf {} zpool export bpool_$poolUUID zpool export rpool_$poolUUID
Reboot
Until this is fixed, we need to manually load the key and mount root dataset with
zfs load-key -a # enter password mount -t zfs rpool_$UUID/ROOT/default /sysroot
Disk space stat
Without optional swap or cryptsetup:
- bpool used 25.2M
- rpool used 491M
- efi used 416K
Recovery in Live environment
After installing zfs packages, run the following command:
Create a mount point and store encryption password in a variable:
MOUNTPOINT=`mktemp -d` ENCRYPTION_PWD='YOUR DISK ENCRYPTION PASSWORD, 8 MINIMUM'
Find the unique UUID of your pool with
zpool import
Import rpool without mounting datasets: -N
for not mounting all datasets; -R
for alternate root.
poolUUID=abc123 zpool import -N -R $MOUNTPOINT rpool_$poolUUID
Load encryption key
echo $ENCRYPTION_PWD | zfs load-key -a
As canmount=noauto
is set for /
dataset, we have to mount it manually. To find the dataset, use
zfs list rpool_$poolUUID/ROOT
Mount /
dataset
zfs mount rpool_$UUID/ROOT/$dataset
Mount other datasets
zfs mount -a
Import bpool
zpool import -N -R $MOUNTPOINT bpool_$UUID
Find and mount the /boot
dataset, same as above.
zfs list bpool_$UUID/BOOT zfs mount bpool_$UUID/BOOT/$dataset
Chroot
mount --rbind /dev $MOUNTPOINT/dev mount --rbind /proc $MOUNTPOINT/proc mount --rbind /sys $MOUNTPOINT/sys chroot $MOUNTPOINT /bin/sh
After chroot, mount /efi
mount /boot/efi
After fixing the system, don't forget to umount and export the pools:
mount | grep -v zfs | tac | grep $MOUNTPOINT | awk '{print $3}' | \ xargs -i{} umount -lf {} zpool export bpool_$poolUUID zpool export rpool_$poolUUID