mount.zfs-non-legacy

#!/bin/sh
# mount.zfs-non-legacy (part of ossobv/vcutil) // wdoekes/2021-2022
#   // Public Domain
#
# This script is a mount wrapper that adds the 'zfs-non-legacy' type for
# mount(8). It checks whether a particular ZFS dataset is mounted,
# and tries to import the pool and mount it, if it isn't.
#
# Rationale behind this script is the following:
# - systemd has .mount files that can be used in dependencies
# - zfs (non-legacy) doesn't use mount(8), so systemd interfaces poorly with it
# By using this script, we can leverage the power of systemd .mount files
# while using regular ZFS mount points.
#
# Common scenario:
# - /var/lib/docker resides on a separate ZFS pool (data)
# - docker.service needs to depend on that
# If there is no such dependency, docker might start _before_ all zfs imports
# and mounts have completed, data gets written in the root filesystem instead.
#
# >>> Fix in /etc/systemd/system/var-lib-docker.mount <<<
# [Unit]
# Documentation=https://github.com/ossobv/vcutil/blob/main/mount.zfs-non-legacy
# After=zfs-mount.service
# Requires=zfs-mount.service
# [Mount]
# Type=zfs-non-legacy
# What=data/docker
#
# >>> Fix in /etc/systemd/system/docker.service.d/override.conf <<<
# [Unit]
# RequiresMountsFor=/var/lib/docker
#
# Now a start of the docker.service will depend on var-lib-docker.mount.
# That mount unit will call something like this:
# > mount -t zfs-non-legacy data/docker /var/lib/docker
# And that call ends up in this wrapper script:
# - it checks if data is imported, and imports it if it isn't
# - it checks that the mount path is correct
# - it checks whether it is mounted, and if not, it mounts it
#
set -eu

name="$1"  # data/docker
path="$2"  # /var/lib/docker

if test "${name#/}" != "$name"; then  # fileset has a leading slash
    # https://github.com/util-linux/util-linux/commit/
    #   372ce5b74e79470b1bda1fc284c19a313a422361
    # is a fix for fstype = "zfs" only. Here we have
    # fstype = "zfs-non-legacy", so it suffers from the same issue as:
    # https://github.com/openzfs/zfs/pull/11295#issuecomment-757889868
    # > mount -t zfs foo /foo
    # > # will run 'mount.zfs /foo /foo' internally because /foo exists
    # A satisfactory solution to this problem was if we could tell
    # systemd to add the --no-canonicalize mount option:
    # https://github.com/systemd/systemd/issues/18188
    echo "Looks like systemd/systemd#23795 or systemd/systemd#18188 or\
 util-linux/util-linux#1231 is not fixed satisfactorily (we want a\
 --no-canonicalize mount option): args=[$*] -- removing slash from\
 \$1" >&2
    name=${name#/}
fi

if ! command -v zpool >/dev/null || ! command -v zfs >/dev/null; then
    echo "No ZFS installed?" >&2
    exit 1
fi

# Get mounted state, but tries to import the zpool if it was not found.
# Implemented in a function so we can retry this during boot. During boot
# several jobs may be attempting to mount/import stuff, causing intermittent
# failures that will auto-heal.
get_mounted() {
    local name="$1"         # data/docker
    local zpool=${name%%/*} # data
    local found=0
    local mounted
    mounted="$(zfs list -Homounted "$name")" && found=1
    if test $found -eq 0; then
        if ! zpool list -Honame "$zpool" >/dev/null; then
            if ! zpool import "$zpool"; then
                false  # intermittent or permanent failure?
                return
            fi
        fi
        mounted="$(zfs list -Homounted "$name")" || mounted=no-dataset
    fi
    echo "$mounted"
}

# Get mounted value. Retry a couple of times.
mounted=
for retry in 1 2 3 0; do
    mounted=$(get_mounted "$name") && break
    sleep $retry || true
    false
done || mounted=no-pool

# Handle the various possibilities.
case $mounted in
no-dataset|no-pool)
    echo "zfs-non-legacy: failed to load $name: $mounted" >&2
    exit 1
    ;;
no|yes)
    if ! mountpoint=$(zfs list -Homountpoint "$name") ||
            test "$path" != "$mountpoint"; then
        echo "zfs-non-legacy: dest fail for $name: $path ~ $mountpoint" >&2
        exit 1
    fi
    ;;
*)
    echo "zfs-non-legacy: unexpected mounted value for $name: $mounted" >&2
    exit 1
    ;;
esac

# So.. it wasn't mounted? Try that then.
if test "$mounted" = no && ! zfs mount "$name"; then
    echo "zfs-non-legacy: failed to mount $name" >&2
    exit 1
fi

# All good.
exit 0