mirror of
https://github.com/systemd/systemd
synced 2024-11-05 18:25:39 +00:00
7a321b5a21
Having these named differently than the test itself mostly creates unecessary confusion and makes writing logic against the tests harder so let's rename the testsuite-xx units and scripts to just use the test name itself.
407 lines
21 KiB
Bash
Executable file
407 lines
21 KiB
Bash
Executable file
#!/usr/bin/env bash
|
|
# SPDX-License-Identifier: LGPL-2.1-or-later
|
|
# shellcheck disable=SC2235,SC2233
|
|
set -eux
|
|
set -o pipefail
|
|
|
|
# shellcheck source=test/units/generator-utils.sh
|
|
. "$(dirname "$0")/generator-utils.sh"
|
|
|
|
GENERATOR_BIN="/usr/lib/systemd/system-generators/systemd-fstab-generator"
|
|
NETWORK_FS_RX="^(afs|ceph|cifs|gfs|gfs2|ncp|ncpfs|nfs|nfs4|ocfs2|orangefs|pvfs2|smb3|smbfs|davfs|glusterfs|lustre|sshfs)$"
|
|
OUT_DIR="$(mktemp -d /tmp/fstab-generator.XXX)"
|
|
FSTAB="$(mktemp)"
|
|
|
|
at_exit() {
|
|
mountpoint -q /proc/cmdline && umount /proc/cmdline
|
|
rm -fr "${OUT_DIR:?}" "${FSTAB:?}"
|
|
}
|
|
|
|
trap at_exit EXIT
|
|
|
|
test -x "${GENERATOR_BIN:?}"
|
|
|
|
FSTAB_GENERAL=(
|
|
# Valid entries
|
|
"/dev/test2 /nofail ext4 nofail 0 0"
|
|
"/dev/test3 /regular btrfs defaults 0 0"
|
|
"/dev/test4 /x-systemd.requires xfs x-systemd.requires=foo.service 0 0"
|
|
"/dev/test5 /x-systemd.before-after xfs x-systemd.before=foo.service,x-systemd.after=bar.mount 0 0"
|
|
"/dev/test6 /x-systemd.wanted-required-by xfs x-systemd.wanted-by=foo.service,x-systemd.required-by=bar.device 0 0"
|
|
"/dev/test7 /x-systemd.requires-mounts-for xfs x-systemd.requires-mounts-for=/foo/bar/baz 0 0"
|
|
"/dev/test8 /x-systemd.automount-idle-timeout vfat x-systemd.automount,x-systemd.idle-timeout=50s 0 0"
|
|
"/dev/test9 /x-systemd.makefs xfs x-systemd.makefs 0 0"
|
|
"/dev/test10 /x-systemd.growfs xfs x-systemd.growfs 0 0"
|
|
"/dev/test11 /_netdev ext4 defaults,_netdev 0 0"
|
|
"/dev/test12 /_rwonly ext4 x-systemd.rw-only 0 0"
|
|
"/dev/test13 /chaos1 zfs x-systemd.rw-only,x-systemd.requires=hello.service,x-systemd.after=my.device 0 0"
|
|
"/dev/test14 /chaos2 zfs x.systemd.wanted-by=foo.service,x-systemd.growfs,x-systemd.makefs 0 0"
|
|
"/dev/test15 /fstype/auto auto defaults 0 0"
|
|
"/dev/test16 /fsck/me ext4 defaults 0 1"
|
|
"/dev/test17 /also/fsck/me ext4 defaults,x-systemd.requires-mounts-for=/var/lib/foo 0 99"
|
|
"/dev/test18 /swap swap defaults 0 0"
|
|
"/dev/test19 /swap/makefs swap defaults,x-systemd.makefs 0 0"
|
|
"/dev/test20 /var xfs defaults,x-systemd.device-timeout=1h 0 0"
|
|
"/dev/test21 /usr ext4 defaults 0 1"
|
|
"/dev/test22 /initrd/mount ext2 defaults,x-systemd.rw-only,x-initrd.mount 0 1"
|
|
"/dev/test23 /initrd/mount/nofail ext3 defaults,nofail,x-initrd.mount 0 1"
|
|
"/dev/test24 /initrd/mount/deps ext4 x-initrd.mount,x-systemd.before=early.service,x-systemd.after=late.service 0 1"
|
|
|
|
# Incomplete, but valid entries
|
|
"/dev/incomplete1 /incomplete1"
|
|
"/dev/incomplete2 /incomplete2 ext4"
|
|
"/dev/incomplete3 /incomplete3 ext4 defaults"
|
|
"/dev/incomplete4 /incomplete4 ext4 defaults 0"
|
|
|
|
# Remote filesystems
|
|
"/dev/remote1 /nfs nfs bg 0 0"
|
|
"/dev/remote2 /nfs4 nfs4 bg 0 0"
|
|
"bar.tld:/store /remote/storage nfs ro,x-systemd.wanted-by=store.service 0 0"
|
|
"user@host.tld:/remote/dir /remote/top-secret sshfs rw,x-systemd.before=naughty.service 0 0"
|
|
"foo.tld:/hello /hello/world ceph defaults 0 0"
|
|
"//192.168.0.1/storage /cifs-storage cifs automount,nofail 0 0"
|
|
)
|
|
|
|
FSTAB_GENERAL_ROOT=(
|
|
# rootfs with bunch of options we should ignore and fsck enabled
|
|
"/dev/test1 / ext4 noauto,nofail,x-systemd.automount,x-systemd.wanted-by=foo,x-systemd.required-by=bar 0 1"
|
|
"${FSTAB_GENERAL[@]}"
|
|
)
|
|
|
|
FSTAB_MINIMAL=(
|
|
"/dev/loop1 /foo/bar ext3 defaults 0 0"
|
|
)
|
|
|
|
FSTAB_DUPLICATE=(
|
|
"/dev/dup1 / ext4 defaults 0 1"
|
|
"/dev/dup2 / ext4 defaults,x-systemd.requires=foo.mount 0 2"
|
|
)
|
|
|
|
FSTAB_INVALID=(
|
|
# Ignored entries
|
|
"/dev/ignored1 /sys/fs/cgroup/foo ext4 defaults 0 0"
|
|
"/dev/ignored2 /sys/fs/selinux ext4 defaults 0 0"
|
|
"/dev/ignored3 /dev/console ext4 defaults 0 0"
|
|
"/dev/ignored4 /proc/kmsg ext4 defaults 0 0"
|
|
"/dev/ignored5 /proc/sys ext4 defaults 0 0"
|
|
"/dev/ignored6 /proc/sys/kernel/random/boot_id ext4 defaults 0 0"
|
|
"/dev/ignored7 /run/host ext4 defaults 0 0"
|
|
"/dev/ignored8 /run/host/foo ext4 defaults 0 0"
|
|
"/dev/ignored9 /autofs autofs defaults 0 0"
|
|
"/dev/invalid1 not-a-path ext4 defaults 0 0"
|
|
""
|
|
"/dev/invalid1"
|
|
" "
|
|
"\\"
|
|
"$"
|
|
)
|
|
|
|
check_fstab_mount_units() {
|
|
local what where fstype opts passno unit
|
|
local item opt split_options filtered_options supp service device arg
|
|
local array_name="${1:?}"
|
|
local out_dir="${2:?}/normal"
|
|
# Get a reference to the array from its name
|
|
local -n fstab_entries="$array_name"
|
|
|
|
# Running the checks in a container is pretty much useless, since we don't
|
|
# generate any mounts, but don't skip the whole test to test the "skip"
|
|
# paths as well
|
|
in_container && return 0
|
|
|
|
for item in "${fstab_entries[@]}"; do
|
|
# Don't use a pipe here, as it would make the variables out of scope
|
|
read -r what where fstype opts _ passno <<< "$item"
|
|
|
|
# Skip non-initrd mounts in initrd
|
|
if in_initrd_host && ! [[ "$opts" =~ x-initrd.mount ]]; then
|
|
continue
|
|
fi
|
|
|
|
if [[ "$fstype" == swap ]]; then
|
|
unit="$(systemd-escape --suffix=swap --path "${what:?}")"
|
|
cat "$out_dir/$unit"
|
|
|
|
grep -qE "^What=$what$" "$out_dir/$unit"
|
|
if [[ "$opts" != defaults ]]; then
|
|
grep -qE "^Options=$opts$" "$out_dir/$unit"
|
|
fi
|
|
|
|
if [[ "$opts" =~ x-systemd.makefs ]]; then
|
|
service="$(systemd-escape --template=systemd-mkswap@.service --path "$what")"
|
|
test -e "$out_dir/$service"
|
|
fi
|
|
|
|
continue
|
|
fi
|
|
|
|
# If we're parsing host's fstab in initrd, prefix all mount targets
|
|
# with /sysroot
|
|
in_initrd_host && where="/sysroot${where:?}"
|
|
unit="$(systemd-escape --suffix=mount --path "${where:?}")"
|
|
cat "$out_dir/$unit"
|
|
|
|
# Check the general stuff
|
|
grep -qE "^What=$what$" "$out_dir/$unit"
|
|
grep -qE "^Where=$where$" "$out_dir/$unit"
|
|
if [[ -n "$fstype" ]] && [[ "$fstype" != auto ]]; then
|
|
grep -qE "^Type=$fstype$" "$out_dir/$unit"
|
|
fi
|
|
if [[ -n "$opts" ]] && [[ "$opts" != defaults ]]; then
|
|
# Some options are not propagated to the generated unit
|
|
if [[ "$where" == / || "$where" == /usr ]]; then
|
|
filtered_options="$(opt_filter "$opts" "(noauto|nofail|x-systemd.(wanted-by=|required-by=|automount|device-timeout=))")"
|
|
else
|
|
filtered_options="$(opt_filter "$opts" "^x-systemd.device-timeout=")"
|
|
fi
|
|
|
|
if [[ "${filtered_options[*]}" != defaults ]]; then
|
|
grep -qE "^Options=.*$filtered_options.*$" "$out_dir/$unit"
|
|
fi
|
|
fi
|
|
|
|
if ! [[ "$opts" =~ (noauto|x-systemd.(wanted-by=|required-by=|automount)) ]]; then
|
|
# We don't create the Requires=/Wants= symlinks for noauto/automount mounts
|
|
# and for mounts that use x-systemd.wanted-by=/required-by=
|
|
if in_initrd_host; then
|
|
if [[ "$where" == / ]] || ! [[ "$opts" =~ nofail ]]; then
|
|
link_eq "$out_dir/initrd-fs.target.requires/$unit" "../$unit"
|
|
else
|
|
link_eq "$out_dir/initrd-fs.target.wants/$unit" "../$unit"
|
|
fi
|
|
elif [[ "$fstype" =~ $NETWORK_FS_RX || "$opts" =~ _netdev ]]; then
|
|
# Units with network filesystems should have a Requires= dependency
|
|
# on the remote-fs.target, unless they use nofail or are an nfs "bg"
|
|
# mounts, in which case the dependency is downgraded to Wants=
|
|
if [[ "$opts" =~ nofail ]] || [[ "$fstype" =~ ^(nfs|nfs4) && "$opts" =~ bg ]]; then
|
|
link_eq "$out_dir/remote-fs.target.wants/$unit" "../$unit"
|
|
else
|
|
link_eq "$out_dir/remote-fs.target.requires/$unit" "../$unit"
|
|
fi
|
|
else
|
|
# Similarly, local filesystems should have a Requires= dependency on
|
|
# the local-fs.target, unless they use nofail, in which case the
|
|
# dependency is downgraded to Wants=. Rootfs is a special case,
|
|
# since we always ignore nofail there
|
|
if [[ "$where" == / ]] || ! [[ "$opts" =~ nofail ]]; then
|
|
link_eq "$out_dir/local-fs.target.requires/$unit" "../$unit"
|
|
else
|
|
link_eq "$out_dir/local-fs.target.wants/$unit" "../$unit"
|
|
fi
|
|
fi
|
|
fi
|
|
|
|
if [[ "${passno:=0}" -ne 0 ]]; then
|
|
# Generate systemd-fsck@.service dependencies, if applicable
|
|
if in_initrd && [[ "$where" == / || "$where" == /usr ]]; then
|
|
continue
|
|
fi
|
|
|
|
if [[ "$where" == / ]]; then
|
|
link_endswith "$out_dir/local-fs.target.wants/systemd-fsck-root.service" "/lib/systemd/system/systemd-fsck-root.service"
|
|
else
|
|
service="$(systemd-escape --template=systemd-fsck@.service --path "$what")"
|
|
grep -qE "^After=$service$" "$out_dir/$unit"
|
|
if [[ "$where" == /usr ]]; then
|
|
grep -qE "^Wants=$service$" "$out_dir/$unit"
|
|
else
|
|
grep -qE "^Requires=$service$" "$out_dir/$unit"
|
|
fi
|
|
fi
|
|
fi
|
|
|
|
# Check various x-systemd options
|
|
#
|
|
# First, split them into an array to make splitting them even further
|
|
# easier
|
|
IFS="," read -ra split_options <<< "$opts"
|
|
# and process them one by one.
|
|
#
|
|
# Note: the "machinery" below might (and probably does) miss some
|
|
# combinations of supported options, so tread carefully
|
|
for opt in "${split_options[@]}"; do
|
|
if [[ "$opt" =~ ^x-systemd.requires= ]]; then
|
|
service="$(opt_get_arg "$opt")"
|
|
grep -qE "^Requires=$service$" "$out_dir/$unit"
|
|
grep -qE "^After=$service$" "$out_dir/$unit"
|
|
elif [[ "$opt" =~ ^x-systemd.before= ]]; then
|
|
service="$(opt_get_arg "$opt")"
|
|
grep -qE "^Before=$service$" "$out_dir/$unit"
|
|
elif [[ "$opt" =~ ^x-systemd.after= ]]; then
|
|
service="$(opt_get_arg "$opt")"
|
|
grep -qE "^After=$service$" "$out_dir/$unit"
|
|
elif [[ "$opt" =~ ^x-systemd.wanted-by= ]]; then
|
|
service="$(opt_get_arg "$opt")"
|
|
if [[ "$where" == / ]]; then
|
|
# This option is ignored for rootfs mounts
|
|
(! link_eq "$out_dir/$service.wants/$unit" "../$unit")
|
|
else
|
|
link_eq "$out_dir/$service.wants/$unit" "../$unit"
|
|
fi
|
|
elif [[ "$opt" =~ ^x-systemd.required-by= ]]; then
|
|
service="$(opt_get_arg "$opt")"
|
|
if [[ "$where" == / ]]; then
|
|
# This option is ignored for rootfs mounts
|
|
(! link_eq "$out_dir/$service.requires/$unit" "../$unit")
|
|
else
|
|
link_eq "$out_dir/$service.requires/$unit" "../$unit"
|
|
fi
|
|
elif [[ "$opt" =~ ^x-systemd.requires-mounts-for= ]]; then
|
|
arg="$(opt_get_arg "$opt")"
|
|
grep -qE "^RequiresMountsFor=$arg$" "$out_dir/$unit"
|
|
elif [[ "$opt" == x-systemd.device-bound ]]; then
|
|
# This is implied for fstab mounts
|
|
:
|
|
elif [[ "$opt" == x-systemd.automount ]]; then
|
|
# The $unit should have an accompanying automount unit
|
|
supp="$(systemd-escape --suffix=automount --path "$where")"
|
|
if [[ "$where" == / ]]; then
|
|
# This option is ignored for rootfs mounts
|
|
test ! -e "$out_dir/$supp"
|
|
(! link_eq "$out_dir/local-fs.target.requires/$supp" "../$supp")
|
|
else
|
|
test -e "$out_dir/$supp"
|
|
link_eq "$out_dir/local-fs.target.requires/$supp" "../$supp"
|
|
fi
|
|
elif [[ "$opt" =~ ^x-systemd.idle-timeout= ]]; then
|
|
# The timeout applies to the automount unit, not the original
|
|
# mount one
|
|
arg="$(opt_get_arg "$opt")"
|
|
supp="$(systemd-escape --suffix=automount --path "$where")"
|
|
grep -qE "^TimeoutIdleSec=$arg$" "$out_dir/$supp"
|
|
elif [[ "$opt" =~ ^x-systemd.device-timeout= ]]; then
|
|
arg="$(opt_get_arg "$opt")"
|
|
device="$(systemd-escape --suffix=device --path "$what")"
|
|
grep -qE "^JobRunningTimeoutSec=$arg$" "$out_dir/${device}.d/50-device-timeout.conf"
|
|
elif [[ "$opt" == x-systemd.makefs ]]; then
|
|
service="$(systemd-escape --template=systemd-makefs@.service --path "$what")"
|
|
test -e "$out_dir/$service"
|
|
link_eq "$out_dir/${unit}.requires/$service" "../$service"
|
|
elif [[ "$opt" == x-systemd.rw-only ]]; then
|
|
grep -qE "^ReadWriteOnly=yes$" "$out_dir/$unit"
|
|
elif [[ "$opt" == x-systemd.growfs ]]; then
|
|
service="$(systemd-escape --template=systemd-growfs@.service --path "$where")"
|
|
link_endswith "$out_dir/${unit}.wants/$service" "/lib/systemd/system/systemd-growfs@.service"
|
|
elif [[ "$opt" == bg ]] && [[ "$fstype" =~ ^(nfs|nfs4)$ ]]; then
|
|
# We "convert" nfs bg mounts to fg, so we can do the job-control
|
|
# ourselves
|
|
grep -qE "^Options=.*\bx-systemd.mount-timeout=infinity\b" "$out_dir/$unit"
|
|
grep -qE "^Options=.*\bfg\b.*" "$out_dir/$unit"
|
|
elif [[ "$opt" =~ ^x-systemd\. ]]; then
|
|
echo >&2 "Unhandled mount option: $opt"
|
|
exit 1
|
|
fi
|
|
done
|
|
done
|
|
}
|
|
|
|
: "fstab-generator: regular"
|
|
printf "%s\n" "${FSTAB_GENERAL_ROOT[@]}" >"$FSTAB"
|
|
cat "$FSTAB"
|
|
SYSTEMD_FSTAB="$FSTAB" run_and_list "$GENERATOR_BIN" "$OUT_DIR"
|
|
check_fstab_mount_units FSTAB_GENERAL_ROOT "$OUT_DIR"
|
|
|
|
# Skip the rest when running in a container, as it makes little sense to check
|
|
# initrd-related stuff there and fstab-generator might have a bit strange
|
|
# behavior during certain tests, like https://github.com/systemd/systemd/issues/27156
|
|
if in_container; then
|
|
echo "Running in a container, skipping the rest of the fstab-generator tests..."
|
|
exit 0
|
|
fi
|
|
|
|
# In this mode we treat the entries as "regular" ones
|
|
: "fstab-generator: initrd - initrd fstab"
|
|
printf "%s\n" "${FSTAB_GENERAL[@]}" >"$FSTAB"
|
|
cat "$FSTAB"
|
|
SYSTEMD_IN_INITRD=1 SYSTEMD_FSTAB="$FSTAB" SYSTEMD_SYSROOT_FSTAB=/dev/null run_and_list "$GENERATOR_BIN" "$OUT_DIR"
|
|
SYSTEMD_IN_INITRD=1 SYSTEMD_FSTAB="$FSTAB" SYSTEMD_SYSROOT_FSTAB=/dev/null check_fstab_mount_units FSTAB_GENERAL "$OUT_DIR"
|
|
|
|
# In this mode we prefix the mount target with /sysroot and ignore all mounts
|
|
# that don't have the x-initrd.mount flag
|
|
: "fstab-generator: initrd - host fstab"
|
|
printf "%s\n" "${FSTAB_GENERAL_ROOT[@]}" >"$FSTAB"
|
|
cat "$FSTAB"
|
|
SYSTEMD_IN_INITRD=1 SYSTEMD_FSTAB=/dev/null SYSTEMD_SYSROOT_FSTAB="$FSTAB" run_and_list "$GENERATOR_BIN" "$OUT_DIR"
|
|
SYSTEMD_IN_INITRD=1 SYSTEMD_FSTAB=/dev/null SYSTEMD_SYSROOT_FSTAB="$FSTAB" check_fstab_mount_units FSTAB_GENERAL_ROOT "$OUT_DIR"
|
|
|
|
# Check the default stuff that we (almost) always create in initrd
|
|
: "fstab-generator: initrd default"
|
|
SYSTEMD_PROC_CMDLINE="root=/dev/sda2" SYSTEMD_IN_INITRD=1 SYSTEMD_FSTAB=/dev/null SYSTEMD_SYSROOT_FSTAB=/dev/null run_and_list "$GENERATOR_BIN" "$OUT_DIR"
|
|
test -e "$OUT_DIR/normal/sysroot.mount"
|
|
test -e "$OUT_DIR/normal/systemd-fsck-root.service"
|
|
link_eq "$OUT_DIR/normal/initrd-root-fs.target.requires/sysroot.mount" "../sysroot.mount"
|
|
link_eq "$OUT_DIR/normal/initrd-root-fs.target.requires/sysroot.mount" "../sysroot.mount"
|
|
|
|
: "fstab-generator: run as systemd-sysroot-fstab-check in initrd"
|
|
ln -svf "$GENERATOR_BIN" /tmp/systemd-sysroot-fstab-check
|
|
(! /tmp/systemd-sysroot-fstab-check foo)
|
|
(! SYSTEMD_IN_INITRD=0 /tmp/systemd-sysroot-fstab-check)
|
|
printf "%s\n" "${FSTAB_GENERAL[@]}" >"$FSTAB"
|
|
SYSTEMD_IN_INITRD=1 SYSTEMD_SYSROOT_FSTAB="$FSTAB" /tmp/systemd-sysroot-fstab-check
|
|
|
|
: "fstab-generator: duplicate"
|
|
printf "%s\n" "${FSTAB_DUPLICATE[@]}" >"$FSTAB"
|
|
cat "$FSTAB"
|
|
(! SYSTEMD_FSTAB="$FSTAB" run_and_list "$GENERATOR_BIN" "$OUT_DIR")
|
|
|
|
: "fstab-generator: invalid"
|
|
printf "%s\n" "${FSTAB_INVALID[@]}" >"$FSTAB"
|
|
cat "$FSTAB"
|
|
# Don't care about the exit code here
|
|
SYSTEMD_PROC_CMDLINE="" SYSTEMD_FSTAB="$FSTAB" run_and_list "$GENERATOR_BIN" "$OUT_DIR" || :
|
|
# No mounts should get created here
|
|
[[ "$(find "$OUT_DIR" -name "*.mount" | wc -l)" -eq 0 ]]
|
|
|
|
: "fstab-generator: kernel args - fstab=0"
|
|
printf "%s\n" "${FSTAB_MINIMAL[@]}" >"$FSTAB"
|
|
SYSTEMD_FSTAB="$FSTAB" SYSTEMD_PROC_CMDLINE="fstab=0" run_and_list "$GENERATOR_BIN" "$OUT_DIR"
|
|
(! SYSTEMD_FSTAB="$FSTAB" check_fstab_mount_units FSTAB_MINIMAL "$OUT_DIR")
|
|
SYSTEMD_IN_INITRD=1 SYSTEMD_FSTAB="$FSTAB" SYSTEMD_PROC_CMDLINE="fstab=0" run_and_list "$GENERATOR_BIN" "$OUT_DIR"
|
|
(! SYSTEMD_IN_INITRD=1 SYSTEMD_FSTAB="$FSTAB" check_fstab_mount_units FSTAB_MINIMAL "$OUT_DIR")
|
|
|
|
: "fstab-generator: kernel args - rd.fstab=0"
|
|
printf "%s\n" "${FSTAB_MINIMAL[@]}" >"$FSTAB"
|
|
SYSTEMD_FSTAB="$FSTAB" SYSTEMD_PROC_CMDLINE="rd.fstab=0" run_and_list "$GENERATOR_BIN" "$OUT_DIR"
|
|
SYSTEMD_FSTAB="$FSTAB" check_fstab_mount_units FSTAB_MINIMAL "$OUT_DIR"
|
|
SYSTEMD_IN_INITRD=1 SYSTEMD_FSTAB="$FSTAB" SYSTEMD_PROC_CMDLINE="rd.fstab=0" run_and_list "$GENERATOR_BIN" "$OUT_DIR"
|
|
(! SYSTEMD_IN_INITRD=1 SYSTEMD_FSTAB="$FSTAB" check_fstab_mount_units FSTAB_MINIMAL "$OUT_DIR")
|
|
|
|
: "fstab-generator: kernel args - systemd.swap=0"
|
|
printf "%s\n" "${FSTAB_GENERAL_ROOT[@]}" >"$FSTAB"
|
|
cat "$FSTAB"
|
|
SYSTEMD_FSTAB="$FSTAB" SYSTEMD_PROC_CMDLINE="systemd.swap=0" run_and_list "$GENERATOR_BIN" "$OUT_DIR"
|
|
# No swap units should get created here
|
|
[[ "$(find "$OUT_DIR" -name "*.swap" | wc -l)" -eq 0 ]]
|
|
|
|
# Possible TODO
|
|
# - combine the rootfs & usrfs arguments and mix them with fstab entries
|
|
# - systemd.volatile=
|
|
: "fstab-generator: kernel args - root= + rootfstype= + rootflags="
|
|
# shellcheck disable=SC2034
|
|
EXPECTED_FSTAB=(
|
|
"/dev/disk/by-label/rootfs / ext4 noexec,ro 0 1"
|
|
)
|
|
CMDLINE="root=LABEL=rootfs rootfstype=ext4 rootflags=noexec"
|
|
SYSTEMD_IN_INITRD=1 SYSTEMD_FSTAB=/dev/null SYSTEMD_SYSROOT_FSTAB=/dev/null SYSTEMD_PROC_CMDLINE="$CMDLINE" run_and_list "$GENERATOR_BIN" "$OUT_DIR"
|
|
# The /proc/cmdline here is a dummy value to tell the in_initrd_host() function
|
|
# we're parsing host's fstab, but it's all on the kernel cmdline instead
|
|
SYSTEMD_IN_INITRD=1 SYSTEMD_SYSROOT_FSTAB=/proc/cmdline check_fstab_mount_units EXPECTED_FSTAB "$OUT_DIR"
|
|
|
|
# This is a very basic sanity test that involves manual checks, since adding it
|
|
# to the check_fstab_mount_units() function would make it way too complex
|
|
# (yet another possible TODO)
|
|
: "fstab-generator: kernel args - mount.usr= + mount.usrfstype= + mount.usrflags="
|
|
CMDLINE="mount.usr=UUID=be780f43-8803-4a76-9732-02ceda6e9808 mount.usrfstype=ext4 mount.usrflags=noexec,nodev"
|
|
SYSTEMD_IN_INITRD=1 SYSTEMD_FSTAB=/dev/null SYSTEMD_SYSROOT_FSTAB=/dev/null SYSTEMD_PROC_CMDLINE="$CMDLINE" run_and_list "$GENERATOR_BIN" "$OUT_DIR"
|
|
cat "$OUT_DIR/normal/sysroot-usr.mount" "$OUT_DIR/normal/sysusr-usr.mount"
|
|
# The general idea here is to mount the device to /sysusr/usr and then
|
|
# bind-mount /sysusr/usr to /sysroot/usr
|
|
grep -qE "^What=/dev/disk/by-uuid/be780f43-8803-4a76-9732-02ceda6e9808$" "$OUT_DIR/normal/sysusr-usr.mount"
|
|
grep -qE "^Where=/sysusr/usr$" "$OUT_DIR/normal/sysusr-usr.mount"
|
|
grep -qE "^Type=ext4$" "$OUT_DIR/normal/sysusr-usr.mount"
|
|
grep -qE "^Options=noexec,nodev,ro$" "$OUT_DIR/normal/sysusr-usr.mount"
|
|
link_eq "$OUT_DIR/normal/initrd-usr-fs.target.requires/sysusr-usr.mount" "../sysusr-usr.mount"
|
|
grep -qE "^What=/sysusr/usr$" "$OUT_DIR/normal/sysroot-usr.mount"
|
|
grep -qE "^Where=/sysroot/usr$" "$OUT_DIR/normal/sysroot-usr.mount"
|
|
grep -qE "^Options=bind$" "$OUT_DIR/normal/sysroot-usr.mount"
|
|
link_eq "$OUT_DIR/normal/initrd-fs.target.requires/sysroot-usr.mount" "../sysroot-usr.mount"
|