Files
systemd/test/units/TEST-19-CGROUP.delegate.sh
Lennart Poettering 1d5fd2e621 test: add testcase that verifies we can safely delete subcgroups owned by other users if we own the parent
This is a test for the previous commits: we create an unpriv, delegated cgroup in
--user mode, then create a subcgroup that is owned by some other user
(to mimic the case where an unpriv user got a userns with delegated UIDs
assigned), and then try to stop the unit. traditionally this would fail,
because our unpriv systemd --user instance can't remove the subcrroup
owned by someone else. With the earlier patches this is addressed.
2025-01-08 15:27:25 +01:00

158 lines
6.2 KiB
Bash
Executable File

#!/usr/bin/env bash
# SPDX-License-Identifier: LGPL-2.1-or-later
# shellcheck disable=SC2235
set -eux
set -o pipefail
# Test cgroup delegation in the unified hierarchy
# shellcheck source=test/units/test-control.sh
. "$(dirname "$0")"/test-control.sh
# shellcheck source=test/units/util.sh
. "$(dirname "$0")"/util.sh
if [[ "$(get_cgroup_hierarchy)" != unified ]]; then
echo "Skipping $0 as we're not running with the unified cgroup hierarchy"
exit 0
fi
testcase_controllers() {
systemd-run --wait \
--unit=test-0.service \
--property="DynamicUser=1" \
--property="Delegate=" \
test -w /sys/fs/cgroup/system.slice/test-0.service/ -a \
-w /sys/fs/cgroup/system.slice/test-0.service/cgroup.procs -a \
-w /sys/fs/cgroup/system.slice/test-0.service/cgroup.subtree_control
systemd-run --wait \
--unit=test-1.service \
--property="DynamicUser=1" \
--property="Delegate=memory pids" \
grep -q memory /sys/fs/cgroup/system.slice/test-1.service/cgroup.controllers
systemd-run --wait \
--unit=test-2.service \
--property="DynamicUser=1" \
--property="Delegate=memory pids" \
grep -q pids /sys/fs/cgroup/system.slice/test-2.service/cgroup.controllers
# "io" is not among the controllers enabled by default for all units, verify that
grep -qv io /sys/fs/cgroup/system.slice/cgroup.controllers
# Run a service with "io" enabled, and verify it works
systemd-run --wait \
--unit=test-3.service \
--property="IOAccounting=yes" \
--property="Slice=system-foo-bar-baz.slice" \
grep -q io /sys/fs/cgroup/system.slice/system-foo.slice/system-foo-bar.slice/system-foo-bar-baz.slice/test-3.service/cgroup.controllers
# We want to check if "io" is removed again from the controllers
# list. However, PID 1 (rightfully) does this asynchronously. In order
# to force synchronization on this, let's start a short-lived service
# which requires PID 1 to refresh the cgroup tree, so that we can
# verify that this all works.
systemd-run --wait --unit=test-4.service true
# And now check again, "io" should have vanished
grep -qv io /sys/fs/cgroup/system.slice/cgroup.controllers
}
testcase_attributes() {
# Test if delegation also works for some of the more recent attrs the kernel might or might not support
for attr in cgroup.threads memory.oom.group memory.reclaim ; do
if grep -q "$attr" /sys/kernel/cgroup/delegate ; then
systemd-run --wait \
--unit=test-0.service \
--property="MemoryAccounting=1" \
--property="DynamicUser=1" \
--property="Delegate=" \
test -w /sys/fs/cgroup/system.slice/test-0.service/ -a \
-w /sys/fs/cgroup/system.slice/test-0.service/"$attr"
fi
done
}
testcase_scope_unpriv_delegation() {
# Check that unprivileged delegation works for scopes
useradd test
trap "userdel -r test" RETURN
systemd-run --uid=test \
--property="User=test" \
--property="Delegate=yes" \
--slice workload.slice \
--unit test-workload0.scope\
--scope \
test -w /sys/fs/cgroup/workload.slice/test-workload0.scope -a \
-w /sys/fs/cgroup/workload.slice/test-workload0.scope/cgroup.procs -a \
-w /sys/fs/cgroup/workload.slice/test-workload0.scope/cgroup.subtree_control
}
testcase_user_unpriv_delegation() {
# Check that delegation works for unpriv users, and that we can insert a
# subcgroup owned by a different user (which can happen in case unpriv
# userns where a UID range was delegated), which is still cleaned up
# correctly when it goes down.
run0 -u testuser systemd-run --user \
--property="Delegate=yes" \
--unit=test-chown-subcgroup \
--service-type=exec \
sleep infinity
TESTUID=$(id -u testuser)
CGROUP="/sys/fs/cgroup/user.slice/user-$TESTUID.slice/user@$TESTUID.service/app.slice/test-chown-subcgroup.service"
test -d "$CGROUP"
# Create a subcgroup, and make it owned by some unrelated user
SUBCGROUP="$CGROUP/subcgroup"
mkdir "$SUBCGROUP"
chown 1:1 "$SUBCGROUP"
# Make sure the subcgroup is not empty (empty dirs owned by other users can
# be removed if one owns the dir they are contained in, after all)
mkdir "$SUBCGROUP"/filler
run0 -u testuser systemctl stop --user test-chown-subcgroup.service
# Verify that the subcgroup got correctly removed
(! test -e "$CGROUP")
systemctl stop user@testuser.service
}
testcase_subgroup() {
# Verify that DelegateSubgroup= affects ownership correctly
unit="test-subgroup-$RANDOM.service"
systemd-run --wait \
--unit="$unit" \
--property="DynamicUser=1" \
--property="Delegate=pids" \
--property="DelegateSubgroup=foo" \
test -w "/sys/fs/cgroup/system.slice/$unit" -a \
-w "/sys/fs/cgroup/system.slice/$unit/foo"
# Check that for the subgroup also attributes that aren't covered by
# regular (i.e. main cgroup) delegation ownership rules are delegated properly
if test -f /sys/fs/cgroup/cgroup.max.depth; then
unit="test-subgroup-$RANDOM.service"
systemd-run --wait \
--unit="$unit" \
--property="DynamicUser=1" \
--property="Delegate=pids" \
--property="DelegateSubgroup=zzz" \
test -w "/sys/fs/cgroup/system.slice/$unit/zzz/cgroup.max.depth"
fi
# Check that the invoked process itself is also in the subgroup
unit="test-subgroup-$RANDOM.service"
systemd-run --wait \
--unit="$unit" \
--property="DynamicUser=1" \
--property="Delegate=pids" \
--property="DelegateSubgroup=bar" \
grep -q -x -F "0::/system.slice/$unit/bar" /proc/self/cgroup
}
run_testcases