Merge pull request #27766 from rphibel/cleanup-cgroups-before-cleaning-units

Don't GC unit if it is in cgroup_empty_queue

src/core/cgroup.c

@@ -3142,7 +3142,9 @@ static int on_cgroup_empty_event(sd_event_source *s, void *userdata) {
 
         unit_add_to_gc_queue(u);
 
-        if (UNIT_VTABLE(u)->notify_cgroup_empty)
+        if (IN_SET(unit_active_state(u), UNIT_INACTIVE, UNIT_FAILED))
+                unit_prune_cgroup(u);
+        else if (UNIT_VTABLE(u)->notify_cgroup_empty)
                 UNIT_VTABLE(u)->notify_cgroup_empty(u);
 
         return 0;

src/core/scope.c

@@ -629,11 +629,6 @@ static void scope_notify_cgroup_empty_event(Unit *u) {
 
         if (IN_SET(s->state, SCOPE_RUNNING, SCOPE_ABANDONED, SCOPE_STOP_SIGTERM, SCOPE_STOP_SIGKILL))
                 scope_enter_dead(s, SCOPE_SUCCESS);
-
-        /* If the cgroup empty notification comes when the unit is not active, we must have failed to clean
-         * up the cgroup earlier and should do it now. */
-        if (IN_SET(s->state, SCOPE_DEAD, SCOPE_FAILED))
-                unit_prune_cgroup(u);
 }
 
 static void scope_notify_cgroup_oom_event(Unit *u, bool managed_oom) {

src/core/service.c

@@ -3648,12 +3648,7 @@ static void service_notify_cgroup_empty_event(Unit *u) {
 
         /* If the cgroup empty notification comes when the unit is not active, we must have failed to clean
          * up the cgroup earlier and should do it now. */
-        case SERVICE_DEAD:
-        case SERVICE_FAILED:
-        case SERVICE_DEAD_BEFORE_AUTO_RESTART:
-        case SERVICE_FAILED_BEFORE_AUTO_RESTART:
         case SERVICE_AUTO_RESTART:
-        case SERVICE_DEAD_RESOURCES_PINNED:
                 unit_prune_cgroup(u);
                 break;
 

src/core/unit.c

@@ -441,6 +441,9 @@ bool unit_may_gc(Unit *u) {
         if (u->perpetual)
                 return false;
 
+        if (u->in_cgroup_empty_queue)
+                return false;
+
         if (sd_bus_track_count(u->bus_track) > 0)
                 return false;
 

src/test/test-execute.c

@@ -206,6 +206,17 @@ static bool is_inaccessible_available(void) {
         return true;
 }
 
+static void start_parent_slices(Unit *unit) {
+        Unit *slice;
+
+        slice = UNIT_GET_SLICE(unit);
+        if (slice) {
+                start_parent_slices(slice);
+                int r = unit_start(slice, NULL);
+                assert_se(r >= 0 || r == -EALREADY);
+        }
+}
+
 static void _test(const char *file, unsigned line, const char *func,
                   Manager *m, const char *unit_name, int status_expected, int code_expected) {
         Unit *unit;
@@ -213,6 +224,9 @@ static void _test(const char *file, unsigned line, const char *func,
         assert_se(unit_name);
 
         assert_se(manager_load_startable_unit_or_warn(m, unit_name, NULL, &unit) >= 0);
+        /* We need to start the slices as well otherwise the slice cgroups might be pruned
+         * in on_cgroup_empty_event. */
+        start_parent_slices(unit);
         assert_se(unit_start(unit, NULL) >= 0);
         check_main_result(file, line, func, m, unit, status_expected, code_expected);
 }

test/units/testsuite-19.cleanup-slice.sh

@@ -0,0 +1,49 @@
+#!/usr/bin/env bash
+# SPDX-License-Identifier: LGPL-2.1-or-later
+set -eux
+set -o pipefail
+
+# shellcheck source=test/units/util.sh
+. "$(dirname "$0")"/util.sh
+
+export SYSTEMD_LOG_LEVEL=debug
+
+# Create service with KillMode=none inside a slice
+cat <<EOF >/run/systemd/system/test19cleanup.service
+[Unit]
+Description=Test 19 cleanup Service
+[Service]
+Slice=test19cleanup.slice
+Type=exec
+ExecStart=sleep infinity
+KillMode=none
+EOF
+cat <<EOF >/run/systemd/system/test19cleanup.slice
+[Unit]
+Description=Test 19 cleanup Slice
+EOF
+
+# Start service
+systemctl start test19cleanup.service
+assert_rc 0 systemd-cgls /test19cleanup.slice
+
+pid=$(systemctl show --property MainPID --value test19cleanup)
+ps "$pid"
+
+# Stop slice
+# The sleep process will not be killed because of KillMode=none
+# Since there is still a process running under it, the /test19cleanup.slice cgroup won't be removed
+systemctl stop test19cleanup.slice
+
+ps "$pid"
+
+# Kill sleep process manually
+kill -s TERM "$pid"
+while kill -0 "$pid" 2>/dev/null; do sleep 0.1; done
+
+timeout 30 bash -c 'while systemd-cgls /test19cleanup.slice/test19cleanup.service >& /dev/null; do sleep .5; done'
+assert_rc 1 systemd-cgls /test19cleanup.slice/test19cleanup.service
+
+# Check that empty cgroup /test19cleanup.slice has been removed
+timeout 30 bash -c 'while systemd-cgls /test19cleanup.slice >& /dev/null; do sleep .5; done'
+assert_rc 1 systemd-cgls /test19cleanup.slice