core: add DelegateSubgroup= setting

This implements a minimal subset of #24961, but in a lot more
restrictive way: we only allow one level of subcgroup (as that's enough
to address the no-processes in inner cgroups rule), and does not change
anything about threaded cgroup logic or similar, or make any of this new
behaviour mandatory.

All this does is this: all non-control processes we invoke for a unit
we'll invoke in a subgroup by the specified name.

We'll later port all our current services that use cgroup delegation
over to this, i.e. user@.service, systemd-nspawn@.service and
systemd-udevd.service.
This commit is contained in:
Lennart Poettering
2023-04-27 12:18:32 +02:00
parent 1a56b0c05d
commit a8b993dc11
11 changed files with 168 additions and 15 deletions
+34 -9
View File
@@ -4178,8 +4178,12 @@ static int compile_suggested_paths(const ExecContext *c, const ExecParameters *p
return 0;
}
static int exec_parameters_get_cgroup_path(const ExecParameters *params, char **ret) {
bool using_subcgroup;
static int exec_parameters_get_cgroup_path(
const ExecParameters *params,
const CGroupContext *c,
char **ret) {
const char *subgroup = NULL;
char *p;
assert(params);
@@ -4197,16 +4201,22 @@ static int exec_parameters_get_cgroup_path(const ExecParameters *params, char **
* this is not necessary, the cgroup is still empty. We distinguish these cases with the EXEC_CONTROL_CGROUP
* flag, which is only passed for the former statements, not for the latter. */
using_subcgroup = FLAGS_SET(params->flags, EXEC_CONTROL_CGROUP|EXEC_CGROUP_DELEGATE|EXEC_IS_CONTROL);
if (using_subcgroup)
p = path_join(params->cgroup_path, ".control");
if (FLAGS_SET(params->flags, EXEC_CGROUP_DELEGATE) && (FLAGS_SET(params->flags, EXEC_CONTROL_CGROUP) || c->delegate_subgroup)) {
if (FLAGS_SET(params->flags, EXEC_IS_CONTROL))
subgroup = ".control";
else
subgroup = c->delegate_subgroup;
}
if (subgroup)
p = path_join(params->cgroup_path, subgroup);
else
p = strdup(params->cgroup_path);
if (!p)
return -ENOMEM;
*ret = p;
return using_subcgroup;
return !!subgroup;
}
static int exec_context_cpu_affinity_from_numa(const ExecContext *c, CPUSet *ret) {
@@ -4705,7 +4715,7 @@ static int exec_child(
if (params->cgroup_path) {
_cleanup_free_ char *p = NULL;
r = exec_parameters_get_cgroup_path(params, &p);
r = exec_parameters_get_cgroup_path(params, cgroup_context, &p);
if (r < 0) {
*exit_status = EXIT_CGROUP;
return log_unit_error_errno(unit, r, "Failed to acquire cgroup path: %m");
@@ -4880,11 +4890,26 @@ static int exec_child(
* touch a single hierarchy too. */
if (params->flags & EXEC_CGROUP_DELEGATE) {
_cleanup_free_ char *p = NULL;
r = cg_set_access(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, uid, gid);
if (r < 0) {
*exit_status = EXIT_CGROUP;
return log_unit_error_errno(unit, r, "Failed to adjust control group access: %m");
}
r = exec_parameters_get_cgroup_path(params, cgroup_context, &p);
if (r < 0) {
*exit_status = EXIT_CGROUP;
return log_unit_error_errno(unit, r, "Failed to acquire cgroup path: %m");
}
if (r > 0) {
r = cg_set_access(SYSTEMD_CGROUP_CONTROLLER, p, uid, gid);
if (r < 0) {
*exit_status = EXIT_CGROUP;
return log_unit_error_errno(unit, r, "Failed to adjust control subgroup access: %m");
}
}
}
if (cgroup_context && cg_unified() > 0 && is_pressure_supported() > 0) {
@@ -5635,13 +5660,13 @@ int exec_spawn(Unit *unit,
log_command_line(unit, "About to execute", command->path, command->argv);
if (params->cgroup_path) {
r = exec_parameters_get_cgroup_path(params, &subcgroup_path);
r = exec_parameters_get_cgroup_path(params, cgroup_context, &subcgroup_path);
if (r < 0)
return log_unit_error_errno(unit, r, "Failed to acquire subcgroup path: %m");
if (r > 0) { /* We are using a child cgroup */
r = cg_create(SYSTEMD_CGROUP_CONTROLLER, subcgroup_path);
if (r < 0)
return log_unit_error_errno(unit, r, "Failed to create control group '%s': %m", subcgroup_path);
return log_unit_error_errno(unit, r, "Failed to create subcgroup '%s': %m", subcgroup_path);
/* Normally we would not propagate the xattrs to children but since we created this
* sub-cgroup internally we should do it. */