[RFC PATCH 2/2] bpf: Add OOM policy test

From: Chuyi Zhou
Date: Fri Aug 04 2023 - 05:39:17 EST


This patch adds a test which implements a priority-based policy through
bpf_select_task.

The BPF program, oom_policy.c, compares the cgroup priority of two tasks
and select the lower one. The userspace program test_oom_policy.c
maintains a priority map by using cgroup id as the keys and priority as
the values. We could protect certain cgroups from oom-killer by setting
higher priority.

Signed-off-by: Chuyi Zhou <zhouchuyi@xxxxxxxxxxxxx>
---
.../bpf/prog_tests/test_oom_policy.c | 140 ++++++++++++++++++
.../testing/selftests/bpf/progs/oom_policy.c | 77 ++++++++++
2 files changed, 217 insertions(+)
create mode 100644 tools/testing/selftests/bpf/prog_tests/test_oom_policy.c
create mode 100644 tools/testing/selftests/bpf/progs/oom_policy.c

diff --git a/tools/testing/selftests/bpf/prog_tests/test_oom_policy.c b/tools/testing/selftests/bpf/prog_tests/test_oom_policy.c
new file mode 100644
index 000000000000..2400cc48ba83
--- /dev/null
+++ b/tools/testing/selftests/bpf/prog_tests/test_oom_policy.c
@@ -0,0 +1,140 @@
+// SPDX-License-Identifier: GPL-2.0-only
+#define _GNU_SOURCE
+
+#include <stdio.h>
+#include <fcntl.h>
+#include <unistd.h>
+#include <stdlib.h>
+#include <signal.h>
+#include <sys/stat.h>
+#include <test_progs.h>
+#include <bpf/btf.h>
+#include <bpf/bpf.h>
+
+#include "cgroup_helpers.h"
+#include "oom_policy.skel.h"
+
+static int map_fd;
+static int cg_nr;
+struct {
+ const char *path;
+ int fd;
+ unsigned long long id;
+} cgs[] = {
+ { "/cg1" },
+ { "/cg2" },
+};
+
+
+static struct oom_policy *open_load_oom_policy_skel(void)
+{
+ struct oom_policy *skel;
+ int err;
+
+ skel = oom_policy__open();
+ if (!ASSERT_OK_PTR(skel, "skel_open"))
+ return NULL;
+
+ err = oom_policy__load(skel);
+ if (!ASSERT_OK(err, "skel_load"))
+ goto cleanup;
+
+ return skel;
+
+cleanup:
+ oom_policy__destroy(skel);
+ return NULL;
+}
+
+static void run_memory_consume(unsigned long long consume_size, int idx)
+{
+ char *buf;
+
+ join_parent_cgroup(cgs[idx].path);
+ buf = malloc(consume_size);
+ memset(buf, 0, consume_size);
+ sleep(2);
+ exit(0);
+}
+
+static int set_cgroup_prio(unsigned long long cg_id, int prio)
+{
+ int err;
+
+ err = bpf_map_update_elem(map_fd, &cg_id, &prio, BPF_ANY);
+ ASSERT_EQ(err, 0, "update_map");
+ return err;
+}
+
+static int prepare_cgroup_environment(void)
+{
+ int err;
+
+ err = setup_cgroup_environment();
+ if (err)
+ goto clean_cg_env;
+ for (int i = 0; i < cg_nr; i++) {
+ err = cgs[i].fd = create_and_get_cgroup(cgs[i].path);
+ if (!ASSERT_GE(cgs[i].fd, 0, "cg_create"))
+ goto clean_cg_env;
+ cgs[i].id = get_cgroup_id(cgs[i].path);
+ }
+ return 0;
+clean_cg_env:
+ cleanup_cgroup_environment();
+ return err;
+}
+
+void test_oom_policy(void)
+{
+ struct oom_policy *skel;
+ struct bpf_link *link;
+ int err;
+ int victim_pid;
+ unsigned long long victim_cg_id;
+
+ link = NULL;
+ cg_nr = ARRAY_SIZE(cgs);
+
+ skel = open_load_oom_policy_skel();
+ err = oom_policy__attach(skel);
+ if (!ASSERT_OK(err, "oom_policy__attach"))
+ goto cleanup;
+
+ map_fd = bpf_object__find_map_fd_by_name(skel->obj, "cg_map");
+ if (!ASSERT_GE(map_fd, 0, "find map"))
+ goto cleanup;
+
+ err = prepare_cgroup_environment();
+ if (!ASSERT_EQ(err, 0, "prepare cgroup env"))
+ goto cleanup;
+
+ write_cgroup_file("/", "memory.max", "10M");
+
+ /*
+ * Set higher priority to cg2 and lower to cg1, so we would select
+ * task under cg1 as victim.(see oom_policy.c)
+ */
+ set_cgroup_prio(cgs[0].id, 10);
+ set_cgroup_prio(cgs[1].id, 50);
+
+ victim_cg_id = cgs[0].id;
+ victim_pid = fork();
+
+ if (victim_pid == 0)
+ run_memory_consume(1024 * 1024 * 4, 0);
+
+ if (fork() == 0)
+ run_memory_consume(1024 * 1024 * 8, 1);
+
+ while (wait(NULL) > 0)
+ ;
+
+ ASSERT_EQ(skel->bss->victim_pid, victim_pid, "victim_pid");
+ ASSERT_EQ(skel->bss->victim_cg_id, victim_cg_id, "victim_cgid");
+
+cleanup:
+ bpf_link__destroy(link);
+ oom_policy__destroy(skel);
+ cleanup_cgroup_environment();
+}
diff --git a/tools/testing/selftests/bpf/progs/oom_policy.c b/tools/testing/selftests/bpf/progs/oom_policy.c
new file mode 100644
index 000000000000..d269ea52bcb2
--- /dev/null
+++ b/tools/testing/selftests/bpf/progs/oom_policy.c
@@ -0,0 +1,77 @@
+// SPDX-License-Identifier: GPL-2.0-only
+#include <vmlinux.h>
+#include <bpf/bpf_tracing.h>
+#include <bpf/bpf_helpers.h>
+
+char _license[] SEC("license") = "GPL";
+struct {
+ __uint(type, BPF_MAP_TYPE_HASH);
+ __type(key, int);
+ __type(value, int);
+ __uint(max_entries, 24);
+} cg_map SEC(".maps");
+
+unsigned int victim_pid;
+u64 victim_cg_id;
+
+enum bpf_select_ret {
+ BPF_SELECT_DISABLE,
+ BPF_SELECT_TASK,
+ BPF_SELECT_CHOSEN,
+};
+
+static __always_inline u64 task_cgroup_id(struct task_struct *task)
+{
+ struct kernfs_node *node;
+ struct task_group *tg;
+
+ if (!task)
+ return 0;
+
+ tg = task->sched_task_group;
+ node = tg->css.cgroup->kn;
+
+ return node->id;
+}
+
+SEC("fentry/oom_kill_process")
+int BPF_PROG(oom_kill_process_k, struct oom_control *oc, const char *message)
+{
+ struct task_struct *victim = oc->chosen;
+
+ if (!victim)
+ return 0;
+
+ victim_pid = victim->pid;
+ victim_cg_id = task_cgroup_id(victim);
+ return 0;
+}
+
+SEC("fmod_ret/bpf_select_task")
+int BPF_PROG(select_task_test, struct oom_control *oc, struct task_struct *task, long points)
+{
+ u64 chosen_cg_id, task_cg_id;
+ int chosen_cg_prio, task_cg_prio;
+ struct task_struct *chosen;
+ int *val;
+
+ chosen = oc->chosen;
+ chosen_cg_id = task_cgroup_id(chosen);
+ task_cg_id = task_cgroup_id(task);
+ chosen_cg_prio = task_cg_prio = 0;
+ val = bpf_map_lookup_elem(&cg_map, &chosen_cg_id);
+ if (val)
+ chosen_cg_prio = *val;
+ val = bpf_map_lookup_elem(&cg_map, &task_cg_id);
+ if (val)
+ task_cg_prio = *val;
+
+ if (chosen_cg_prio > task_cg_prio)
+ return BPF_SELECT_TASK;
+ if (chosen_cg_prio < task_cg_prio)
+ return BPF_SELECT_CHOSEN;
+
+ return BPF_SELECT_DISABLE;
+
+}
+
--
2.20.1