1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * This test makes sure BPF stats collection using rstat works correctly.
4 * The test uses 3 BPF progs:
5 * (a) counter: This BPF prog is invoked every time we attach a process to a
6 * cgroup and locklessly increments a percpu counter.
7 * The program then calls cgroup_rstat_updated() to inform rstat
8 * of an update on the (cpu, cgroup) pair.
9 *
10 * (b) flusher: This BPF prog is invoked when an rstat flush is ongoing, it
11 * aggregates all percpu counters to a total counter, and also
12 * propagates the changes to the ancestor cgroups.
13 *
14 * (c) dumper: This BPF prog is a cgroup_iter. It is used to output the total
15 * counter of a cgroup through reading a file in userspace.
16 *
17 * The test sets up a cgroup hierarchy, and the above programs. It spawns a few
18 * processes in the leaf cgroups and makes sure all the counters are aggregated
19 * correctly.
20 *
21 * Copyright 2022 Google LLC.
22 */
23#include <asm-generic/errno.h>
24#include <errno.h>
25#include <sys/types.h>
26#include <sys/mount.h>
27#include <sys/stat.h>
28#include <unistd.h>
29
30#include <test_progs.h>
31#include <bpf/libbpf.h>
32#include <bpf/bpf.h>
33
34#include "cgroup_helpers.h"
35#include "cgroup_hierarchical_stats.skel.h"
36
37#define PAGE_SIZE 4096
38#define MB(x) (x << 20)
39
40#define PROCESSES_PER_CGROUP 3
41
42#define BPFFS_ROOT "/sys/fs/bpf/"
43#define BPFFS_ATTACH_COUNTERS BPFFS_ROOT "attach_counters/"
44
45#define CG_ROOT_NAME "root"
46#define CG_ROOT_ID 1
47
48#define CGROUP_PATH(p, n) {.path = p"/"n, .name = n}
49
50static struct {
51 const char *path, *name;
52 unsigned long long id;
53 int fd;
54} cgroups[] = {
55 CGROUP_PATH("/", "test"),
56 CGROUP_PATH("/test", "child1"),
57 CGROUP_PATH("/test", "child2"),
58 CGROUP_PATH("/test/child1", "child1_1"),
59 CGROUP_PATH("/test/child1", "child1_2"),
60 CGROUP_PATH("/test/child2", "child2_1"),
61 CGROUP_PATH("/test/child2", "child2_2"),
62};
63
64#define N_CGROUPS ARRAY_SIZE(cgroups)
65#define N_NON_LEAF_CGROUPS 3
66
67static int root_cgroup_fd;
68static bool mounted_bpffs;
69
70/* reads file at 'path' to 'buf', returns 0 on success. */
71static int read_from_file(const char *path, char *buf, size_t size)
72{
73 int fd, len;
74
75 fd = open(path, O_RDONLY);
76 if (fd < 0)
77 return fd;
78
79 len = read(fd, buf, size);
80 close(fd);
81 if (len < 0)
82 return len;
83
84 buf[len] = 0;
85 return 0;
86}
87
88/* mounts bpffs and mkdir for reading stats, returns 0 on success. */
89static int setup_bpffs(void)
90{
91 int err;
92
93 /* Mount bpffs */
94 err = mount("bpf", BPFFS_ROOT, "bpf", 0, NULL);
95 mounted_bpffs = !err;
96 if (ASSERT_FALSE(err && errno != EBUSY, "mount"))
97 return err;
98
99 /* Create a directory to contain stat files in bpffs */
100 err = mkdir(BPFFS_ATTACH_COUNTERS, 0755);
101 if (!ASSERT_OK(err, "mkdir"))
102 return err;
103
104 return 0;
105}
106
107static void cleanup_bpffs(void)
108{
109 /* Remove created directory in bpffs */
110 ASSERT_OK(rmdir(BPFFS_ATTACH_COUNTERS), "rmdir "BPFFS_ATTACH_COUNTERS);
111
112 /* Unmount bpffs, if it wasn't already mounted when we started */
113 if (mounted_bpffs)
114 return;
115
116 ASSERT_OK(umount(BPFFS_ROOT), "unmount bpffs");
117}
118
119/* sets up cgroups, returns 0 on success. */
120static int setup_cgroups(void)
121{
122 int i, fd, err;
123
124 err = setup_cgroup_environment();
125 if (!ASSERT_OK(err, "setup_cgroup_environment"))
126 return err;
127
128 root_cgroup_fd = get_root_cgroup();
129 if (!ASSERT_GE(root_cgroup_fd, 0, "get_root_cgroup"))
130 return root_cgroup_fd;
131
132 for (i = 0; i < N_CGROUPS; i++) {
133 fd = create_and_get_cgroup(cgroups[i].path);
134 if (!ASSERT_GE(fd, 0, "create_and_get_cgroup"))
135 return fd;
136
137 cgroups[i].fd = fd;
138 cgroups[i].id = get_cgroup_id(cgroups[i].path);
139 }
140 return 0;
141}
142
143static void cleanup_cgroups(void)
144{
145 close(root_cgroup_fd);
146 for (int i = 0; i < N_CGROUPS; i++)
147 close(cgroups[i].fd);
148 cleanup_cgroup_environment();
149}
150
151/* Sets up cgroup hiearchary, returns 0 on success. */
152static int setup_hierarchy(void)
153{
154 return setup_bpffs() || setup_cgroups();
155}
156
157static void destroy_hierarchy(void)
158{
159 cleanup_cgroups();
160 cleanup_bpffs();
161}
162
163static int attach_processes(void)
164{
165 int i, j, status;
166
167 /* In every leaf cgroup, attach 3 processes */
168 for (i = N_NON_LEAF_CGROUPS; i < N_CGROUPS; i++) {
169 for (j = 0; j < PROCESSES_PER_CGROUP; j++) {
170 pid_t pid;
171
172 /* Create child and attach to cgroup */
173 pid = fork();
174 if (pid == 0) {
175 if (join_parent_cgroup(cgroups[i].path))
176 exit(EACCES);
177 exit(0);
178 }
179
180 /* Cleanup child */
181 waitpid(pid, &status, 0);
182 if (!ASSERT_TRUE(WIFEXITED(status), "child process exited"))
183 return 1;
184 if (!ASSERT_EQ(WEXITSTATUS(status), 0,
185 "child process exit code"))
186 return 1;
187 }
188 }
189 return 0;
190}
191
192static unsigned long long
193get_attach_counter(unsigned long long cgroup_id, const char *file_name)
194{
195 unsigned long long attach_counter = 0, id = 0;
196 static char buf[128], path[128];
197
198 /* For every cgroup, read the file generated by cgroup_iter */
199 snprintf(buf: path, size: 128, fmt: "%s%s", BPFFS_ATTACH_COUNTERS, file_name);
200 if (!ASSERT_OK(read_from_file(path, buf, size: 128), "read cgroup_iter"))
201 return 0;
202
203 /* Check the output file formatting */
204 ASSERT_EQ(sscanf(buf, "cg_id: %llu, attach_counter: %llu\n",
205 &id, &attach_counter), 2, "output format");
206
207 /* Check that the cgroup_id is displayed correctly */
208 ASSERT_EQ(id, cgroup_id, "cgroup_id");
209 /* Check that the counter is non-zero */
210 ASSERT_GT(attach_counter, 0, "attach counter non-zero");
211 return attach_counter;
212}
213
214static void check_attach_counters(void)
215{
216 unsigned long long attach_counters[N_CGROUPS], root_attach_counter;
217 int i;
218
219 for (i = 0; i < N_CGROUPS; i++)
220 attach_counters[i] = get_attach_counter(cgroup_id: cgroups[i].id,
221 file_name: cgroups[i].name);
222
223 /* Read stats for root too */
224 root_attach_counter = get_attach_counter(CG_ROOT_ID, CG_ROOT_NAME);
225
226 /* Check that all leafs cgroups have an attach counter of 3 */
227 for (i = N_NON_LEAF_CGROUPS; i < N_CGROUPS; i++)
228 ASSERT_EQ(attach_counters[i], PROCESSES_PER_CGROUP,
229 "leaf cgroup attach counter");
230
231 /* Check that child1 == child1_1 + child1_2 */
232 ASSERT_EQ(attach_counters[1], attach_counters[3] + attach_counters[4],
233 "child1_counter");
234 /* Check that child2 == child2_1 + child2_2 */
235 ASSERT_EQ(attach_counters[2], attach_counters[5] + attach_counters[6],
236 "child2_counter");
237 /* Check that test == child1 + child2 */
238 ASSERT_EQ(attach_counters[0], attach_counters[1] + attach_counters[2],
239 "test_counter");
240 /* Check that root >= test */
241 ASSERT_GE(root_attach_counter, attach_counters[1], "root_counter");
242}
243
244/* Creates iter link and pins in bpffs, returns 0 on success, -errno on failure.
245 */
246static int setup_cgroup_iter(struct cgroup_hierarchical_stats *obj,
247 int cgroup_fd, const char *file_name)
248{
249 DECLARE_LIBBPF_OPTS(bpf_iter_attach_opts, opts);
250 union bpf_iter_link_info linfo = {};
251 struct bpf_link *link;
252 static char path[128];
253 int err;
254
255 /*
256 * Create an iter link, parameterized by cgroup_fd. We only want to
257 * traverse one cgroup, so set the traversal order to "self".
258 */
259 linfo.cgroup.cgroup_fd = cgroup_fd;
260 linfo.cgroup.order = BPF_CGROUP_ITER_SELF_ONLY;
261 opts.link_info = &linfo;
262 opts.link_info_len = sizeof(linfo);
263 link = bpf_program__attach_iter(prog: obj->progs.dumper, opts: &opts);
264 if (!ASSERT_OK_PTR(link, "attach_iter"))
265 return -EFAULT;
266
267 /* Pin the link to a bpffs file */
268 snprintf(buf: path, size: 128, fmt: "%s%s", BPFFS_ATTACH_COUNTERS, file_name);
269 err = bpf_link__pin(link, path);
270 ASSERT_OK(err, "pin cgroup_iter");
271
272 /* Remove the link, leaving only the ref held by the pinned file */
273 bpf_link__destroy(link);
274 return err;
275}
276
277/* Sets up programs for collecting stats, returns 0 on success. */
278static int setup_progs(struct cgroup_hierarchical_stats **skel)
279{
280 int i, err;
281
282 *skel = cgroup_hierarchical_stats__open_and_load();
283 if (!ASSERT_OK_PTR(*skel, "open_and_load"))
284 return 1;
285
286 /* Attach cgroup_iter program that will dump the stats to cgroups */
287 for (i = 0; i < N_CGROUPS; i++) {
288 err = setup_cgroup_iter(obj: *skel, cgroup_fd: cgroups[i].fd, file_name: cgroups[i].name);
289 if (!ASSERT_OK(err, "setup_cgroup_iter"))
290 return err;
291 }
292
293 /* Also dump stats for root */
294 err = setup_cgroup_iter(obj: *skel, cgroup_fd: root_cgroup_fd, CG_ROOT_NAME);
295 if (!ASSERT_OK(err, "setup_cgroup_iter"))
296 return err;
297
298 bpf_program__set_autoattach(prog: (*skel)->progs.dumper, autoattach: false);
299 err = cgroup_hierarchical_stats__attach(*skel);
300 if (!ASSERT_OK(err, "attach"))
301 return err;
302
303 return 0;
304}
305
306static void destroy_progs(struct cgroup_hierarchical_stats *skel)
307{
308 static char path[128];
309 int i;
310
311 for (i = 0; i < N_CGROUPS; i++) {
312 /* Delete files in bpffs that cgroup_iters are pinned in */
313 snprintf(buf: path, size: 128, fmt: "%s%s", BPFFS_ATTACH_COUNTERS,
314 cgroups[i].name);
315 ASSERT_OK(remove(path), "remove cgroup_iter pin");
316 }
317
318 /* Delete root file in bpffs */
319 snprintf(buf: path, size: 128, fmt: "%s%s", BPFFS_ATTACH_COUNTERS, CG_ROOT_NAME);
320 ASSERT_OK(remove(path), "remove cgroup_iter root pin");
321 cgroup_hierarchical_stats__destroy(skel);
322}
323
324void test_cgroup_hierarchical_stats(void)
325{
326 struct cgroup_hierarchical_stats *skel = NULL;
327
328 if (setup_hierarchy())
329 goto hierarchy_cleanup;
330 if (setup_progs(&skel))
331 goto cleanup;
332 if (attach_processes())
333 goto cleanup;
334 check_attach_counters();
335cleanup:
336 destroy_progs(skel);
337hierarchy_cleanup:
338 destroy_hierarchy();
339}
340

source code of linux/tools/testing/selftests/bpf/prog_tests/cgroup_hierarchical_stats.c