1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Functions to manage eBPF programs attached to cgroup subsystems
4  *
5  * Copyright 2022 Google LLC.
6  */
7 #include <asm-generic/errno.h>
8 #include <errno.h>
9 #include <sys/types.h>
10 #include <sys/mount.h>
11 #include <sys/stat.h>
12 #include <unistd.h>
13 
14 #include <test_progs.h>
15 #include <bpf/libbpf.h>
16 #include <bpf/bpf.h>
17 
18 #include "cgroup_helpers.h"
19 #include "cgroup_hierarchical_stats.skel.h"
20 
21 #define PAGE_SIZE 4096
22 #define MB(x) (x << 20)
23 
24 #define BPFFS_ROOT "/sys/fs/bpf/"
25 #define BPFFS_VMSCAN BPFFS_ROOT"vmscan/"
26 
27 #define CG_ROOT_NAME "root"
28 #define CG_ROOT_ID 1
29 
30 #define CGROUP_PATH(p, n) {.path = p"/"n, .name = n}
31 
32 static struct {
33 	const char *path, *name;
34 	unsigned long long id;
35 	int fd;
36 } cgroups[] = {
37 	CGROUP_PATH("/", "test"),
38 	CGROUP_PATH("/test", "child1"),
39 	CGROUP_PATH("/test", "child2"),
40 	CGROUP_PATH("/test/child1", "child1_1"),
41 	CGROUP_PATH("/test/child1", "child1_2"),
42 	CGROUP_PATH("/test/child2", "child2_1"),
43 	CGROUP_PATH("/test/child2", "child2_2"),
44 };
45 
46 #define N_CGROUPS ARRAY_SIZE(cgroups)
47 #define N_NON_LEAF_CGROUPS 3
48 
49 static int root_cgroup_fd;
50 static bool mounted_bpffs;
51 
52 /* reads file at 'path' to 'buf', returns 0 on success. */
53 static int read_from_file(const char *path, char *buf, size_t size)
54 {
55 	int fd, len;
56 
57 	fd = open(path, O_RDONLY);
58 	if (fd < 0)
59 		return fd;
60 
61 	len = read(fd, buf, size);
62 	close(fd);
63 	if (len < 0)
64 		return len;
65 
66 	buf[len] = 0;
67 	return 0;
68 }
69 
70 /* mounts bpffs and mkdir for reading stats, returns 0 on success. */
71 static int setup_bpffs(void)
72 {
73 	int err;
74 
75 	/* Mount bpffs */
76 	err = mount("bpf", BPFFS_ROOT, "bpf", 0, NULL);
77 	mounted_bpffs = !err;
78 	if (ASSERT_FALSE(err && errno != EBUSY, "mount"))
79 		return err;
80 
81 	/* Create a directory to contain stat files in bpffs */
82 	err = mkdir(BPFFS_VMSCAN, 0755);
83 	if (!ASSERT_OK(err, "mkdir"))
84 		return err;
85 
86 	return 0;
87 }
88 
89 static void cleanup_bpffs(void)
90 {
91 	/* Remove created directory in bpffs */
92 	ASSERT_OK(rmdir(BPFFS_VMSCAN), "rmdir "BPFFS_VMSCAN);
93 
94 	/* Unmount bpffs, if it wasn't already mounted when we started */
95 	if (mounted_bpffs)
96 		return;
97 
98 	ASSERT_OK(umount(BPFFS_ROOT), "unmount bpffs");
99 }
100 
101 /* sets up cgroups, returns 0 on success. */
102 static int setup_cgroups(void)
103 {
104 	int i, fd, err;
105 
106 	err = setup_cgroup_environment();
107 	if (!ASSERT_OK(err, "setup_cgroup_environment"))
108 		return err;
109 
110 	root_cgroup_fd = get_root_cgroup();
111 	if (!ASSERT_GE(root_cgroup_fd, 0, "get_root_cgroup"))
112 		return root_cgroup_fd;
113 
114 	for (i = 0; i < N_CGROUPS; i++) {
115 		fd = create_and_get_cgroup(cgroups[i].path);
116 		if (!ASSERT_GE(fd, 0, "create_and_get_cgroup"))
117 			return fd;
118 
119 		cgroups[i].fd = fd;
120 		cgroups[i].id = get_cgroup_id(cgroups[i].path);
121 
122 		/*
123 		 * Enable memcg controller for the entire hierarchy.
124 		 * Note that stats are collected for all cgroups in a hierarchy
125 		 * with memcg enabled anyway, but are only exposed for cgroups
126 		 * that have memcg enabled.
127 		 */
128 		if (i < N_NON_LEAF_CGROUPS) {
129 			err = enable_controllers(cgroups[i].path, "memory");
130 			if (!ASSERT_OK(err, "enable_controllers"))
131 				return err;
132 		}
133 	}
134 	return 0;
135 }
136 
137 static void cleanup_cgroups(void)
138 {
139 	close(root_cgroup_fd);
140 	for (int i = 0; i < N_CGROUPS; i++)
141 		close(cgroups[i].fd);
142 	cleanup_cgroup_environment();
143 }
144 
145 /* Sets up cgroup hiearchary, returns 0 on success. */
146 static int setup_hierarchy(void)
147 {
148 	return setup_bpffs() || setup_cgroups();
149 }
150 
151 static void destroy_hierarchy(void)
152 {
153 	cleanup_cgroups();
154 	cleanup_bpffs();
155 }
156 
157 static int reclaimer(const char *cgroup_path, size_t size)
158 {
159 	static char size_buf[128];
160 	char *buf, *ptr;
161 	int err;
162 
163 	/* Join cgroup in the parent process workdir */
164 	if (join_parent_cgroup(cgroup_path))
165 		return EACCES;
166 
167 	/* Allocate memory */
168 	buf = malloc(size);
169 	if (!buf)
170 		return ENOMEM;
171 
172 	/* Write to memory to make sure it's actually allocated */
173 	for (ptr = buf; ptr < buf + size; ptr += PAGE_SIZE)
174 		*ptr = 1;
175 
176 	/* Try to reclaim memory */
177 	snprintf(size_buf, 128, "%lu", size);
178 	err = write_cgroup_file_parent(cgroup_path, "memory.reclaim", size_buf);
179 
180 	free(buf);
181 	/* memory.reclaim returns EAGAIN if the amount is not fully reclaimed */
182 	if (err && errno != EAGAIN)
183 		return errno;
184 
185 	return 0;
186 }
187 
188 static int induce_vmscan(void)
189 {
190 	int i, status;
191 
192 	/*
193 	 * In every leaf cgroup, run a child process that allocates some memory
194 	 * and attempts to reclaim some of it.
195 	 */
196 	for (i = N_NON_LEAF_CGROUPS; i < N_CGROUPS; i++) {
197 		pid_t pid;
198 
199 		/* Create reclaimer child */
200 		pid = fork();
201 		if (pid == 0) {
202 			status = reclaimer(cgroups[i].path, MB(5));
203 			exit(status);
204 		}
205 
206 		/* Cleanup reclaimer child */
207 		waitpid(pid, &status, 0);
208 		ASSERT_TRUE(WIFEXITED(status), "reclaimer exited");
209 		ASSERT_EQ(WEXITSTATUS(status), 0, "reclaim exit code");
210 	}
211 	return 0;
212 }
213 
214 static unsigned long long
215 get_cgroup_vmscan_delay(unsigned long long cgroup_id, const char *file_name)
216 {
217 	unsigned long long vmscan = 0, id = 0;
218 	static char buf[128], path[128];
219 
220 	/* For every cgroup, read the file generated by cgroup_iter */
221 	snprintf(path, 128, "%s%s", BPFFS_VMSCAN, file_name);
222 	if (!ASSERT_OK(read_from_file(path, buf, 128), "read cgroup_iter"))
223 		return 0;
224 
225 	/* Check the output file formatting */
226 	ASSERT_EQ(sscanf(buf, "cg_id: %llu, total_vmscan_delay: %llu\n",
227 			 &id, &vmscan), 2, "output format");
228 
229 	/* Check that the cgroup_id is displayed correctly */
230 	ASSERT_EQ(id, cgroup_id, "cgroup_id");
231 	/* Check that the vmscan reading is non-zero */
232 	ASSERT_GT(vmscan, 0, "vmscan_reading");
233 	return vmscan;
234 }
235 
236 static void check_vmscan_stats(void)
237 {
238 	unsigned long long vmscan_readings[N_CGROUPS], vmscan_root;
239 	int i;
240 
241 	for (i = 0; i < N_CGROUPS; i++) {
242 		vmscan_readings[i] = get_cgroup_vmscan_delay(cgroups[i].id,
243 							     cgroups[i].name);
244 	}
245 
246 	/* Read stats for root too */
247 	vmscan_root = get_cgroup_vmscan_delay(CG_ROOT_ID, CG_ROOT_NAME);
248 
249 	/* Check that child1 == child1_1 + child1_2 */
250 	ASSERT_EQ(vmscan_readings[1], vmscan_readings[3] + vmscan_readings[4],
251 		  "child1_vmscan");
252 	/* Check that child2 == child2_1 + child2_2 */
253 	ASSERT_EQ(vmscan_readings[2], vmscan_readings[5] + vmscan_readings[6],
254 		  "child2_vmscan");
255 	/* Check that test == child1 + child2 */
256 	ASSERT_EQ(vmscan_readings[0], vmscan_readings[1] + vmscan_readings[2],
257 		  "test_vmscan");
258 	/* Check that root >= test */
259 	ASSERT_GE(vmscan_root, vmscan_readings[1], "root_vmscan");
260 }
261 
262 /* Creates iter link and pins in bpffs, returns 0 on success, -errno on failure.
263  */
264 static int setup_cgroup_iter(struct cgroup_hierarchical_stats *obj,
265 			     int cgroup_fd, const char *file_name)
266 {
267 	DECLARE_LIBBPF_OPTS(bpf_iter_attach_opts, opts);
268 	union bpf_iter_link_info linfo = {};
269 	struct bpf_link *link;
270 	static char path[128];
271 	int err;
272 
273 	/*
274 	 * Create an iter link, parameterized by cgroup_fd. We only want to
275 	 * traverse one cgroup, so set the traversal order to "self".
276 	 */
277 	linfo.cgroup.cgroup_fd = cgroup_fd;
278 	linfo.cgroup.order = BPF_CGROUP_ITER_SELF_ONLY;
279 	opts.link_info = &linfo;
280 	opts.link_info_len = sizeof(linfo);
281 	link = bpf_program__attach_iter(obj->progs.dump_vmscan, &opts);
282 	if (!ASSERT_OK_PTR(link, "attach_iter"))
283 		return -EFAULT;
284 
285 	/* Pin the link to a bpffs file */
286 	snprintf(path, 128, "%s%s", BPFFS_VMSCAN, file_name);
287 	err = bpf_link__pin(link, path);
288 	ASSERT_OK(err, "pin cgroup_iter");
289 
290 	/* Remove the link, leaving only the ref held by the pinned file */
291 	bpf_link__destroy(link);
292 	return err;
293 }
294 
295 /* Sets up programs for collecting stats, returns 0 on success. */
296 static int setup_progs(struct cgroup_hierarchical_stats **skel)
297 {
298 	int i, err;
299 
300 	*skel = cgroup_hierarchical_stats__open_and_load();
301 	if (!ASSERT_OK_PTR(*skel, "open_and_load"))
302 		return 1;
303 
304 	/* Attach cgroup_iter program that will dump the stats to cgroups */
305 	for (i = 0; i < N_CGROUPS; i++) {
306 		err = setup_cgroup_iter(*skel, cgroups[i].fd, cgroups[i].name);
307 		if (!ASSERT_OK(err, "setup_cgroup_iter"))
308 			return err;
309 	}
310 
311 	/* Also dump stats for root */
312 	err = setup_cgroup_iter(*skel, root_cgroup_fd, CG_ROOT_NAME);
313 	if (!ASSERT_OK(err, "setup_cgroup_iter"))
314 		return err;
315 
316 	bpf_program__set_autoattach((*skel)->progs.dump_vmscan, false);
317 	err = cgroup_hierarchical_stats__attach(*skel);
318 	if (!ASSERT_OK(err, "attach"))
319 		return err;
320 
321 	return 0;
322 }
323 
324 static void destroy_progs(struct cgroup_hierarchical_stats *skel)
325 {
326 	static char path[128];
327 	int i;
328 
329 	for (i = 0; i < N_CGROUPS; i++) {
330 		/* Delete files in bpffs that cgroup_iters are pinned in */
331 		snprintf(path, 128, "%s%s", BPFFS_VMSCAN,
332 			 cgroups[i].name);
333 		ASSERT_OK(remove(path), "remove cgroup_iter pin");
334 	}
335 
336 	/* Delete root file in bpffs */
337 	snprintf(path, 128, "%s%s", BPFFS_VMSCAN, CG_ROOT_NAME);
338 	ASSERT_OK(remove(path), "remove cgroup_iter root pin");
339 	cgroup_hierarchical_stats__destroy(skel);
340 }
341 
342 void test_cgroup_hierarchical_stats(void)
343 {
344 	struct cgroup_hierarchical_stats *skel = NULL;
345 
346 	if (setup_hierarchy())
347 		goto hierarchy_cleanup;
348 	if (setup_progs(&skel))
349 		goto cleanup;
350 	if (induce_vmscan())
351 		goto cleanup;
352 	check_vmscan_stats();
353 cleanup:
354 	destroy_progs(skel);
355 hierarchy_cleanup:
356 	destroy_hierarchy();
357 }
358