162306a36Sopenharmony_ci/* SPDX-License-Identifier: GPL-2.0 */
262306a36Sopenharmony_ci
362306a36Sopenharmony_ci#define _GNU_SOURCE
462306a36Sopenharmony_ci
562306a36Sopenharmony_ci#include <errno.h>
662306a36Sopenharmony_ci#include <fcntl.h>
762306a36Sopenharmony_ci#include <linux/limits.h>
862306a36Sopenharmony_ci#include <poll.h>
962306a36Sopenharmony_ci#include <signal.h>
1062306a36Sopenharmony_ci#include <stdio.h>
1162306a36Sopenharmony_ci#include <stdlib.h>
1262306a36Sopenharmony_ci#include <string.h>
1362306a36Sopenharmony_ci#include <sys/inotify.h>
1462306a36Sopenharmony_ci#include <sys/stat.h>
1562306a36Sopenharmony_ci#include <sys/types.h>
1662306a36Sopenharmony_ci#include <sys/wait.h>
1762306a36Sopenharmony_ci#include <unistd.h>
1862306a36Sopenharmony_ci
1962306a36Sopenharmony_ci#include "cgroup_util.h"
2062306a36Sopenharmony_ci#include "../clone3/clone3_selftests.h"
2162306a36Sopenharmony_ci
2262306a36Sopenharmony_ci/* Returns read len on success, or -errno on failure. */
2362306a36Sopenharmony_cistatic ssize_t read_text(const char *path, char *buf, size_t max_len)
2462306a36Sopenharmony_ci{
2562306a36Sopenharmony_ci	ssize_t len;
2662306a36Sopenharmony_ci	int fd;
2762306a36Sopenharmony_ci
2862306a36Sopenharmony_ci	fd = open(path, O_RDONLY);
2962306a36Sopenharmony_ci	if (fd < 0)
3062306a36Sopenharmony_ci		return -errno;
3162306a36Sopenharmony_ci
3262306a36Sopenharmony_ci	len = read(fd, buf, max_len - 1);
3362306a36Sopenharmony_ci
3462306a36Sopenharmony_ci	if (len >= 0)
3562306a36Sopenharmony_ci		buf[len] = 0;
3662306a36Sopenharmony_ci
3762306a36Sopenharmony_ci	close(fd);
3862306a36Sopenharmony_ci	return len < 0 ? -errno : len;
3962306a36Sopenharmony_ci}
4062306a36Sopenharmony_ci
4162306a36Sopenharmony_ci/* Returns written len on success, or -errno on failure. */
4262306a36Sopenharmony_cistatic ssize_t write_text(const char *path, char *buf, ssize_t len)
4362306a36Sopenharmony_ci{
4462306a36Sopenharmony_ci	int fd;
4562306a36Sopenharmony_ci
4662306a36Sopenharmony_ci	fd = open(path, O_WRONLY | O_APPEND);
4762306a36Sopenharmony_ci	if (fd < 0)
4862306a36Sopenharmony_ci		return -errno;
4962306a36Sopenharmony_ci
5062306a36Sopenharmony_ci	len = write(fd, buf, len);
5162306a36Sopenharmony_ci	close(fd);
5262306a36Sopenharmony_ci	return len < 0 ? -errno : len;
5362306a36Sopenharmony_ci}
5462306a36Sopenharmony_ci
5562306a36Sopenharmony_cichar *cg_name(const char *root, const char *name)
5662306a36Sopenharmony_ci{
5762306a36Sopenharmony_ci	size_t len = strlen(root) + strlen(name) + 2;
5862306a36Sopenharmony_ci	char *ret = malloc(len);
5962306a36Sopenharmony_ci
6062306a36Sopenharmony_ci	snprintf(ret, len, "%s/%s", root, name);
6162306a36Sopenharmony_ci
6262306a36Sopenharmony_ci	return ret;
6362306a36Sopenharmony_ci}
6462306a36Sopenharmony_ci
6562306a36Sopenharmony_cichar *cg_name_indexed(const char *root, const char *name, int index)
6662306a36Sopenharmony_ci{
6762306a36Sopenharmony_ci	size_t len = strlen(root) + strlen(name) + 10;
6862306a36Sopenharmony_ci	char *ret = malloc(len);
6962306a36Sopenharmony_ci
7062306a36Sopenharmony_ci	snprintf(ret, len, "%s/%s_%d", root, name, index);
7162306a36Sopenharmony_ci
7262306a36Sopenharmony_ci	return ret;
7362306a36Sopenharmony_ci}
7462306a36Sopenharmony_ci
7562306a36Sopenharmony_cichar *cg_control(const char *cgroup, const char *control)
7662306a36Sopenharmony_ci{
7762306a36Sopenharmony_ci	size_t len = strlen(cgroup) + strlen(control) + 2;
7862306a36Sopenharmony_ci	char *ret = malloc(len);
7962306a36Sopenharmony_ci
8062306a36Sopenharmony_ci	snprintf(ret, len, "%s/%s", cgroup, control);
8162306a36Sopenharmony_ci
8262306a36Sopenharmony_ci	return ret;
8362306a36Sopenharmony_ci}
8462306a36Sopenharmony_ci
8562306a36Sopenharmony_ci/* Returns 0 on success, or -errno on failure. */
8662306a36Sopenharmony_ciint cg_read(const char *cgroup, const char *control, char *buf, size_t len)
8762306a36Sopenharmony_ci{
8862306a36Sopenharmony_ci	char path[PATH_MAX];
8962306a36Sopenharmony_ci	ssize_t ret;
9062306a36Sopenharmony_ci
9162306a36Sopenharmony_ci	snprintf(path, sizeof(path), "%s/%s", cgroup, control);
9262306a36Sopenharmony_ci
9362306a36Sopenharmony_ci	ret = read_text(path, buf, len);
9462306a36Sopenharmony_ci	return ret >= 0 ? 0 : ret;
9562306a36Sopenharmony_ci}
9662306a36Sopenharmony_ci
9762306a36Sopenharmony_ciint cg_read_strcmp(const char *cgroup, const char *control,
9862306a36Sopenharmony_ci		   const char *expected)
9962306a36Sopenharmony_ci{
10062306a36Sopenharmony_ci	size_t size;
10162306a36Sopenharmony_ci	char *buf;
10262306a36Sopenharmony_ci	int ret;
10362306a36Sopenharmony_ci
10462306a36Sopenharmony_ci	/* Handle the case of comparing against empty string */
10562306a36Sopenharmony_ci	if (!expected)
10662306a36Sopenharmony_ci		return -1;
10762306a36Sopenharmony_ci	else
10862306a36Sopenharmony_ci		size = strlen(expected) + 1;
10962306a36Sopenharmony_ci
11062306a36Sopenharmony_ci	buf = malloc(size);
11162306a36Sopenharmony_ci	if (!buf)
11262306a36Sopenharmony_ci		return -1;
11362306a36Sopenharmony_ci
11462306a36Sopenharmony_ci	if (cg_read(cgroup, control, buf, size)) {
11562306a36Sopenharmony_ci		free(buf);
11662306a36Sopenharmony_ci		return -1;
11762306a36Sopenharmony_ci	}
11862306a36Sopenharmony_ci
11962306a36Sopenharmony_ci	ret = strcmp(expected, buf);
12062306a36Sopenharmony_ci	free(buf);
12162306a36Sopenharmony_ci	return ret;
12262306a36Sopenharmony_ci}
12362306a36Sopenharmony_ci
12462306a36Sopenharmony_ciint cg_read_strstr(const char *cgroup, const char *control, const char *needle)
12562306a36Sopenharmony_ci{
12662306a36Sopenharmony_ci	char buf[PAGE_SIZE];
12762306a36Sopenharmony_ci
12862306a36Sopenharmony_ci	if (cg_read(cgroup, control, buf, sizeof(buf)))
12962306a36Sopenharmony_ci		return -1;
13062306a36Sopenharmony_ci
13162306a36Sopenharmony_ci	return strstr(buf, needle) ? 0 : -1;
13262306a36Sopenharmony_ci}
13362306a36Sopenharmony_ci
13462306a36Sopenharmony_cilong cg_read_long(const char *cgroup, const char *control)
13562306a36Sopenharmony_ci{
13662306a36Sopenharmony_ci	char buf[128];
13762306a36Sopenharmony_ci
13862306a36Sopenharmony_ci	if (cg_read(cgroup, control, buf, sizeof(buf)))
13962306a36Sopenharmony_ci		return -1;
14062306a36Sopenharmony_ci
14162306a36Sopenharmony_ci	return atol(buf);
14262306a36Sopenharmony_ci}
14362306a36Sopenharmony_ci
14462306a36Sopenharmony_cilong cg_read_key_long(const char *cgroup, const char *control, const char *key)
14562306a36Sopenharmony_ci{
14662306a36Sopenharmony_ci	char buf[PAGE_SIZE];
14762306a36Sopenharmony_ci	char *ptr;
14862306a36Sopenharmony_ci
14962306a36Sopenharmony_ci	if (cg_read(cgroup, control, buf, sizeof(buf)))
15062306a36Sopenharmony_ci		return -1;
15162306a36Sopenharmony_ci
15262306a36Sopenharmony_ci	ptr = strstr(buf, key);
15362306a36Sopenharmony_ci	if (!ptr)
15462306a36Sopenharmony_ci		return -1;
15562306a36Sopenharmony_ci
15662306a36Sopenharmony_ci	return atol(ptr + strlen(key));
15762306a36Sopenharmony_ci}
15862306a36Sopenharmony_ci
15962306a36Sopenharmony_cilong cg_read_lc(const char *cgroup, const char *control)
16062306a36Sopenharmony_ci{
16162306a36Sopenharmony_ci	char buf[PAGE_SIZE];
16262306a36Sopenharmony_ci	const char delim[] = "\n";
16362306a36Sopenharmony_ci	char *line;
16462306a36Sopenharmony_ci	long cnt = 0;
16562306a36Sopenharmony_ci
16662306a36Sopenharmony_ci	if (cg_read(cgroup, control, buf, sizeof(buf)))
16762306a36Sopenharmony_ci		return -1;
16862306a36Sopenharmony_ci
16962306a36Sopenharmony_ci	for (line = strtok(buf, delim); line; line = strtok(NULL, delim))
17062306a36Sopenharmony_ci		cnt++;
17162306a36Sopenharmony_ci
17262306a36Sopenharmony_ci	return cnt;
17362306a36Sopenharmony_ci}
17462306a36Sopenharmony_ci
17562306a36Sopenharmony_ci/* Returns 0 on success, or -errno on failure. */
17662306a36Sopenharmony_ciint cg_write(const char *cgroup, const char *control, char *buf)
17762306a36Sopenharmony_ci{
17862306a36Sopenharmony_ci	char path[PATH_MAX];
17962306a36Sopenharmony_ci	ssize_t len = strlen(buf), ret;
18062306a36Sopenharmony_ci
18162306a36Sopenharmony_ci	snprintf(path, sizeof(path), "%s/%s", cgroup, control);
18262306a36Sopenharmony_ci	ret = write_text(path, buf, len);
18362306a36Sopenharmony_ci	return ret == len ? 0 : ret;
18462306a36Sopenharmony_ci}
18562306a36Sopenharmony_ci
18662306a36Sopenharmony_ciint cg_write_numeric(const char *cgroup, const char *control, long value)
18762306a36Sopenharmony_ci{
18862306a36Sopenharmony_ci	char buf[64];
18962306a36Sopenharmony_ci	int ret;
19062306a36Sopenharmony_ci
19162306a36Sopenharmony_ci	ret = sprintf(buf, "%lu", value);
19262306a36Sopenharmony_ci	if (ret < 0)
19362306a36Sopenharmony_ci		return ret;
19462306a36Sopenharmony_ci
19562306a36Sopenharmony_ci	return cg_write(cgroup, control, buf);
19662306a36Sopenharmony_ci}
19762306a36Sopenharmony_ci
19862306a36Sopenharmony_ciint cg_find_unified_root(char *root, size_t len)
19962306a36Sopenharmony_ci{
20062306a36Sopenharmony_ci	char buf[10 * PAGE_SIZE];
20162306a36Sopenharmony_ci	char *fs, *mount, *type;
20262306a36Sopenharmony_ci	const char delim[] = "\n\t ";
20362306a36Sopenharmony_ci
20462306a36Sopenharmony_ci	if (read_text("/proc/self/mounts", buf, sizeof(buf)) <= 0)
20562306a36Sopenharmony_ci		return -1;
20662306a36Sopenharmony_ci
20762306a36Sopenharmony_ci	/*
20862306a36Sopenharmony_ci	 * Example:
20962306a36Sopenharmony_ci	 * cgroup /sys/fs/cgroup cgroup2 rw,seclabel,noexec,relatime 0 0
21062306a36Sopenharmony_ci	 */
21162306a36Sopenharmony_ci	for (fs = strtok(buf, delim); fs; fs = strtok(NULL, delim)) {
21262306a36Sopenharmony_ci		mount = strtok(NULL, delim);
21362306a36Sopenharmony_ci		type = strtok(NULL, delim);
21462306a36Sopenharmony_ci		strtok(NULL, delim);
21562306a36Sopenharmony_ci		strtok(NULL, delim);
21662306a36Sopenharmony_ci		strtok(NULL, delim);
21762306a36Sopenharmony_ci
21862306a36Sopenharmony_ci		if (strcmp(type, "cgroup2") == 0) {
21962306a36Sopenharmony_ci			strncpy(root, mount, len);
22062306a36Sopenharmony_ci			return 0;
22162306a36Sopenharmony_ci		}
22262306a36Sopenharmony_ci	}
22362306a36Sopenharmony_ci
22462306a36Sopenharmony_ci	return -1;
22562306a36Sopenharmony_ci}
22662306a36Sopenharmony_ci
22762306a36Sopenharmony_ciint cg_create(const char *cgroup)
22862306a36Sopenharmony_ci{
22962306a36Sopenharmony_ci	return mkdir(cgroup, 0755);
23062306a36Sopenharmony_ci}
23162306a36Sopenharmony_ci
23262306a36Sopenharmony_ciint cg_wait_for_proc_count(const char *cgroup, int count)
23362306a36Sopenharmony_ci{
23462306a36Sopenharmony_ci	char buf[10 * PAGE_SIZE] = {0};
23562306a36Sopenharmony_ci	int attempts;
23662306a36Sopenharmony_ci	char *ptr;
23762306a36Sopenharmony_ci
23862306a36Sopenharmony_ci	for (attempts = 10; attempts >= 0; attempts--) {
23962306a36Sopenharmony_ci		int nr = 0;
24062306a36Sopenharmony_ci
24162306a36Sopenharmony_ci		if (cg_read(cgroup, "cgroup.procs", buf, sizeof(buf)))
24262306a36Sopenharmony_ci			break;
24362306a36Sopenharmony_ci
24462306a36Sopenharmony_ci		for (ptr = buf; *ptr; ptr++)
24562306a36Sopenharmony_ci			if (*ptr == '\n')
24662306a36Sopenharmony_ci				nr++;
24762306a36Sopenharmony_ci
24862306a36Sopenharmony_ci		if (nr >= count)
24962306a36Sopenharmony_ci			return 0;
25062306a36Sopenharmony_ci
25162306a36Sopenharmony_ci		usleep(100000);
25262306a36Sopenharmony_ci	}
25362306a36Sopenharmony_ci
25462306a36Sopenharmony_ci	return -1;
25562306a36Sopenharmony_ci}
25662306a36Sopenharmony_ci
25762306a36Sopenharmony_ciint cg_killall(const char *cgroup)
25862306a36Sopenharmony_ci{
25962306a36Sopenharmony_ci	char buf[PAGE_SIZE];
26062306a36Sopenharmony_ci	char *ptr = buf;
26162306a36Sopenharmony_ci
26262306a36Sopenharmony_ci	/* If cgroup.kill exists use it. */
26362306a36Sopenharmony_ci	if (!cg_write(cgroup, "cgroup.kill", "1"))
26462306a36Sopenharmony_ci		return 0;
26562306a36Sopenharmony_ci
26662306a36Sopenharmony_ci	if (cg_read(cgroup, "cgroup.procs", buf, sizeof(buf)))
26762306a36Sopenharmony_ci		return -1;
26862306a36Sopenharmony_ci
26962306a36Sopenharmony_ci	while (ptr < buf + sizeof(buf)) {
27062306a36Sopenharmony_ci		int pid = strtol(ptr, &ptr, 10);
27162306a36Sopenharmony_ci
27262306a36Sopenharmony_ci		if (pid == 0)
27362306a36Sopenharmony_ci			break;
27462306a36Sopenharmony_ci		if (*ptr)
27562306a36Sopenharmony_ci			ptr++;
27662306a36Sopenharmony_ci		else
27762306a36Sopenharmony_ci			break;
27862306a36Sopenharmony_ci		if (kill(pid, SIGKILL))
27962306a36Sopenharmony_ci			return -1;
28062306a36Sopenharmony_ci	}
28162306a36Sopenharmony_ci
28262306a36Sopenharmony_ci	return 0;
28362306a36Sopenharmony_ci}
28462306a36Sopenharmony_ci
28562306a36Sopenharmony_ciint cg_destroy(const char *cgroup)
28662306a36Sopenharmony_ci{
28762306a36Sopenharmony_ci	int ret;
28862306a36Sopenharmony_ci
28962306a36Sopenharmony_ci	if (!cgroup)
29062306a36Sopenharmony_ci		return 0;
29162306a36Sopenharmony_ciretry:
29262306a36Sopenharmony_ci	ret = rmdir(cgroup);
29362306a36Sopenharmony_ci	if (ret && errno == EBUSY) {
29462306a36Sopenharmony_ci		cg_killall(cgroup);
29562306a36Sopenharmony_ci		usleep(100);
29662306a36Sopenharmony_ci		goto retry;
29762306a36Sopenharmony_ci	}
29862306a36Sopenharmony_ci
29962306a36Sopenharmony_ci	if (ret && errno == ENOENT)
30062306a36Sopenharmony_ci		ret = 0;
30162306a36Sopenharmony_ci
30262306a36Sopenharmony_ci	return ret;
30362306a36Sopenharmony_ci}
30462306a36Sopenharmony_ci
30562306a36Sopenharmony_ciint cg_enter(const char *cgroup, int pid)
30662306a36Sopenharmony_ci{
30762306a36Sopenharmony_ci	char pidbuf[64];
30862306a36Sopenharmony_ci
30962306a36Sopenharmony_ci	snprintf(pidbuf, sizeof(pidbuf), "%d", pid);
31062306a36Sopenharmony_ci	return cg_write(cgroup, "cgroup.procs", pidbuf);
31162306a36Sopenharmony_ci}
31262306a36Sopenharmony_ci
31362306a36Sopenharmony_ciint cg_enter_current(const char *cgroup)
31462306a36Sopenharmony_ci{
31562306a36Sopenharmony_ci	return cg_write(cgroup, "cgroup.procs", "0");
31662306a36Sopenharmony_ci}
31762306a36Sopenharmony_ci
31862306a36Sopenharmony_ciint cg_enter_current_thread(const char *cgroup)
31962306a36Sopenharmony_ci{
32062306a36Sopenharmony_ci	return cg_write(cgroup, "cgroup.threads", "0");
32162306a36Sopenharmony_ci}
32262306a36Sopenharmony_ci
32362306a36Sopenharmony_ciint cg_run(const char *cgroup,
32462306a36Sopenharmony_ci	   int (*fn)(const char *cgroup, void *arg),
32562306a36Sopenharmony_ci	   void *arg)
32662306a36Sopenharmony_ci{
32762306a36Sopenharmony_ci	int pid, retcode;
32862306a36Sopenharmony_ci
32962306a36Sopenharmony_ci	pid = fork();
33062306a36Sopenharmony_ci	if (pid < 0) {
33162306a36Sopenharmony_ci		return pid;
33262306a36Sopenharmony_ci	} else if (pid == 0) {
33362306a36Sopenharmony_ci		char buf[64];
33462306a36Sopenharmony_ci
33562306a36Sopenharmony_ci		snprintf(buf, sizeof(buf), "%d", getpid());
33662306a36Sopenharmony_ci		if (cg_write(cgroup, "cgroup.procs", buf))
33762306a36Sopenharmony_ci			exit(EXIT_FAILURE);
33862306a36Sopenharmony_ci		exit(fn(cgroup, arg));
33962306a36Sopenharmony_ci	} else {
34062306a36Sopenharmony_ci		waitpid(pid, &retcode, 0);
34162306a36Sopenharmony_ci		if (WIFEXITED(retcode))
34262306a36Sopenharmony_ci			return WEXITSTATUS(retcode);
34362306a36Sopenharmony_ci		else
34462306a36Sopenharmony_ci			return -1;
34562306a36Sopenharmony_ci	}
34662306a36Sopenharmony_ci}
34762306a36Sopenharmony_ci
34862306a36Sopenharmony_cipid_t clone_into_cgroup(int cgroup_fd)
34962306a36Sopenharmony_ci{
35062306a36Sopenharmony_ci#ifdef CLONE_ARGS_SIZE_VER2
35162306a36Sopenharmony_ci	pid_t pid;
35262306a36Sopenharmony_ci
35362306a36Sopenharmony_ci	struct __clone_args args = {
35462306a36Sopenharmony_ci		.flags = CLONE_INTO_CGROUP,
35562306a36Sopenharmony_ci		.exit_signal = SIGCHLD,
35662306a36Sopenharmony_ci		.cgroup = cgroup_fd,
35762306a36Sopenharmony_ci	};
35862306a36Sopenharmony_ci
35962306a36Sopenharmony_ci	pid = sys_clone3(&args, sizeof(struct __clone_args));
36062306a36Sopenharmony_ci	/*
36162306a36Sopenharmony_ci	 * Verify that this is a genuine test failure:
36262306a36Sopenharmony_ci	 * ENOSYS -> clone3() not available
36362306a36Sopenharmony_ci	 * E2BIG  -> CLONE_INTO_CGROUP not available
36462306a36Sopenharmony_ci	 */
36562306a36Sopenharmony_ci	if (pid < 0 && (errno == ENOSYS || errno == E2BIG))
36662306a36Sopenharmony_ci		goto pretend_enosys;
36762306a36Sopenharmony_ci
36862306a36Sopenharmony_ci	return pid;
36962306a36Sopenharmony_ci
37062306a36Sopenharmony_cipretend_enosys:
37162306a36Sopenharmony_ci#endif
37262306a36Sopenharmony_ci	errno = ENOSYS;
37362306a36Sopenharmony_ci	return -ENOSYS;
37462306a36Sopenharmony_ci}
37562306a36Sopenharmony_ci
37662306a36Sopenharmony_ciint clone_reap(pid_t pid, int options)
37762306a36Sopenharmony_ci{
37862306a36Sopenharmony_ci	int ret;
37962306a36Sopenharmony_ci	siginfo_t info = {
38062306a36Sopenharmony_ci		.si_signo = 0,
38162306a36Sopenharmony_ci	};
38262306a36Sopenharmony_ci
38362306a36Sopenharmony_ciagain:
38462306a36Sopenharmony_ci	ret = waitid(P_PID, pid, &info, options | __WALL | __WNOTHREAD);
38562306a36Sopenharmony_ci	if (ret < 0) {
38662306a36Sopenharmony_ci		if (errno == EINTR)
38762306a36Sopenharmony_ci			goto again;
38862306a36Sopenharmony_ci		return -1;
38962306a36Sopenharmony_ci	}
39062306a36Sopenharmony_ci
39162306a36Sopenharmony_ci	if (options & WEXITED) {
39262306a36Sopenharmony_ci		if (WIFEXITED(info.si_status))
39362306a36Sopenharmony_ci			return WEXITSTATUS(info.si_status);
39462306a36Sopenharmony_ci	}
39562306a36Sopenharmony_ci
39662306a36Sopenharmony_ci	if (options & WSTOPPED) {
39762306a36Sopenharmony_ci		if (WIFSTOPPED(info.si_status))
39862306a36Sopenharmony_ci			return WSTOPSIG(info.si_status);
39962306a36Sopenharmony_ci	}
40062306a36Sopenharmony_ci
40162306a36Sopenharmony_ci	if (options & WCONTINUED) {
40262306a36Sopenharmony_ci		if (WIFCONTINUED(info.si_status))
40362306a36Sopenharmony_ci			return 0;
40462306a36Sopenharmony_ci	}
40562306a36Sopenharmony_ci
40662306a36Sopenharmony_ci	return -1;
40762306a36Sopenharmony_ci}
40862306a36Sopenharmony_ci
40962306a36Sopenharmony_ciint dirfd_open_opath(const char *dir)
41062306a36Sopenharmony_ci{
41162306a36Sopenharmony_ci	return open(dir, O_DIRECTORY | O_CLOEXEC | O_NOFOLLOW | O_PATH);
41262306a36Sopenharmony_ci}
41362306a36Sopenharmony_ci
41462306a36Sopenharmony_ci#define close_prot_errno(fd)                                                   \
41562306a36Sopenharmony_ci	if (fd >= 0) {                                                         \
41662306a36Sopenharmony_ci		int _e_ = errno;                                               \
41762306a36Sopenharmony_ci		close(fd);                                                     \
41862306a36Sopenharmony_ci		errno = _e_;                                                   \
41962306a36Sopenharmony_ci	}
42062306a36Sopenharmony_ci
42162306a36Sopenharmony_cistatic int clone_into_cgroup_run_nowait(const char *cgroup,
42262306a36Sopenharmony_ci					int (*fn)(const char *cgroup, void *arg),
42362306a36Sopenharmony_ci					void *arg)
42462306a36Sopenharmony_ci{
42562306a36Sopenharmony_ci	int cgroup_fd;
42662306a36Sopenharmony_ci	pid_t pid;
42762306a36Sopenharmony_ci
42862306a36Sopenharmony_ci	cgroup_fd =  dirfd_open_opath(cgroup);
42962306a36Sopenharmony_ci	if (cgroup_fd < 0)
43062306a36Sopenharmony_ci		return -1;
43162306a36Sopenharmony_ci
43262306a36Sopenharmony_ci	pid = clone_into_cgroup(cgroup_fd);
43362306a36Sopenharmony_ci	close_prot_errno(cgroup_fd);
43462306a36Sopenharmony_ci	if (pid == 0)
43562306a36Sopenharmony_ci		exit(fn(cgroup, arg));
43662306a36Sopenharmony_ci
43762306a36Sopenharmony_ci	return pid;
43862306a36Sopenharmony_ci}
43962306a36Sopenharmony_ci
44062306a36Sopenharmony_ciint cg_run_nowait(const char *cgroup,
44162306a36Sopenharmony_ci		  int (*fn)(const char *cgroup, void *arg),
44262306a36Sopenharmony_ci		  void *arg)
44362306a36Sopenharmony_ci{
44462306a36Sopenharmony_ci	int pid;
44562306a36Sopenharmony_ci
44662306a36Sopenharmony_ci	pid = clone_into_cgroup_run_nowait(cgroup, fn, arg);
44762306a36Sopenharmony_ci	if (pid > 0)
44862306a36Sopenharmony_ci		return pid;
44962306a36Sopenharmony_ci
45062306a36Sopenharmony_ci	/* Genuine test failure. */
45162306a36Sopenharmony_ci	if (pid < 0 && errno != ENOSYS)
45262306a36Sopenharmony_ci		return -1;
45362306a36Sopenharmony_ci
45462306a36Sopenharmony_ci	pid = fork();
45562306a36Sopenharmony_ci	if (pid == 0) {
45662306a36Sopenharmony_ci		char buf[64];
45762306a36Sopenharmony_ci
45862306a36Sopenharmony_ci		snprintf(buf, sizeof(buf), "%d", getpid());
45962306a36Sopenharmony_ci		if (cg_write(cgroup, "cgroup.procs", buf))
46062306a36Sopenharmony_ci			exit(EXIT_FAILURE);
46162306a36Sopenharmony_ci		exit(fn(cgroup, arg));
46262306a36Sopenharmony_ci	}
46362306a36Sopenharmony_ci
46462306a36Sopenharmony_ci	return pid;
46562306a36Sopenharmony_ci}
46662306a36Sopenharmony_ci
46762306a36Sopenharmony_ciint get_temp_fd(void)
46862306a36Sopenharmony_ci{
46962306a36Sopenharmony_ci	return open(".", O_TMPFILE | O_RDWR | O_EXCL);
47062306a36Sopenharmony_ci}
47162306a36Sopenharmony_ci
47262306a36Sopenharmony_ciint alloc_pagecache(int fd, size_t size)
47362306a36Sopenharmony_ci{
47462306a36Sopenharmony_ci	char buf[PAGE_SIZE];
47562306a36Sopenharmony_ci	struct stat st;
47662306a36Sopenharmony_ci	int i;
47762306a36Sopenharmony_ci
47862306a36Sopenharmony_ci	if (fstat(fd, &st))
47962306a36Sopenharmony_ci		goto cleanup;
48062306a36Sopenharmony_ci
48162306a36Sopenharmony_ci	size += st.st_size;
48262306a36Sopenharmony_ci
48362306a36Sopenharmony_ci	if (ftruncate(fd, size))
48462306a36Sopenharmony_ci		goto cleanup;
48562306a36Sopenharmony_ci
48662306a36Sopenharmony_ci	for (i = 0; i < size; i += sizeof(buf))
48762306a36Sopenharmony_ci		read(fd, buf, sizeof(buf));
48862306a36Sopenharmony_ci
48962306a36Sopenharmony_ci	return 0;
49062306a36Sopenharmony_ci
49162306a36Sopenharmony_cicleanup:
49262306a36Sopenharmony_ci	return -1;
49362306a36Sopenharmony_ci}
49462306a36Sopenharmony_ci
49562306a36Sopenharmony_ciint alloc_anon(const char *cgroup, void *arg)
49662306a36Sopenharmony_ci{
49762306a36Sopenharmony_ci	size_t size = (unsigned long)arg;
49862306a36Sopenharmony_ci	char *buf, *ptr;
49962306a36Sopenharmony_ci
50062306a36Sopenharmony_ci	buf = malloc(size);
50162306a36Sopenharmony_ci	for (ptr = buf; ptr < buf + size; ptr += PAGE_SIZE)
50262306a36Sopenharmony_ci		*ptr = 0;
50362306a36Sopenharmony_ci
50462306a36Sopenharmony_ci	free(buf);
50562306a36Sopenharmony_ci	return 0;
50662306a36Sopenharmony_ci}
50762306a36Sopenharmony_ci
50862306a36Sopenharmony_ciint is_swap_enabled(void)
50962306a36Sopenharmony_ci{
51062306a36Sopenharmony_ci	char buf[PAGE_SIZE];
51162306a36Sopenharmony_ci	const char delim[] = "\n";
51262306a36Sopenharmony_ci	int cnt = 0;
51362306a36Sopenharmony_ci	char *line;
51462306a36Sopenharmony_ci
51562306a36Sopenharmony_ci	if (read_text("/proc/swaps", buf, sizeof(buf)) <= 0)
51662306a36Sopenharmony_ci		return -1;
51762306a36Sopenharmony_ci
51862306a36Sopenharmony_ci	for (line = strtok(buf, delim); line; line = strtok(NULL, delim))
51962306a36Sopenharmony_ci		cnt++;
52062306a36Sopenharmony_ci
52162306a36Sopenharmony_ci	return cnt > 1;
52262306a36Sopenharmony_ci}
52362306a36Sopenharmony_ci
52462306a36Sopenharmony_ciint set_oom_adj_score(int pid, int score)
52562306a36Sopenharmony_ci{
52662306a36Sopenharmony_ci	char path[PATH_MAX];
52762306a36Sopenharmony_ci	int fd, len;
52862306a36Sopenharmony_ci
52962306a36Sopenharmony_ci	sprintf(path, "/proc/%d/oom_score_adj", pid);
53062306a36Sopenharmony_ci
53162306a36Sopenharmony_ci	fd = open(path, O_WRONLY | O_APPEND);
53262306a36Sopenharmony_ci	if (fd < 0)
53362306a36Sopenharmony_ci		return fd;
53462306a36Sopenharmony_ci
53562306a36Sopenharmony_ci	len = dprintf(fd, "%d", score);
53662306a36Sopenharmony_ci	if (len < 0) {
53762306a36Sopenharmony_ci		close(fd);
53862306a36Sopenharmony_ci		return len;
53962306a36Sopenharmony_ci	}
54062306a36Sopenharmony_ci
54162306a36Sopenharmony_ci	close(fd);
54262306a36Sopenharmony_ci	return 0;
54362306a36Sopenharmony_ci}
54462306a36Sopenharmony_ci
54562306a36Sopenharmony_ciint proc_mount_contains(const char *option)
54662306a36Sopenharmony_ci{
54762306a36Sopenharmony_ci	char buf[4 * PAGE_SIZE];
54862306a36Sopenharmony_ci	ssize_t read;
54962306a36Sopenharmony_ci
55062306a36Sopenharmony_ci	read = read_text("/proc/mounts", buf, sizeof(buf));
55162306a36Sopenharmony_ci	if (read < 0)
55262306a36Sopenharmony_ci		return read;
55362306a36Sopenharmony_ci
55462306a36Sopenharmony_ci	return strstr(buf, option) != NULL;
55562306a36Sopenharmony_ci}
55662306a36Sopenharmony_ci
55762306a36Sopenharmony_cissize_t proc_read_text(int pid, bool thread, const char *item, char *buf, size_t size)
55862306a36Sopenharmony_ci{
55962306a36Sopenharmony_ci	char path[PATH_MAX];
56062306a36Sopenharmony_ci	ssize_t ret;
56162306a36Sopenharmony_ci
56262306a36Sopenharmony_ci	if (!pid)
56362306a36Sopenharmony_ci		snprintf(path, sizeof(path), "/proc/%s/%s",
56462306a36Sopenharmony_ci			 thread ? "thread-self" : "self", item);
56562306a36Sopenharmony_ci	else
56662306a36Sopenharmony_ci		snprintf(path, sizeof(path), "/proc/%d/%s", pid, item);
56762306a36Sopenharmony_ci
56862306a36Sopenharmony_ci	ret = read_text(path, buf, size);
56962306a36Sopenharmony_ci	return ret < 0 ? -1 : ret;
57062306a36Sopenharmony_ci}
57162306a36Sopenharmony_ci
57262306a36Sopenharmony_ciint proc_read_strstr(int pid, bool thread, const char *item, const char *needle)
57362306a36Sopenharmony_ci{
57462306a36Sopenharmony_ci	char buf[PAGE_SIZE];
57562306a36Sopenharmony_ci
57662306a36Sopenharmony_ci	if (proc_read_text(pid, thread, item, buf, sizeof(buf)) < 0)
57762306a36Sopenharmony_ci		return -1;
57862306a36Sopenharmony_ci
57962306a36Sopenharmony_ci	return strstr(buf, needle) ? 0 : -1;
58062306a36Sopenharmony_ci}
58162306a36Sopenharmony_ci
58262306a36Sopenharmony_ciint clone_into_cgroup_run_wait(const char *cgroup)
58362306a36Sopenharmony_ci{
58462306a36Sopenharmony_ci	int cgroup_fd;
58562306a36Sopenharmony_ci	pid_t pid;
58662306a36Sopenharmony_ci
58762306a36Sopenharmony_ci	cgroup_fd =  dirfd_open_opath(cgroup);
58862306a36Sopenharmony_ci	if (cgroup_fd < 0)
58962306a36Sopenharmony_ci		return -1;
59062306a36Sopenharmony_ci
59162306a36Sopenharmony_ci	pid = clone_into_cgroup(cgroup_fd);
59262306a36Sopenharmony_ci	close_prot_errno(cgroup_fd);
59362306a36Sopenharmony_ci	if (pid < 0)
59462306a36Sopenharmony_ci		return -1;
59562306a36Sopenharmony_ci
59662306a36Sopenharmony_ci	if (pid == 0)
59762306a36Sopenharmony_ci		exit(EXIT_SUCCESS);
59862306a36Sopenharmony_ci
59962306a36Sopenharmony_ci	/*
60062306a36Sopenharmony_ci	 * We don't care whether this fails. We only care whether the initial
60162306a36Sopenharmony_ci	 * clone succeeded.
60262306a36Sopenharmony_ci	 */
60362306a36Sopenharmony_ci	(void)clone_reap(pid, WEXITED);
60462306a36Sopenharmony_ci	return 0;
60562306a36Sopenharmony_ci}
60662306a36Sopenharmony_ci
60762306a36Sopenharmony_cistatic int __prepare_for_wait(const char *cgroup, const char *filename)
60862306a36Sopenharmony_ci{
60962306a36Sopenharmony_ci	int fd, ret = -1;
61062306a36Sopenharmony_ci
61162306a36Sopenharmony_ci	fd = inotify_init1(0);
61262306a36Sopenharmony_ci	if (fd == -1)
61362306a36Sopenharmony_ci		return fd;
61462306a36Sopenharmony_ci
61562306a36Sopenharmony_ci	ret = inotify_add_watch(fd, cg_control(cgroup, filename), IN_MODIFY);
61662306a36Sopenharmony_ci	if (ret == -1) {
61762306a36Sopenharmony_ci		close(fd);
61862306a36Sopenharmony_ci		fd = -1;
61962306a36Sopenharmony_ci	}
62062306a36Sopenharmony_ci
62162306a36Sopenharmony_ci	return fd;
62262306a36Sopenharmony_ci}
62362306a36Sopenharmony_ci
62462306a36Sopenharmony_ciint cg_prepare_for_wait(const char *cgroup)
62562306a36Sopenharmony_ci{
62662306a36Sopenharmony_ci	return __prepare_for_wait(cgroup, "cgroup.events");
62762306a36Sopenharmony_ci}
62862306a36Sopenharmony_ci
62962306a36Sopenharmony_ciint memcg_prepare_for_wait(const char *cgroup)
63062306a36Sopenharmony_ci{
63162306a36Sopenharmony_ci	return __prepare_for_wait(cgroup, "memory.events");
63262306a36Sopenharmony_ci}
63362306a36Sopenharmony_ci
63462306a36Sopenharmony_ciint cg_wait_for(int fd)
63562306a36Sopenharmony_ci{
63662306a36Sopenharmony_ci	int ret = -1;
63762306a36Sopenharmony_ci	struct pollfd fds = {
63862306a36Sopenharmony_ci		.fd = fd,
63962306a36Sopenharmony_ci		.events = POLLIN,
64062306a36Sopenharmony_ci	};
64162306a36Sopenharmony_ci
64262306a36Sopenharmony_ci	while (true) {
64362306a36Sopenharmony_ci		ret = poll(&fds, 1, 10000);
64462306a36Sopenharmony_ci
64562306a36Sopenharmony_ci		if (ret == -1) {
64662306a36Sopenharmony_ci			if (errno == EINTR)
64762306a36Sopenharmony_ci				continue;
64862306a36Sopenharmony_ci
64962306a36Sopenharmony_ci			break;
65062306a36Sopenharmony_ci		}
65162306a36Sopenharmony_ci
65262306a36Sopenharmony_ci		if (ret > 0 && fds.revents & POLLIN) {
65362306a36Sopenharmony_ci			ret = 0;
65462306a36Sopenharmony_ci			break;
65562306a36Sopenharmony_ci		}
65662306a36Sopenharmony_ci	}
65762306a36Sopenharmony_ci
65862306a36Sopenharmony_ci	return ret;
65962306a36Sopenharmony_ci}
660