static void test_cg_tests(void) {
int all, hybrid, systemd, r;
r = cg_unified_flush();
if (r == -ENOMEDIUM) {
log_notice_errno(r, "Skipping cg hierarchy tests: %m");
return;
}
assert_se(r == 0);
all = cg_all_unified();
assert_se(IN_SET(all, 0, 1));
hybrid = cg_hybrid_unified();
assert_se(IN_SET(hybrid, 0, 1));
systemd = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
assert_se(IN_SET(systemd, 0, 1));
if (all) {
assert_se(systemd);
assert_se(!hybrid);
} else if (hybrid) {
assert_se(systemd);
assert_se(!all);
} else
assert_se(!systemd);
}
static void show_cg_info(const char *controller, const char *path) {
if (cg_all_unified() == 0 && controller && !streq(controller, SYSTEMD_CGROUP_CONTROLLER))
printf("Controller %s; ", controller);
printf("Control group %s:\n", isempty(path) ? "/" : path);
fflush(stdout);
}
static int process(
const char *controller,
const char *path,
Hashmap *a,
Hashmap *b,
unsigned iteration,
Group **ret) {
Group *g;
int r;
assert(controller);
assert(path);
assert(a);
g = hashmap_get(a, path);
if (!g) {
g = hashmap_get(b, path);
if (!g) {
g = new0(Group, 1);
if (!g)
return -ENOMEM;
g->path = strdup(path);
if (!g->path) {
group_free(g);
return -ENOMEM;
}
r = hashmap_put(a, g->path, g);
if (r < 0) {
group_free(g);
return r;
}
} else {
r = hashmap_move_one(a, b, path);
if (r < 0)
return r;
g->cpu_valid = g->memory_valid = g->io_valid = g->n_tasks_valid = false;
}
}
if (streq(controller, SYSTEMD_CGROUP_CONTROLLER) && IN_SET(arg_count, COUNT_ALL_PROCESSES, COUNT_USERSPACE_PROCESSES)) {
_cleanup_fclose_ FILE *f = NULL;
pid_t pid;
r = cg_enumerate_processes(controller, path, &f);
if (r == -ENOENT)
return 0;
if (r < 0)
return r;
g->n_tasks = 0;
while (cg_read_pid(f, &pid) > 0) {
if (arg_count == COUNT_USERSPACE_PROCESSES && is_kernel_thread(pid) > 0)
continue;
g->n_tasks++;
}
if (g->n_tasks > 0)
g->n_tasks_valid = true;
} else if (streq(controller, "pids") && arg_count == COUNT_PIDS) {
_cleanup_free_ char *p = NULL, *v = NULL;
r = cg_get_path(controller, path, "pids.current", &p);
if (r < 0)
return r;
r = read_one_line_file(p, &v);
if (r == -ENOENT)
return 0;
if (r < 0)
return r;
r = safe_atou64(v, &g->n_tasks);
if (r < 0)
return r;
if (g->n_tasks > 0)
g->n_tasks_valid = true;
} else if (streq(controller, "cpu") || streq(controller, "cpuacct")) {
_cleanup_free_ char *p = NULL, *v = NULL;
uint64_t new_usage;
nsec_t timestamp;
if (cg_all_unified() > 0) {
const char *keys[] = { "usage_usec", NULL };
_cleanup_free_ char *val = NULL;
if (!streq(controller, "cpu"))
return 0;
r = cg_get_keyed_attribute("cpu", path, "cpu.stat", keys, &val);
if (r == -ENOENT)
return 0;
if (r < 0)
return r;
r = safe_atou64(val, &new_usage);
if (r < 0)
return r;
new_usage *= NSEC_PER_USEC;
} else {
if (!streq(controller, "cpuacct"))
return 0;
r = cg_get_path(controller, path, "cpuacct.usage", &p);
if (r < 0)
return r;
r = read_one_line_file(p, &v);
if (r == -ENOENT)
return 0;
if (r < 0)
return r;
r = safe_atou64(v, &new_usage);
if (r < 0)
return r;
}
timestamp = now_nsec(CLOCK_MONOTONIC);
if (g->cpu_iteration == iteration - 1 &&
(nsec_t) new_usage > g->cpu_usage) {
nsec_t x, y;
x = timestamp - g->cpu_timestamp;
if (x < 1)
x = 1;
y = (nsec_t) new_usage - g->cpu_usage;
g->cpu_fraction = (double) y / (double) x;
g->cpu_valid = true;
}
g->cpu_usage = (nsec_t) new_usage;
g->cpu_timestamp = timestamp;
g->cpu_iteration = iteration;
} else if (streq(controller, "memory")) {
_cleanup_free_ char *p = NULL, *v = NULL;
if (cg_all_unified() <= 0)
r = cg_get_path(controller, path, "memory.usage_in_bytes", &p);
else
r = cg_get_path(controller, path, "memory.current", &p);
if (r < 0)
return r;
r = read_one_line_file(p, &v);
if (r == -ENOENT)
return 0;
if (r < 0)
return r;
r = safe_atou64(v, &g->memory);
if (r < 0)
return r;
if (g->memory > 0)
g->memory_valid = true;
} else if ((streq(controller, "io") && cg_all_unified() > 0) ||
(streq(controller, "blkio") && cg_all_unified() <= 0)) {
_cleanup_fclose_ FILE *f = NULL;
_cleanup_free_ char *p = NULL;
bool unified = cg_all_unified() > 0;
uint64_t wr = 0, rd = 0;
nsec_t timestamp;
r = cg_get_path(controller, path, unified ? "io.stat" : "blkio.io_service_bytes", &p);
if (r < 0)
return r;
f = fopen(p, "re");
if (!f) {
if (errno == ENOENT)
return 0;
return -errno;
}
for (;;) {
char line[LINE_MAX], *l;
uint64_t k, *q;
if (!fgets(line, sizeof(line), f))
break;
/* Trim and skip the device */
l = strstrip(line);
l += strcspn(l, WHITESPACE);
l += strspn(l, WHITESPACE);
if (unified) {
while (!isempty(l)) {
if (sscanf(l, "rbytes=%" SCNu64, &k))
rd += k;
else if (sscanf(l, "wbytes=%" SCNu64, &k))
wr += k;
l += strcspn(l, WHITESPACE);
l += strspn(l, WHITESPACE);
}
} else {
if (first_word(l, "Read")) {
l += 4;
q = &rd;
} else if (first_word(l, "Write")) {
l += 5;
q = ≀
} else
continue;
l += strspn(l, WHITESPACE);
r = safe_atou64(l, &k);
if (r < 0)
continue;
*q += k;
}
}
timestamp = now_nsec(CLOCK_MONOTONIC);
if (g->io_iteration == iteration - 1) {
uint64_t x, yr, yw;
x = (uint64_t) (timestamp - g->io_timestamp);
if (x < 1)
x = 1;
if (rd > g->io_input)
yr = rd - g->io_input;
else
yr = 0;
if (wr > g->io_output)
yw = wr - g->io_output;
else
yw = 0;
if (yr > 0 || yw > 0) {
g->io_input_bps = (yr * 1000000000ULL) / x;
g->io_output_bps = (yw * 1000000000ULL) / x;
g->io_valid = true;
}
}
g->io_input = rd;
g->io_output = wr;
g->io_timestamp = timestamp;
g->io_iteration = iteration;
}
if (ret)
*ret = g;
return 0;
}
/* Mount legacy cgroup hierarchy when cgroup namespaces are unsupported. */
static int mount_legacy_cgns_unsupported(
const char *dest,
CGroupUnified unified_requested,
bool userns,
uid_t uid_shift,
uid_t uid_range,
const char *selinux_apifs_context) {
_cleanup_set_free_free_ Set *controllers = NULL;
const char *cgroup_root;
int r;
cgroup_root = prefix_roota(dest, "/sys/fs/cgroup");
(void) mkdir_p(cgroup_root, 0755);
/* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
r = path_is_mount_point(cgroup_root, dest, AT_SYMLINK_FOLLOW);
if (r < 0)
return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
if (r == 0) {
_cleanup_free_ char *options = NULL;
r = tmpfs_patch_options("mode=755", uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &options);
if (r < 0)
return log_oom();
r = mount_verbose(LOG_ERR, "tmpfs", cgroup_root, "tmpfs",
MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options);
if (r < 0)
return r;
}
r = cg_all_unified();
if (r < 0)
return r;
if (r > 0)
goto skip_controllers;
r = cg_kernel_controllers(&controllers);
if (r < 0)
return log_error_errno(r, "Failed to determine cgroup controllers: %m");
for (;;) {
_cleanup_free_ char *controller = NULL, *origin = NULL, *combined = NULL;
controller = set_steal_first(controllers);
if (!controller)
break;
origin = prefix_root("/sys/fs/cgroup/", controller);
if (!origin)
return log_oom();
r = readlink_malloc(origin, &combined);
if (r == -EINVAL) {
/* Not a symbolic link, but directly a single cgroup hierarchy */
r = mount_legacy_cgroup_hierarchy(dest, controller, controller, true);
if (r < 0)
return r;
} else if (r < 0)
return log_error_errno(r, "Failed to read link %s: %m", origin);
else {
_cleanup_free_ char *target = NULL;
target = prefix_root(dest, origin);
if (!target)
return log_oom();
/* A symbolic link, a combination of controllers in one hierarchy */
if (!filename_is_valid(combined)) {
log_warning("Ignoring invalid combined hierarchy %s.", combined);
continue;
}
r = mount_legacy_cgroup_hierarchy(dest, combined, combined, true);
if (r < 0)
return r;
r = symlink_idempotent(combined, target, false);
if (r == -EINVAL)
return log_error_errno(r, "Invalid existing symlink for combined hierarchy: %m");
if (r < 0)
return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m");
}
}
skip_controllers:
if (unified_requested >= CGROUP_UNIFIED_SYSTEMD) {
r = mount_legacy_cgroup_hierarchy(dest, SYSTEMD_CGROUP_CONTROLLER_HYBRID, "unified", false);
if (r < 0)
return r;
}
r = mount_legacy_cgroup_hierarchy(dest, SYSTEMD_CGROUP_CONTROLLER_LEGACY, "systemd", false);
if (r < 0)
return r;
return mount_verbose(LOG_ERR, NULL, cgroup_root, NULL,
MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
}
/* Mount a legacy cgroup hierarchy when cgroup namespaces are supported. */
static int mount_legacy_cgns_supported(
const char *dest,
CGroupUnified unified_requested,
bool userns,
uid_t uid_shift,
uid_t uid_range,
const char *selinux_apifs_context) {
_cleanup_set_free_free_ Set *controllers = NULL;
const char *cgroup_root = "/sys/fs/cgroup", *c;
int r;
(void) mkdir_p(cgroup_root, 0755);
/* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
r = path_is_mount_point(cgroup_root, dest, AT_SYMLINK_FOLLOW);
if (r < 0)
return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
if (r == 0) {
_cleanup_free_ char *options = NULL;
/* When cgroup namespaces are enabled and user namespaces are
* used then the mount of the cgroupfs is done *inside* the new
* user namespace. We're root in the new user namespace and the
* kernel will happily translate our uid/gid to the correct
* uid/gid as seen from e.g. /proc/1/mountinfo. So we simply
* pass uid 0 and not uid_shift to tmpfs_patch_options().
*/
r = tmpfs_patch_options("mode=755", 0, selinux_apifs_context, &options);
if (r < 0)
return log_oom();
r = mount_verbose(LOG_ERR, "tmpfs", cgroup_root, "tmpfs",
MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options);
if (r < 0)
return r;
}
r = cg_all_unified();
if (r < 0)
return r;
if (r > 0)
goto skip_controllers;
r = get_process_controllers(&controllers);
if (r < 0)
return log_error_errno(r, "Failed to determine cgroup controllers: %m");
for (;;) {
_cleanup_free_ const char *controller = NULL;
controller = set_steal_first(controllers);
if (!controller)
break;
r = mount_legacy_cgroup_hierarchy("", controller, controller, !userns);
if (r < 0)
return r;
/* When multiple hierarchies are co-mounted, make their
* constituting individual hierarchies a symlink to the
* co-mount.
*/
c = controller;
for (;;) {
_cleanup_free_ char *target = NULL, *tok = NULL;
r = extract_first_word(&c, &tok, ",", 0);
if (r < 0)
return log_error_errno(r, "Failed to extract co-mounted cgroup controller: %m");
if (r == 0)
break;
if (streq(controller, tok))
break;
target = prefix_root("/sys/fs/cgroup/", tok);
if (!target)
return log_oom();
r = symlink_idempotent(controller, target, false);
if (r == -EINVAL)
return log_error_errno(r, "Invalid existing symlink for combined hierarchy: %m");
if (r < 0)
return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m");
}
}
skip_controllers:
if (unified_requested >= CGROUP_UNIFIED_SYSTEMD) {
r = mount_legacy_cgroup_hierarchy("", SYSTEMD_CGROUP_CONTROLLER_HYBRID, "unified", false);
if (r < 0)
return r;
}
r = mount_legacy_cgroup_hierarchy("", SYSTEMD_CGROUP_CONTROLLER_LEGACY, "systemd", false);
if (r < 0)
return r;
if (!userns)
return mount_verbose(LOG_ERR, NULL, cgroup_root, NULL,
MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
return 0;
}
